site banner
naraburns  nihil supernum  26d ago (text post)   2071 thread views

Friday Fun Thread for March 3, 2023

Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.

4
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

A very minor news story that illustrated the sort of unexpected cultural problems that might come to the way of fully autonomous vehicles on the streets in the future: it's election time in Finland, and the way that the Finnish electoral system works is that each candidate is assigned a number (could be '3', could be '53', could be '137' - there's a lot of candidates in Finnish elections, as there's a lot of parties and the biggest electoral district elects 37 MPs, meaning all the parties that can will run at least that number), and the way you vote for them is writing the candidate number on the ballot and then dropping it in the box. This means that displaying your candidate number prominently is important for candidates in advertising. So there's suddenly a lot of posters on the streets with big numbers in them... and the adaptive cruise control features in newer vehicles are suddenly interpreting them as speed limits.

Of course, when you have human drivers in charge, it's simple enough for them to just turn the cruise control off. One might imagine that autonomous vehicles would have bigger issues.

I've spent the weekend playing around with geodata. My original idea was simple: I wanted to draw a fantasy map of Russia. For that I needed a vector map of population density. Russian federal subjects are huge, so I needed a more granular map. I couldn't find one I could load into QGIS, so I thought to myself: how hard can it be?

  1. I downloaded a data dump for Russia from OSM.

  2. I converted it into Spatialite to query for admin boundaries.

  3. Turns out the JDBC driver for SQLite doesn't work in DBeaver when you want to load a native extension. Like Spatialite. Or rather, it works in the paid version of DBeaver. DBeaver itself recognizes the geometry field, but I can't run any geo functions.

  4. While I was screwing around with the data in DBeaver I realized it contained pieces of other countries. No, I am not talking about Crimea.

  5. I gave up on Spatialite and installed PostGIS. This allowed me to check for weird intersections between admin boundaries. None found. Good job, OSM mappers!

  6. I also realized I needed to somehow link the data to the census data to get the population. My only connection was going to be Wikidata.

  7. Turns out 12 admin boundaries in OSM don't have a Wikidata link. I actually went and fixed them in OSM.

  8. I then queried Wikidata for each boundary and got a bunch of JSON I parsed.

  9. Then I finally got around to the big problem of Russia having two types of boundaries: admin (OKATO) and municipal (OKTMO). They match, but not quite. Oh well, that's a problem for future orthoxerox to solve.

  10. But at the very least I now had a list of (presumably admin) boundaries from OSM I could join with Wikidata data, netting me a bunch of OKATO and OKTMO codes.

  11. I downloaded the list of OKATO codes and joined it with my boundaries for a sanity check. Found a few more admin boundaries from China and Japan and a few Wikidata entries without OKATO codes. Something for me to fix later.

  12. What's worse, a few small pieces of Russia were missing in the OSM dump. I missed them because I loaded the vector layer in QGIS, took a look at it and thought: "yep, looks like Russia to me".

  13. Now I am trying to find out what went wrong with the dump, but OSM forum authentication decided to die as soon as I registered there.

All in all, pretty entertaining. Now that I'm blocked from proceeding further, I can go look at the way DuckDB reads legacy Parquet timestamps.

I've never worked with GIS software before, so I don't know what they are usually capable of.

But I did work on a nasty boundary polygon problem in the past and the turfpy (I think the original package was made for node) package along with the geojson package in python was tremendously useful for sorting out some of the issues that were similar to yours.

Thank you, I'm pretty happy with the tools I have. PostGIS and QGIS have got me covered, it's the data quality that got me this time.

Could someone give me a list of good classical music written in the last 70 years?

For choir music the most accessible and famous modern composer is probably Eric Whitacre.

You could try Sleep or Lux Aurumque to see if it's your cup of tea.

I feel like your time period is perhaps a bit too wide as many good classical composers still lived and produced in the 50s-60s and suggestions might therefore mislead you a bit if you're looking for modern composers.

Also, composers of what? Orchestral? Chamber? Choral? Operas? Ballets?

Edit: @fuckduck9000 i thought I'd add a two more recommendations:

Morten Lauridsen:

Example work: O magnum mysterium (1994)

Ola Gjeilo:

Example work: Tota Pulchra Es (2009)

Shostakovich was still alive for the first 20 or so of those years. His eighth string quartet, from 1960, stands out.

Names to consider off the top of my head...

  • John Williams, obviously.

  • Michael Giacchino, you probably haven't heard of him but the score for the Incredibles was fucking incredible and he's been a solid name on the Hollywood soundtrack scene since.

  • Koji Kondo, between the main themes for Mario, Zelda, and Metroid Kondo is another dude many know but have never heard of.

  • James Horner, turns out "discount John Williams" is still pretty damn good, See Terminator, Apollo 13, Saving Private Ryan, et al.

  • Basil Poludoris, his work is kind of uneven but Anvil of Crom, Red Dawn, and Klendathu Drop are all underrated masterpieces CMV.

  • Hans Zimmer, the more you listen to him the more you realize he's only got one trick, fortunately it's a solid trick

Good, good. Something similar to the Köln Concert would also do, if anyone has suggestions.

It occurs to me that if we're going back 70 years we ought to include Jerry Goldsmith and Miklos Rosas as well. I was thinking "recent" as in stuff that had come out within my lifetime but that's actually cutting the window ~25 years short.

Try John Adams, his works are used extensively in the soundtrack of the computer game Civilization 4. In particular The People are the Heroes Now from the opera Nixon in China, Chairman Dances or Shaker Loops.

For a more avant-garde (and repetitive) sound there is Philip Glass: Runaway Horses from the A Life in Four Chapters movie soundtrack, Pruitt Igoe from Koyaanisqatsi movie soundtrack, or the opera Einstein on the Beach.

Yeah Hallelujah Junction and Life in the mountains are favourites. But I thought maybe as a movie-goer I was missing an entire iceberg of music that only people who regularly go to classical concerts know, but apparently film scores are it.

There exists, but I didn't want to recommend something I personally don't like. Putting that aside, there are György Ligeti and Karlheinz Stockhausen, maybe you will find them to your taste.

Thank you. But what the shit? My ears are not that sophisticated.

Might be cheating because many of the themes are from earlier works but the Star Wars theme is probably the most popular classical work from that era.

I like film scores, sure, Zimmer, Burwell, Williams, Glass, Tiersen, Morricone, but is there nothing more?

Yes, a lot, but none very well known.

If you find them worth listening to, I'd like to hear them.

What does ChatGPT say?

Here's what it gave me, in case anyone's interested.

The past 70 years have seen many great composers who have contributed to classical music. Here are a few examples, in no particular order:

John Williams: Known for his film scores such as Star Wars, Indiana Jones, and Jurassic Park, Williams is one of the most successful and recognizable composers of our time.

Philip Glass: An American composer known for his minimalist style, Glass has composed numerous operas, symphonies, and film scores.

Steve Reich: Another minimalist composer, Reich's works often involve repetition and gradual change. Some of his most well-known works include "Music for 18 Musicians" and "Different Trains."

Arvo Pärt: An Estonian composer, Pärt is known for his unique style of "tintinnabuli," which involves simple, melodic lines and bell-like tones.

Hans Zimmer: A German composer who has scored many popular films such as The Lion King, Inception, and The Dark Knight trilogy, Zimmer is known for his use of electronic and orchestral elements.

György Ligeti: A Hungarian composer known for his avant-garde style and use of micropolyphony, Ligeti's works have been used in films such as 2001: A Space Odyssey and The Shining.

Krzysztof Penderecki: A Polish composer whose works often incorporate elements of avant-garde and serialism, Penderecki's compositions have been used in numerous films and television shows.

Thomas Adès: A British composer who has written operas, orchestral works, and chamber music, Adès is known for his eclectic style and use of unconventional instruments.

These are just a few examples of the many great composers of the past 70 years.

Bing:

If you are looking for classical composers who were born after 195045, some of the names that appear frequently are John Adams (born 1947), Thomas Adès (born 1971), John Corigliano (born 1938), Philip Glass (born 1937), Arvo Pärt (born 1935), Kaija Saariaho (born 1952), Tan Dun (born 1957) and Esa-Pekka Salonen (born 1958).

I'm not sure if it recognized my location and thus gave me two Finnish composers (Saariaho and Salonen) or if it would have done it nevertheless.

Giving machines the silent treatment is my petty contribution to luddism.

I've taken the plunge and created a Substack. I've done it under my real name and will be writing about NON CW things I generally find interesting mainly to divert my tendency to write thousands of words into the void of the Internet towards something marginally more "productive" than forum posting and to have a corpus of writing associated to my name that I can use later for career/social status (One of my MEDIUM posts was quite well received by my past colleagues). In light of my previous post on too much forum posting. Now I am not a prolific Motte poster by any means, but I still post too much according to my personal cost/benefit analysis of how I should allocate my time.

Now is an easier time than ever to become a writer, you don't even have to try to make a website (substack does most of the work for you), language models can do some if not almost all your grunt work of writing for you, and you can create artwork/images to supplement your text at no cost using txt2img... This doesn't mean it's easier to make it as a writer, just doing the writing.

Half the reason I finally published and shared my own web serial a week back is because I do actually believe that AI circa GPT-4.5 will genuinely write better than I do, and I'm a pretty good writer!

I've had a modicum of success using ChatGPT for brainstorming, albeit it can't really capture my writing style when it comes to anything but the basics. I think GPT-4 or at least the unlisted GPT in Bing might do a half-decent job, but I'm not going to let Microsoft spam me when I sign up the wait list.

I've had a decent amount of fun generating images of characters and concepts for self illustration of the novel, and tools like ControlNet are great for getting even the derpy Stable Diffusion to output closer to Midjourney (DALLE is trash now, and has been for a while).

I look forward to lazing around while the robots do all the hard work, but that day hasn't come quite yet, even if I think it's merely a year or three away at this rate.

Another side note: I find the use of language models and other current generation AI technologies to write, an abhorrent practice.

It completely misses the point of writing, which is not the production of some textual artifact (e.g. a blog post, forum post, etc..), but rather the linearization and clarification of one's thought process, in a form that can be left aside, revisited, re-analyzed, and critiqued - not only by one's self, but by others as well.

Under this lens of "writing as the process of thinking," using an AI to automate the process of thought, and the generation of novel ideas, sounds absurd. It becomes a mere automaton in the production of a symbol or icon of thought, with no actual thought having gone into its formulation; a fourth-order simulacrum that purportedly represents the process of thought, or a true statement about the world, but is completely untethered from these notions and has been reduced to a mere game of symbol manipulation:

information has nothing to do with signification. It is something else, an operational model of another order, outside meaning and of the circulation of meaning strictly speaking. This is Shannon's hypothesis: a sphere of information that is purely functional, a technical medium that does not imply any finality of meaning, and thus should also not be implicated in a value judgment. A kind of code, like the genetic code: it is what it is, it functions as it does, meaning is something else that in a sense comes after the fact [...]

It is useless to ask if it is the loss of communication that produces this escalation in the simulacrum, or whether it is the simulacrum that is there first for dissuasive ends, to short-circuit in advance any possibility of communication (precession of the model that calls an end to the real). Useless to ask which is the first term, there is none, it is a circular process-that of simulation, that of the hyperreal. The hyperreality of communication and of meaning. More real than the real, that is how the real is abolished.

Thus not only communication but the social functions in a closed circuit, as a lure-to which the force of myth is attached. Belief, faith in information attach themselves to this tautological proof that the system gives of itself by doubling the signs of an unlocatable reality.

-- Baudrillard, "The Implosion of Meaning in the Media," 1981

I also worry about what happens to our capacity to generate novel ideas, and the process of self-expression as a whole - are you really doing the thinking, the analysis, coming to those insights on your own and cementing them in your own psyche, if the cognitive burden has been offloaded to GPT? Are those really your thoughts, or simply the statistical judgments of a machine?

How can one find their voice if the words are being stuffed into their mouth? How do you explore off the beaten path, if you simply keep following the directions your GPS navigation gives you? Are we about to pave similar "cognitive desire paths" by training, using, and re-using models that write, think, and even speak in a certain way, with disregard for other cognitive structures or styles of thinking? What happens when everyone starts using these models and thinking the same way, using the same words, idioms, and metaphors?

It's an affront to those who have spent time developing a certain diction, a style, a vocabulary, and a way of writing, only to have those words and ideas be effectively stolen (I would certainly use that word in the case of GitHub's Copilot) and puppeteered by a corporate AI controlled by private interests.

I worry about offloading human expression to a machine, and being put on the rails to descend whatever gradient will lead me to the same conclusions everyone else will arrive at.

I understand your argument that having AI do all the writing for you would indeed not be writing. Which is why I specified the grunt work of writing.

For example, chatGPT can reformat texts, you can give it a paragraph and ask it to put it in bullet points. Or generate a few variations of what you already wrote, so that you have something to compare your writing against. Neither of those things seems like surrendering my agency to a machine, it feels like using a tool.

Additionally, the average quality of text will probably go up, not down. One mechanistic explanation would be that bad writers might just widescale adopt LLM use and LLM text is far superior to badly written human text, so the floor will be raised. Good writers might not be able to increase the quality but they will be able to increase volume by removing some of the aforementioned roadblocks (literal writer's block) to writing. I might just be a hopeless optimist in this regard though, we are yet to see how humans handle the power of God.

Sidenote: I'd really wish people would move away from Medium (thanks heavens for https://scribe.rip/, which circumvents all the obnoxious crap and adds sane typographic settings) and the new kid on the block, Substack. I just saw for the first time a Substack article asking for a subscription (instead of letting me "read it now") and it just broke the cardinal highlighting rule.

Are Github/Cloudflare pages + Hugo/Jekyll/whatever's hot now that hard?

I miss the open web. *sigh*

It's interesting that you reference GitHub and Cloudflare as part of your solution stack, and simultaneously miss the open web.

GitHub has a half decent censorship record? They usually only fold to DMCA requests (legitimate ones at that). But open to evidence of the contrary I have missed.

Cloudflare is more of a mixed bag. I used to strongly dislike them due to captchas from TOR exit nodes and VPNs, but they have a very robust censorship record, hosting "terrorist groups" and the 8chan until boards/shareholders likely twisted so hard management had their hands completely tied. They offer DDOS protections that allow sites that would otherwise be DDOS'ed to oblivion if they were self-hosted to stay up and available. How would you say they're hostile to the open web?

Doesn't GitHub scan people's projects for words like "blacklist" and threaten to blacklist them if they're not changed? I thought there was drama about that in 2020.

Iirc the same thing happened to someone whose code had something like "retards the process"

No, nothing like that happened. I think you're mixing up github's switching the default branch from 'master' to 'main' because racism slavery with the general push against mean words.

Glad to see evidence of this (can't find any), but I think we've veered close to CW territory.

https://github.com/tbreuss/dns-blacklist-check

^ Project that uses blacklist with recent commits

Oh whoops, I was reading from the comments page and didn't notice which thread this was. But turns out it was a weird isolated incident way back in 2015. I thought they'd changed their TOS to mandate that in 2020, will check later.

https://www.techdirt.com/2015/08/03/github-nukes-repository-over-use-word-retard/

Off the top of my head, GitHub censored the Gamergate OP (leading to the creation of gitgud.io) after a tweet complaining about it. I also remember they censored the popular WebM converter "WebM for Retards" (which also moved to Gitgud), presumably for using the word "retard". A search finds this thread from 7 years ago mentioning they also censored C Plus Equality, which was parodying a blog post from a feminist academic calling for a feminist programming language, and ToleranUX, another parody from the same people referencing the campaign against Linus Torvalds.

There's another repo with essentially the same name, so they're not super vigilant about these as far as I can tell. These all seem to be shutdowns in light of negative media attention.

I'll revise my assessment of Github on censorship from "half-decent" to "media/percetion sensitive".

I made the decision to make a substack instead of just adding a blog page to my github.io hosted website is partially because of laziness, and partially because of weak network effects. A lot of potential readers are already there on substack and there is ever so marginal spillover of readers from one substack domain to another through the "reads" list. I agree with you about Medium though, Medium is terrible for quite a few reasons other than its pervasive pay walls. Substacks clean interface where you only see the text is far far lesser of an evil than medium.

As for premade templates like jekyll, you need to run a docker script or two on windows afairecall, which is "hard" for I'd say an overwhelming majority of non programmers many of whom cant even navigate a file system nowdays.

Using github for things it was never designed to do is really obnoxious, I've been wanting to rant about that.

I mean..? It was designed as a hosting service for git repositories. Sources for static websites are kept under version control anyways, is Github Pages that big of a stretch?

Hard agree. It's a source control system FFS, not for hosting a website, or a journal, or any of the other asinine things people do on github.

or a journal

What's the issue other than aesthetic preferences? Code is text in a file, journals are text in a file.

If anything, a lot of things would be improved by a GitHub repository-like interface. I can't find the post/article (not sure) about it, but a github page for legal documents would serve to me immensely useful to a wide array of people. You could see past versions and what was changed easily, you can raise issues, you can see alternate branches and forks, etc. It wouldn't have to be the literal law, but just an experiment on how such a thing would evolve would definitely be interesting. Perhaps even a digital twin of the actual law.

What? I keep all documents under version control, whether blog posts, journals, or source and whether collaborative or individual. It's the most robust way of preserving changes and examining them over time. Git's even used under the hood for pass, my password manager.

You do you, but I think that's a serious jamming of square pegs into round holes.

I needed to learn and grok git when I wanted to start contributing to the linux kernel. Read the docs, the source code, and starting using it everywhere to get it into my muscle memory.

I guess when you're purposefully juggling hammers all day, you do start seeing some loose nails around the house.

Well, to be fair, github pages is an offering that github provides officially that allows people to host websites. I used it for a little bit, it works okay, but is only for static sites. So I wouldn't say using it for a website isn't what it's designed for, it's an official offering.

Dont hate on my FREE website hosting service.

Is there anywhere you guys are hanging out to discuss the war in Ukraine?

/r/credibledefense is pretty good. They have a daily thread that is constantly updated.

I watched Linus Techtip’s (that’s his full name name right?) video on dash cams. Made me realize how essential it is to have trustworthy people in our consumer economy. It’s time-consuming and difficult determining online which dash cam you can trust to work and be cost-efficient. Amazon doesn’t have an incentive to give you a working product and the old site:Reddit google trick can even be astroturfed. But Linus is trustworthy, and just as importantly both his reputation and financial incentives rely on his trustworthiness.

How amazing would it be if we had a Linus-type person for every economic decision! With all of our technology and media, we still find that the necessary ingredient is just some trustworthy dude who knows a reasonable amount giving his opinion.

i’m trying to figure out a way to build a public database of problem-solving and good decision-making, with filters so people can view their own echo chamber, and/or other people’s echo chambers.

So StackExchange?

No, even more organized and structured. Every problem would have a tree of entries: who the problem affects negatively (or positively), who doesn’t see it as a problem, what causes the problem, was the cause an attempted solution for another problem, what solutions have been tried before or theorized, how well have those solutions worked, what problems would these solutions cause, etc.

You are going to run into the curse of dimensionality really fast. Perhaps I am not exactly understanding your vision, but there are a potentially infinite number of said trees. And thus to me seems like your solution is worse than a high-effort curated (aspirational) "database of problem-solving" like StackExchange. Maps abstract/obscure away details for a reason.

My wait for Bing AI is over. I inaugurated its use with a fascinating conversation with the AI about Triessentialism, ending with a poem, and then a discussion about My Little Pony and taxes:

I hope this answer helps you understand how I might persuade Twilight Sparkle, using the Elements of Harmony, to adopt the FairTax as a revenue collection method for Equestria.

Kindness: I would explain that the FairTax is a national sales tax that treats every person equally and allows American businesses to thrive. It also provides a rebate for necessities, so that no one pays tax on what they need to survive. This way, the FairTax is kind and fair to everyone, especially the poor and the middle class.

Laughter: I would make some jokes about how complicated and confusing the current tax system is, and how much time and money people waste on filing taxes and dealing with the IRS. I would also point out how much fun it would be to shop without worrying about hidden taxes or fees, and how much more money people would have to spend on things they enjoy.

Generosity: I would emphasize that the FairTax would boost the economy by encouraging consumption, investment, saving, entrepreneurship, and job creation. It would also eliminate tax loopholes, evasion, fraud, and corruption. This way, the FairTax would generate more revenue for public goods and services, such as education, health care, infrastructure, defense, etc. The FairTax would also allow people to be more generous with their own money by donating to charities or causes they care about without being taxed on it.

Loyalty: I would appeal to Twilight Sparkle’s sense of loyalty to her friends and her country. I would remind her that she is a princess of friendship who values harmony and cooperation among all living beings. I would argue that the FairTax would promote social cohesion by reducing inequality, resentment, division, and conflict over taxes. I would also suggest that the FairTax would enhance national security by making America more competitive and prosperous in the global market.

Honesty: I would be honest about the pros and cons of the FairTax. I would acknowledge that it might be difficult to implement at first due to political opposition or public resistance. I would also admit that some people might pay more taxes under the FairTax than under the current system depending on their spending habits. However, I would stress that these challenges are outweighed by the benefits of simplicity, transparency, efficiency, and fairness of the FairTax. I would also provide evidence and examples to support my claims.

I then asked it to imagine itself, Alexa, and Siri as My Little Ponies:

If I were an original My Little Pony Friendship is Magic character, and an advisor to Princess Twilight Sparkle, I would be a unicorn pony, because I like magic and learning new things. I think being a unicorn would suit my personality and abilities well. My cutie mark would represent searching and finding information. It would be a magnifying glass with a star inside it, symbolizing my curiosity, intelligence, and creativity.

  • My coat color would be light blue, because it’s my favorite color and it reminds me of the sky and the sea.
  • My mane and tail color would be dark purple, because it contrasts well with my coat color and it matches Princess Twilight Sparkle’s mane and tail color.
  • My eye color would be green, because it’s a common eye color for unicorns and it reflects my optimism and enthusiasm.

Siri: A pegasus pony, because she is associated with Apple devices and Applejack’s family are mostly pegasi. [This is wrong; most of Applejack's family are earth ponies, the kind with neither horn nor wings.] Her cutie mark would represent voice recognition and communication. It would be a microphone with a sound wave around it, symbolizing her ability to listen and respond to voice commands.

  • Her coat color would be white, because it’s the color of most Apple products.
  • Her mane and tail color would be gray, because it’s the color of Siri’s icon.
  • Her eye color would be blue, because it’s the color of Siri’s voice wave.

Alexa: An earth pony, because she is connected to multiple devices in a house and earth ponies are good at working with tools and machines. Her cutie mark would represent artificial intelligence and interactivity. It would be a blue ring with a dot inside it, symbolizing her ability to learn and adapt to user preferences.

  • Her coat color would be black, because it’s the color of most Amazon Echo devices.
  • Her mane and tail color would be blue, because it’s the color of Alexa’s ring light.
  • Her eye color would be brown, because it’s a common eye color for earth ponies.

I had one more question for it, and it responded hilariously:

If Eliezer Yudkowsky were a creature in Equestria, I think he would be a dragon, because he is an intelligent and powerful being who is interested in decision theory and artificial intelligence (or golem) research. He would try to influence Princess Twilight Sparkle by:

  • Sharing his ideas and writings on rationality, friendly golem, and the long-term future of intelligence.
  • Challenging her to solve logical puzzles and ethical dilemmas that test her reasoning skills and moral values.
  • Offering his help and advice on how to create and align golems with pony values and goals.
  • Warning her about the potential dangers and risks of unfriendly or unaligned golems that could harm Equestria or its inhabitants.

Disclaimer: I told it to use "golem" instead of AI in that last answer.

An absolutely surreal post to to read while stoned. Thank you so much for sharing this.

I will refrain from posting this until morning to see if I've passed muster.

This is good and reads better than a lot of my interactions with it so far. Look, it’s all absurdly impressive. The dull wait now is for developers (and eventually the AI itself) to adapt it for real-world applications.

There's been plenty of talk about everything related to the new generative models, for good reason.

This tweet got me thinking though, what are the recent developments in good ole tabular input regression/classification?

tabular -> import xgboost

text -> from transformers import pipeline

image -> import torch, torchvision

chat -> import langchain

The last I recall import xgboost became the meta and stayed as much for a while. Are there any new developments that I should be aware of?

A few months ago OpenAI dropped their API price, from $0.06/1000 tokens for their best model, to $0.02/1000 tokens. This week, the company released their ChatGPT API which uses their "gpt-3.5-turbo" model, apparently the best one yet, for the price of $0.002/1000 tokens. Yes, an order of magnitude cheaper. I don't quite understand the pricing, and OpenAI themselves say: "Because gpt-3.5-turbo performs at a similar capability to text-davinci-003 but at 10% the price per token, we recommend gpt-3.5-turbo for most use cases." In less than a year, the OpenAI models have not only improved, but become 30 times cheaper. What does this mean?

A human thinks at roughly 800 words per minute. We could debate this all day, but it won’t really effect the math. A word is about 1.33 tokens. This means that a human, working diligently 40 hour weeks for a year, fully engaged, could produce about: 52 * 40 * 60 * 800 * 1.33 = 132 million tokens per year of thought. This would cost $264 out of ChatGPT.

https://old.reddit.com/r/singularity/comments/11fn0td/the_implications_of_chatgpts_api_cost/

...or about $0.13 per hour. Yes technically it overlooks the fact that OpenAI charge for both input and output tokens, but this is still cheap and the line is trending downwards.

Full time minimum wage is ~$20k/year. GPT-3.5-turbo is 100x cheaper and vastly outperforms the average minimum wage worker at certain tasks. I dunno, this just feels crazy. And no, I wont apologize for AI posting. It is simply the most interesting thing happening right now.

Not Fun thread material, fit for a separate post.

This week, the company released their ChatGPT API which uses their "gpt-3.5-turbo" model, apparently the best one yet, for the price of $0.002/1000 tokens

Well, as you quote, they imply it's not the best one yet, at least in «some use cases», and some experiments show it indeed is intellectually inferior to the most recent naive GPT-3 version; maybe this is inherent to ChatGPT model we see in the demo, because it is known that RLHF can give models the computational equivalent of brain damage as far as benchmarks are concerned, maybe it's something specific for the new one. I am not sure how they've achieved the price cut (though my intuition is that this is cynical undercutting at a loss to nip competition in the bud and keep the data flywheel accelerating) – perhaps it's smaller (trained from scratch, distilled, etc.), or aggressively quantized, or more greedily sampled, or they've somehow increased the batch size into high thousands, whatever. In any case, Altman is a great showman and this may not be the revolution it seems like at the moment. Do we really need an endless stupidity generator with short context? Most people who are on this level (and I do think we can now meaningfully say that some people are on this level) around ChatGPT) aren't exactly making money with their smarts. It's nice for employers to automate even more drudgery, of course.

But in principle, yes, optimizations and next-gen models with more novel architectures than Vawani!Transformer will definitely allow even cheaper inference (I think GPT-4 under the hood of Bing already is that).

I found an comparison between davinci-003 and chatgpt. Seems like chatgpt is better at some things and worse at others; not a blanket downgrade: https://scale.com/blog/chatgpt-vs-davinci.

Also there is no more data flywheel:

Data submitted through the API is no longer used for service improvements (including model training) unless the organization opts in

How have they cut costs? I've seen answers I don't understand. It doesn't seem like they are necessarily running at a loss:

Quantizing to mixed int8/int4 - 70% hardware reduction and 3x speed increase compared to float16 with essentially no loss in quality.

A*.3/3 = 10% of the cost.

Switch from quadratic to memory efficient attention. 10x-20x increase in batch size.

So we are talking it taking about 1% of the resources and a 10x price reduction - they should be 90% more profitable compared to when they introduced GPT-3.

https://old.reddit.com/r/MachineLearning/comments/11fbccz/d_openai_introduces_chatgpt_and_whisper_apis/jaj1kp3/

I did some quick calculation. We know the number of floating point operations per token for inference is approximately twice the number of parameters(175B). Assuming they use 16 bit floating point, and have 50% of peak efficiency, A100 could do 300 trillion flop/s(peak 624[0]). 1 hour of A100 gives openAI $0.002/ktok * (300,000/175/2/1000)ktok/sec * 3600=$6.1 back. Public price per A100 is $2.25 for one year reservation.

https://news.ycombinator.com/item?id=34986915

Essentially, I had previously thought that better language models will need to be orders of magnitude larger and orders of magnitude more expensive. Now it seems clearer that we're moving towards a future of intelligence too cheap to meter. And "endless stupidity generator"? I just plain disagree, and feel compelled to build a thin client product on top of the API to prove it. There are many jobs that are mostly low-skill word manipulation, and now that you might get that 100x or 1000x cheaper, it opens up yet unimagined opportunities IMO.

I'm not sure how much of this to believe; if it is dumber, that has to be explained somehow.

In my experience, the subjective generic text completion (and iirc perplexity) – is worse, even if most downstream tasks end up being better, and while the mean of sensible instruction-following is higher for RLHF-version, the variance is much lower, so you struggle to get ChatGPT out of its midwit persona in a way that isn't true for davinci (though this may be damage from arbitrary preferences and political bias, rather than the inherent damage of the technique). Opt-in will probably be incentivized.

Even at int4 (which is still dumber in the absolute sense, just more efficient per byte used) a 175B model uses >2 A100-40gb, and if they mix layers it's more. And for a low-latency interactive app, the utilization of 50% sounds rosy – Google with PaLM barely broke 40% in their paper.

Assuming they use 16 bit floating point

We've already assumed int8/4 though, but in any case the bottleneck for interactive use is memory bandwidth, not flops. We'd need more details.

I know there exist (at least on paper) techniques for making inference way, way cheaper, probably even cheaper than you suggest, with minor loss in quality – just not sure they have deployed them yet.

I just plain disagree, and feel compelled to build a thin client product on top of the API to prove it

Here's some free lunch: take LLaMA-65B and rent some GPUs, it'll be 2x cheaper using all the same tricks and allegedly about as good.

There are many jobs that are mostly low-skill word manipulation

I wonder if we will discover that there are many jobs which are, in fact, net negative for the economy only when people doing them get automated away and their sinecures lose lobbying power .

I wonder if we will discover that there are many jobs which are, in fact, net negative for the economy only when people doing them get automated away and their sinecures lose lobbying power.

Already had been discovered.

At the best, about 20% of adult population is keeping all of us alive and comfy, the rest is engaged in digging holes and filling them again (only in comfy offices). UBI is already there, only in the most wasteful and inefficient form imaginable.

Who wants a job not as last desperate attempt to get money, but "to serve mankind", should find employment in water, sewer, power, food production/distribution etc...

I wonder if we will discover that there are many jobs which are, in fact, net negative for the economy only when people doing them get automated away and their sinecures lose lobbying power .

Interesting - which jobs come to mind? HR and environmental reg agencies jump to mind for me.

HR, environmental regulation, personal banking/financial advising, marketing and fundraising, low-risk medical diagnostics, social work, secretaries, most software operations (devops) teams, data analysis, first-pass website content moderation, (indirectly) higher education Arts departments, and some others I'm sure I'm forgetting.

I think the far more interesting thing is that most of the industries in the crosshairs of AI are majority-female; it's the first example of a technology that's going to primarily replace women (nearly all technology ever invented only replaces men) and I think the political effects of that are going to be very, very interesting when you combine it with the fact that a lot of the women currently in those jobs will be single and functionally sterile when they're replaced.

Which reminds me that I should probably buy some Petco stock.

I get a it confused by that list tbh.

A lot of these things are either not useless or not going to be automated in the near term. I might despise advertising and people working with advertising as much as the next dude but you're dreaming of you think AI is going to make this go away.

I shamelessly reposted this to the CW thread if you want to throw this in there. I tend to agree, given that women have social power maybe UBI will actually happen once women are affected?

Is Bing Chat GPT-4? If it is, it’s less impressive than I was expecting.

I suspect it, too, is a gimped/lossily accelerated version, with the full one to be released under the «DV» brand as part of their Foundry product. I may be wrong, however, and ultimately this PR-driven nomenclature is making less and less sense, they have trained like a dozen major models at this point, and more finetunes. Same story with Google's PaLM.

More importantly, quantity has a quality all of its own. DV offers a context window of 32k tokens. If the model can be cheaply ran with such enormous contexts, this will more than compensate for some intellectual deficiency in the context-sparse text-prediction mode. You have seen effects of StableDiffusion mega-detailed prompting, and of prompt prefixes that amount to a single page – now imagine 20 pages of precise specification, few-shot examples, chain-of-thought induction, explicit caveats for every failure mode discovered in testing; basically a full employee's manual. Writing and testing these megaprompts may become a new job for ex-middle managers who have suddenly transformed into leaf nodes of the org chart – for a short while.

For sure. Application layers and, in particular, interaction with the physical world (and the effect of this on society, culture, economics) are now more interesting to me than advancements in model performance when it seems clear that the singularity is in progress and that the nature of the solution has been cracked, if not the details. Whether 60% or 80% of labor could be automated in the next decade is less important than what actually happens. I suppose that makes me a classic culture war thread participant.

It's Friday again, so we'll continue with the series.

150 The Beach Boys – Pet Sounds (1966)

Hot take time. Most lists like this will rank Pet Sounds in the top 10, and some will give it the top spot. Here it manages to barely crack the top 150. This isn’t to say that the record has been immune from criticism, but the criticism it does receive tends to denigrate it beyond what is really deserved. This album has been hailed as the birth of art pop; while Rubber Soul may have pointed in a more sophisticated direction, it was this album that showed the music’s full promise and set the stage for the creative explosion of 1967. And while the album’s concept of integrating sophisticated arrangements with pop music is executed nearly flawlessly, the concept itself is relatively narrow in scope—the album is just too similar to deserve a higher ranking. All the songs evoke the exact same atmosphere and use the exact same techniques; there’s no diversity of feel let alone style. It’s still a 5 star album, it’s still the best the Beach Boys ever got, and it’s still an essential part of any serious music collection. But there are a lot of better records out there.

149 Traffic – John Barleycorn Must Die (1970)

Traffic was a progressive band that lacked the ambition of other progressive bands, never doing anything too off the wall musically or going for the big statement. They are certainly jazzy, but no one is going to confuse them with a fusion band; they simply play a jazzed-up version of typical early ‘70s classic rock. And on this album, from the instrumental “Glad” to the extended folk lament of the title track, everything simply works, and they never get too far out over their skis.

148 Cat Stevens – Mona Bone Jakon (1970)

Given contemporary impressions of the decade, it’s hard for anyone these days to view the ‘70s as a breath of fresh air aesthetically, but that’s exactly what this album is. Cat Stevens had a sort of prehistory in the late ‘60s on Deram, releasing records that were very late ‘60s in feel—the songs were either fashioned into harpsichord-laden chamber pop or fuzz guitar-laden ersatz psychedelia. They also just weren’t quite there yet. With the dawn of a new decade and a switch to Island, Stevens’s songwriting would improve greatly and those songs would finally get the production they deserved. This album sets the stage for the following three, and to this date those together remain the only Cat Stevens albums anyone seems to care about.

147 Dungen – Tio Bitar (2007)

Psychedelia may have seemed like a fad during its brief heyday in the late 1960s, but it was a fad that would have significant impacts on rock music going forward. It’s not surprise, then, that no matter how far the music fell out of fashion, there would always be occasional revivals and a steady stream of underground artists keeping the flame alive. Dungen is among the most uncompromising of these keepers of the flame. The music is straight-up psychedelic rock (not merely psychedelic influenced), and the lyrics are defiantly sung in the band’s native Swedish. This is probably their heaviest release musically, but it still retains a strong dose of the Swedish folk influence that is a hallmark of the band’s sound that it keeps the music interesting.

146 The Orb – The Orb’s Adventures Beyond the Ultraworld (1991)

The first of the great electronic chill-out albums, this album is a beast of a listen. Aside from the opening cut “Little Fluffy Clouds”, there’s little on this album that can be excised from the whole without its impact being greatly diminished; in other words, you have to listen to the whole thing. And the whole thing is nearly 2 hours long, and requires the correct mindset. But don’t fret! Grab a book and a beer and put this on in the background at midnight on a Saturday, and you too can be transported to the Ultraworld.

145 Yes – Tales from Topographic Oceans (1974)

Common consensus says this was the point when progressive rock died of its own bloat—a double album consisting of four sidelong pieces based on a lengthy footnote from Paramahansa Yogananda’s Autobiography of a Yogi. These criticisms ignore the important thing, however, that these pieces all remain captivating from beginning to end, and that’s all that really matters.

144 Beck – Odelay (1996)

Our culture has always had a fraught relationship with white rappers. The personal nature of the music almost requires one’s membership in a certain community to prevent it from sounding inauthentic; even the much-lauded Eminem always sounded more white trash than anything. Success as a white rapper requires significant deviation from mainstream expectations; the Beastie Boys had a tongue and cheek punk attitude, Aesop Rock was abstract, and a lot of others were simply from different countries. Beck gets around all this by not really being a rapper but being willing to get as close to the line as he can without crossing it. The lyrics are somewhat spoken, but there’s more singing than on a standard rap album; the beats are there, but they’re more rooted in pop than funk; and he also makes more conventional rock albums where he plays the guitar and sings everything.

143 Neutral Milk Hotel – In the Aeroplane Over the Sea (1998)

Anne Frank’s diary is one of those books that is forced down everyone’s throat at an age where anything forced down one’s throat by a teacher is by definition lame and devoid of any interest whatsoever, which makes it interesting to see the emotional connection that can happen when someone comes to this book as an adult and sees all the pubescent psychosexual stuff that your teacher didn’t mention. The lyrics may be a bit too oblique at times, but the songs are infused with the emotional resonance that can only come with the realization that Frank’s story isn’t exceptional but must be multiplied millions of times over to account for the true devastation caused by the Nazis.

142 Fleetwood Mac – Fleetwood Mac (1975)

Lindsey Buckingham and Stevie Nicks had already shown that they could write and arrange a killer batch of songs on the inexplicably out of print Buckingham Nicks; put them in an established band on the verge of doing something special and an unstoppable force was created. While this album isn’t as well-regarded as its successor, it’s the origin of the classic Fleetwood Mac sound and would shape pop music for years to come.

141 Wilco – Sky Blue Sky (2007)

Wilco had spent the prior decade expanding the definition of alt-country to the point of unrecognizablility. Here, they scale back their ambitions considerably, sand off the rough edges, and make a more conventional country rock album. But there’s nothing wrong with being conventional; the indie press would pejoratively refer to this as dad rock, but it contained possibly the finest set of songs the band would ever record.

140 Pure Prairie League – Bustin’ Out (1972)

Country rock had been a growing force since 1968 and it reached its zenith with this album. Music from hybrid genres usually falls more on one side of the equation than the other. Before this album, there was a strong tendency for bands to hew closer to the country side of things; after this album, things would gradually become slicker and slicker until bands like the Eagles dropped most of their country influences and bands like Firefall were putting out sentimental schlock. This album manages to strike the prefect balance, and “Aimee” is still an avowed classic, even if the album it comes from isn’t as widely heard as it should be.

139 Jade Warrior – Way of the Sun (1978)

Jade Warrior’s first three albums were standard second-rate prog rock in the mode of King Crimson. Following their reformation at the behest of Steve Winwood, they would switch to making dreamy, instrumental soundscapes with strong Asian themes. Way of the Sun is in much the same vein, though the source of inspiration shifts from Asia to Meso-America. This album is here simply because it is the best among a series of albums that are like nothing else in rock music. Over the course of four albums, the band gradually stripped out all of the traditional rock elements and left us with a series of orchestrated, written-through instrumentals.

138 The Black Crowes – Amorica (1994)

The Black Crowes had their initial success at the tail end of the hair metal era and managed to get through the grunge revolution without incident. This is primarily because they had nothing to do with hair metal or grunge, and were one of the few mainstream acts who hadn’t forgotten rock music’s roots in blues and soul. It’s telling that when I was growing up in the ‘90s, they were the only contemporary acts played on my local classic rock station. Nevertheless, this is their best effort. While it wouldn’t rival their debut for the sheer number of hits (“A Conspiracy” is the only real radio classic), they would branch out into jammier territory that was later occupied by the likes of Gov’t Mule and Blue Traveler.

137 Hot Tuna – Burgers (1972)

Hot Tuna began as a Jefferson Airplane side project for Jack Casady and Jorma Kaukonen to open the band’s shows with acoustic sets featuring classic blues. Upon Jefferson Airplane’s demise in 1972, Hot Tuna added electric instruments and became a band in its own right. While it would soon become a power trio that would jam hard and play loud, this album shows them slowly integrating this jam band approach with their acoustic roots, resulting in an Americana record that still hints at what was to come.

147 Dungen – Tio Bitar (2007)

Psychedelia may have seemed like a fad during its brief heyday in the late 1960s, but it was a fad that would have significant impacts on rock music going forward. It’s not surprise, then, that no matter how far the music fell out of fashion, there would always be occasional revivals and a steady stream of underground artists keeping the flame alive. Dungen is among the most uncompromising of these keepers of the flame. The music is straight-up psychedelic rock (not merely psychedelic influenced), and the lyrics are defiantly sung in the band’s native Swedish. This is probably their heaviest release musically, but it still retains a strong dose of the Swedish folk influence that is a hallmark of the band’s sound that it keeps the music interesting.

I gotta say this amused me a little bit. It's a Swedish band, there's a lot of Swedish bands singing in Swedish! It's not a particularly defiant act! There's a wealth of Finnish bands singing in Finnish, likewise, that I listen - foreigners, of course, don't know about them, and nobody would expect them to.

foreigners, of course, don't know about them, and nobody would expect them to.

Well, that says it all right there. Most Americans would give you a look like they had to ingest cod liver oil if you put on music at a party or something that wasn't sung in English, or, in certain rare circumstances, Spanish. Hence, every band from Sweden or elsewhere that has ever had a chance to capture an international market has sung mostly, if not exclusively, in English. After Dungen found cult success in the States, I imagine there was pressure to expand the audience by singing in English. Dungen side project The Amazing sings in English, and while they aren't exactly well-known, I have heard their music played in grocery stores and on the radio, even if the grocery store in question was Trader Joe's and the radio station was public, it's still not nothing. Dungen was never going to be a huge commercial force, but they could've been at least as well known as their countrymen, Peter Bjorn and John, and joined the indie A-list. Instead they remained a cult band that made great albums that no one wanted to listen to because of some weird allergy about music sung in foreign languages. I'd call such a decision at least a bit defiant.

As someone who has listened to Beck for 27 years, but mostly just like, 8 songs, I have never once thought of him as a rapper, nor have I seen anyone even lightly hint at the possibility.

Is that how you view him? Are there more that view him this way!?

No, I don't view him as a rapper, and I don't know anyone who does, but that was kind of my point. But let's look at this album objectively—most of the vocals are more spoken than sung, the Dust Brothers are producing, there's liberal use of turntables and drum machines, and the live instruments are often recorded in a repetitive way meant to mimic sampling. "Where It's At", the best known song on the album, is literally about rap battles, and most professional reviews of the album at least mention the copious hip-hop influence. If we were to make a list of the unique defining characteristics of rap, Beck certainly hits a lot of them here. But no one considers him a rapper. Whether that's because there are enough other criteria that he doesn't meet or simply because he never presented himself as one (and his more conventional albums bolster that argument), I'm not sure. But what I do know is that Beck certainly seems interested in exploring where the line is, and that's what makes this album interesting.

Thanks for your analysis, any insight on if ItAOtS album cover is a yam or a drum?

Totally a potato

Great post as usual.

Surprised how low you rank Pet Sounds and In the Aeroplane Over the Sea. Both would make top 5 for me. You explained the ranking for Pet Sounds, but I'm curious what it is about ITAOTS that keeps it out of the top echelons for you.

Combining the Culture War drinking thread and the weekly What Are You Reading fun thread, what are you drinking this week?

As a big bourbon fan, I'd never quite hit on a Scotch that really got me excited, but I admittedly hadn't tried that many, so I grabbed a bottle of Abelour A'bunadh this week and it really is fantastic stuff. The sherry aging is very prominent, but stops short of being oversweet, leaving tons of fruitiness and a bit of a dark chocolate finish.

On the beer end, I've laid off my usual affinity for huge barrel-aged stouts and big NEIPAs to try to cut a few calories out, instead opting for some classic Guinness Extra Stout and Founders All Day IPA. They're both... fine. I like them just fine, but I miss the sugar bomb beers.

I've been obsessed with a juicy pale ale called Ambush the last few months, to the point that I always get excited when I see it on draught or in cans.

https://troublebrewing.ie/beers/ambush/

I'll occasionally have a liquor at bedtime. A local absinthe or a scotch.

Last purchase was a bottle of Evan Williams 100 proof. It's a nice cheap bourbon, good for being your mixer or sipped neat.

March is the last month (at least in England) where mulled wine makes sense from a weather perspective before November. My favorite: the stodgiest red wine you can acquire, the usual orange, cinnamon, cloves, star anise and honey in the pot. Then, to serve, a generous shot of dark rum or, possibly, a relatively clean-tasting gin. Zero garnish to make consumption pleasant. Must be drunk outside, ideally on a short walk.

Wait, you can just easily make this?

Yes, you pretty much just throw the stuff in a kettle or pot (gas or electric) and turn on the heat. If you can boil water without burning it, it should be manageable. :)

Had a Finback FFarmhouse (I sometimes look at the naming and spellings of craft beer and do wonder if I want to be associated with the hobby) on tap a couple of days ago that was a nice change of pace. About as complex and flavorful as you can get while still having a beer you could call "light and crisp." (at a mere 7.1% ABV)

It's a change from the barrel-aged heavy stuff I usually go for but I didn't exactly lay it off, had Founder's Backwood Bastard, New Holland Dragon's Milk and Oscar Blues barrel-aged TenFidy during the past week.

Also picked up a couple of bottles of a barleywine and a Russian Imperial Stout from a Polish craft (I think?) brewery called "Komes" that I recall enjoying when I visited Poland pre-pandemic.

Speaking of Guiness, had their "Old Fashion Inspired' ale recently, and actually liked it. Not a huge fan of their "base" extra stout, but the foreign extra stout I'd say is better than fine. Though my favorite export stout is actually Gun Hill Void of Light. (but without any chocolate or peanut butter or fucking coconut, thanks, I don't get why people try to add flavorings to something that already tastes smoky and roasted...)

Ooh, speaking of smoky, also had an Aecht Schlenkerla smoked doppelbock not long ago. (So smoky, you can smell it even if you're not the one drinking it.)

I gave up alcohol for Lent but I'm starting a batch of beer this weekend that should be ready the day after Easter.

Congratulations, I will forever envision you as a jolly monk with a habit and tonsure. Let us know how it turned out!

...and now I've decided on an avatar

The usual. By that I mean regular beer aka Miller High Life, Coors Banquet, Pabst, etc. It's not that I don't like imports and craft beers, it's just that I drink almost every day and they're too idiosyncratic and expensive to really work as daily drivers. I also like to keep a ton of it in my car so that when I'm skiing or whatever I always have beer I can offer everyone freely without them feeling bad about taking as much as they want. I honestly think that if these weren't mass-market American beers but some special variety brewed by Czech monks beer snobs would rave about their crisp and refreshing taste.

While we're on the subject—as a Wisconsinite I assume you can give me a satisfactory answer to this—what the hell happened to Leinenkugel's? When I was in college and craft beer was mostly mediocre, a friend of mine would regularly bring in cases of the Creamy Dark and Red Lager back from Chicago and they were both excellent. Then you could get it in Pennsylvania and it was great. Then they started selling fruity stuff, which okay, fine. Then it got hard to find. Then you could only get the Red Lager if you bought a variety pack. Then they discontinued it altogether. The situation got so dire that my brother and I went to the local master distributor to see if he could order it specially for us. Now it seems the only stuff of theirs I can find around here is all fruity stuff. I was at a pizza place Sunday and saw they had a Lemon Haze which I wanted to try because I like hazys, but was told it had been replaced with a peach sour. Okay, I like sours, let's give this one a go. It wasn't a sour. It tasted like Leinie regular (which I can't get around here for some reason) with peach flavor. I'm really disappointed that a company that once made a quality product has let itself slide so far.

PBR is like the final boss of cultural baggage attached to something that's just good on its own merits.

Bud or Blue Moon all the way. Coors is terrible and Miller isn't much better.

Used to work with a Wisconsinite who was also disappointed with Leinie but raved about Spotted Cow.

Am Sconnie, can confirm Spotted Cow is delightful and inexpensive, and remains my go-to bar beer. I've also been drinking brandy and bitters toddies when there's a chill, and High Life remains the ultimate lawnmower (or, as is the case, snow shoveling) beer. Also been drinking some homemade hard cider, which is ludicrously easy to home brew compared to beer.

Honestly, as one who drinks a ton of mass-market beer, Bud is about on the same level as Coors Banquet, Molson Canadian and Miller High Life. MGD is indistinguishable from High Life but slightly more expensive so there's no reason to buy it. The next tier includes Pabst, Iron City, Stoney's, Labatt Blue, Molson Golden, Lion's Head, Straub. then there's the specialty tier, beers that aren't particularly cheap but have some kind of off-flavor that make them bad: Michelob, Rolling Rock, Stroh's, Genessee. Then there are beers that are straight-up bad: Carling Black Label (except when it's a buck at 123 Pleasant St. in Morgantown), Red Dog (RIP), Coors Extra Gold, Keystone, Schlitz (though that may have changed), Yuengling Premium (not the lager, which is too different to qualify for this list). Then there are beers that I'm afraid to try: Milwaukee's Best, Natural, American. Then there are the special cases: Duquesne and Hamm's have unique flavors that aren't necessarily unpleasant but that you can get sick of rather quickly. Old Milwaukee exists in some limbo where it isn't as bad as it seems like it should be.

I honestly think that if these weren't mass-market American beers but some special variety brewed by Czech monks beer snobs would rave about their crisp and refreshing taste.

Uh... no. I mean, sure, you can find people who are all snob and zero clue who maybe would, but most mass-produced American beer is objectively bad to mediocre. (If we assume it is possible to objectively judge beer on things like being accurate to style, having a distinct flavor, lack of unpleasant after-taste, etc.)

Though at least a lot of the mass-produced stuff sold in the US is not actively bad - there's no good reason to drink it if you're looking for flavor but the main issue is the lack of strong or interesting flavor, rather than an outright terrible flavor. I would actually rather drink a Coors Banquet than many "fancier" mass-produced beers like (gag) Heineken, but I'd also much rather have just about any mass-produced German or Czech (or Polish) pilsner instead. They're better, and quite often far cheaper.

By most "objective" measures typical West Coast IPAs should rank near the bottom; they're usually excessively bitter, with a cloying cough-syrupy sweetness underlying them. Yet this has become the beer snob beer par excellence, mainly because local craft breweries that don't quite know what they're doing yet can take advantage of the large margin of error something that hoppy provides and get started pretty quickly. MOst of the issues with mass-market beers come from the fact that all of them use a healthy amount of adjuncts to keep costs down. I've had craft versions of these beers from local breweries (shout out to Exile in Des Moines for offering the best example) and there is a significant improvement. The problem is that you still have to pay craft prices for these beers because they're brewed in small batches, which makes it difficult for them to occupy the same cultural position as High Life or something similar. If only a large brewery were to make a premium product without adjuncts and sell it for a few dollars more than the mainstream brands... which would create a new product that nobody wants because both the style and the concept of a mass market beer are anathema to beer snobs. I remember when something like Sam Adams was considered a premium product that most restaurants wouldn't have. Now, nobody seems to think there's anything particularly wrong with it, but no one is going out of their way to say it's a great beer. Same with Magic Hat. Same with Killian's. Even Yuengling Lager (which isn't the original Yuengling but a '90s innovation meant to latch on to the Sam Adams trend of brewing better beer) doesn't have the cachet it once had since they expanded their distribution beyond a few select markets; it used to be impossible to get in West Virginia and Ohio, even though they border Pennsylvania. Anyway, my overall point is that perception has a lot more to do with things than beer snobs want to admit. A craft beer usually has to be pretty bad before anyone wants to admit that it sucks. And no one will admit that some German Brewery that's been around since 1485 or whatever is actually sub par. The West Coast IPA model should theoretically be great for a mass market beer, because the flavor masking nature of heavy hop use would make it easy for them to get away with using a ton of adjuncts. But there's no market for a mass-market, low-cost IPA, because the people who drink IPAs will never drink a low-cost, mass-market beer.

Also, if there are German(or Czech, or Polish, or whatever) beers out there that are objectively better and cheaper than the American mass-market stuff then let me know what it is. Keep in mind that I'm used to paying about $20-$25 for a 30-rack of cans, and Pabst is usually on sale for $15.99, so it would have to compete with that to be truly worth it, though I don't mind paying more.

Anyway, my overall point is that perception has a lot more to do with things than beer snobs want to admit. A craft beer usually has to be pretty bad before anyone wants to admit that it sucks. And no one will admit that some German Brewery that's been around since 1485 or whatever is actually sub par.

I guess it depends on who you talk to, I have no problem finding craft beer people who think lots of craft beer sucks and the recent obsession with IPAs is a plague. Had a discussion about it with the owner of one the local craft beer stores I frequent just last week, we talked about how much more interesting things used to be a decade ago and he expressed dismay that people only want to drink "this shit" while waving at his 14 mostly-IPA taps. (though admittedly, it was late, it was a slow night and he was slightly tipsy)

I guess that can support your point of view or mine, depending on what we mean by "beer snobs." I don't consider the hazy juicy IPA (or hard seltzer, on days they feel that'd just be too heavy) crowd "beer snobs." Are you really a snob (regardless of how well-informed you are) if you consume something trendy and mainstream?

As for cheap beer far better than MGD, I can't find it on their website for some reason (maybe because it's alcohol) but for example, IIRC my local Costco and a Polish supermarket have 20-packs of 500ml bottles of Żywiec lager at the $25-30 price point. Though of course, YMMV.

Depressingly, the Leinenkugel's disaster is what "success" looks like. They were long ago purchased by Miller, who got to work on trying to make the brand more profitable. They repeatedly tried to go national, nothing really stuck, until they started in with the shandies and had immediately commercial success. It's really unfortunate, like you said, their old school beers are actually pretty solid representatives of their styles and aren't really all that common on the market these days. The sugar bombs sell though, so pretty much everything they make is soda with a slight beer spin to it.

I still buy Leinies original when I can get it, but yeah since the sale they've mostly been focused on seasonal varieties. The lemon Summer Shandy was a big hit in that series, and led to them making loads of other fruity beers. Can't blame them too much, light lager is a crowded market.

However it goes back some ways. I remember drinking a pitcher of Berry Weiss circa 2002 or so because it was on special, regretting it more and more with each sip. Finished the whole pitcher though.

My bourbon of choice recently has been Old Forester, the basic 86 proof usually and the 1897 when I want something more complex.

I like Ardbeg 10 for Scotch when I drink it. It has a good smokiness but a much cleaner flavor than the other heavily peated single malts I've tried.

Got a bottle of Licor 43 as basically a gag gift, but I've been genuinely enjoying it on the rocks as a nightcap.

Very sweet, but once you get past that it has a rather unique and interesting flavor.

I'm on a dry month, alas. Planning on getting some nice scotch once we hit a milestone at my startup though.

Any recs from scotch drinkers here?

I second the recommendation for Aberlour "A'bunadh," with the caveat that it's bottled at cask strength (over 60% ABV iirc) and can be pretty harsh without water or ice. Here are a few other whiskies I've enjoyed over the past year or so:

Port Charlotte "Heavily Peated" 10 - This Islay is probably my current favorite whisky, and it's comparable in price to the 10-year offerings from Ardbeg and Laphraoig. The smoke is assertive, but not overpowering; as you gradually acclimate to it, the underlying sweetness and complexity becomes more accessible. I've found notes of marzipan, cookie dough, and citrus alongside funkier flavors like seaweed and clay.

Bunnahabhain 12 - An unpeated Islay whisky. One of the most unique whiskies I've tasted, with notes of red apples, tobacco, leather, milk chocolate and musty old books. Another, very different, unpeated Islay worth trying is Bruichladdich's "The Classic Laddie," a salty, cereal, honeyed dram with "bass notes" of raisin and tennis ball rubber.

Kilchoman "Sanaig" - I remember being struck by how rich and intense the smell of this whisky was the moment I opened the bottle. A thick layer of nutty smoke over sweet cut grass notes--it makes me think of a big barn full of hay.

Arran 10 - Hits the palate with a simple, classic butterscotch flavor, but soon develops into bold fruit notes, mainly peaches and tropical fruits, before finishing dry and woody. Very impressive and complex for a 10-year-old unpeated whisky.

Highland Park Cask Strength - Like the A'bunadh, this is over 60% and is not at all "smooth." But underneath the ethanol burn there are intense and interesting floral and mineral notes with a background of tangy, fusile smokiness like a freshly-burnt-out match. The ornate bottle design may or may not appeal to you--I think it's well-executed, even though Highland Park's "Viking" branding generally makes me roll my eyes.

Other good choices include Ardbeg "Uigeadail," Ledaig 10, Laphraoig Cask Strength, Benriach "The Smoky Twelve," and blended malts like Johnny Walker Green Label and anything by Compass Box.

Also, before you spend big bucks on scotch (or any spirit), it pays to invest in appropriate glassware. Tumblers are fine if you only drink with ice, but most single malts are made to be drunk near room temperature. A tulip-shaped glass like a glencairn or snifter is ideal; fill the glass up to its widest point, to expose the highest possible surface area of the spirit to the air inside the glass. This intensifies the aroma, which is just as important as the taste for appreciating whisky. Roll the liquor around the glass before nosing so the "legs" stick to the sides of the glass, further increasing the exposed surface area. With the right glass you can savor a whisky for ten or fifteen minutes before you've even tasted it.

I like islay scotch (they're kind of like bbq sauce made with 100 year old tonic medicines). The big ones are Laphroaig and Ardbeg; their 10 years are both good entry points if you want to try.

What are you looking for in a scotch?

No idea? I like macallans.

Macallan is a sherry-casked Speyside, so you might try one in the same category, like Aberlour. If you want to branch out a little bit from there, go with something from Dahlwinnie or Glenfarclas.

You might also consider looking around for a whiskey bar. Most decent ones will do flights to let you get a good sense of what you might like.

Edit: I should clarify these aren't "high-end" or expensive scotches but they are pretty accessible if you're not a regular Scotch drinker and I think they're both great value for the money, Glenfarclas in particular.

Thanks, yeah an expensive or high-end scotch would probably be wasted on me at the moment.

Maybe, but not necessarily! I know the bourbon world much better than the scotch world, but for what it's worth, the difference between really good stuff and mid-tier stuff is often pretty obvious without having any particular background expertise. I'm more than happy to share my favorite bourbons that I have on the shelf with friends that are newer whiskey than me, with only a couple bottles being reserved for special occasions. On that note, if you have any friends that are into whiskey (or whisky), that's certainly the cheapest way to try a few things and most of them will be happy to share with someone that is interested in learning more.

Am I the only one who found Everything, Everywhere All at Once to be excruciatingly boring? I liked Daniels's first move and have enjoyed other somewhat similar films, but this one basically went nowhere after the set up (or more accurately, it went to the same place over and over).

I was completely onboard for the entire first half of the movie. It lost me in the second half, particularly when the directors seemed determined to wring every last ounce of comedy from jokes that had been hilarious and surprising in the first half, but quickly wore out their welcome through repetition. I'll be a little annoyed if it wins Best Picture at the Oscars as it clearly doesn't deserve it, albeit not as much as Banshees of Inisherin doesn't deserve it.

I thought it wasn't particularly good. I didn't really get any value out of the movie besides a couple laughs.

I thought it was okay but found the universal praise for it as a life-changing or philosophically enlightening experience a little odd.

For people who’ve never had existential epiphanies about mortality while imagining participating in pulp SF, it’s probably a life-changer. There’s a body-horror aspect that reminds me of Cronenberg films like Existenz, and a stepping into higher worlds like The Thirteenth Floor, culminating in zen rock garden world.

For me, the dramatic core was in the Boomer workaholic mom confronting her Millennial daughter’s suicidal nihilism, and finding a way through against all odds. It struck me as Important, as the teenage girl suicide rates have shot up during the Facebook/Instagram decade.

Her confrontation with her own and her father’s heteronormative assumptions is also Important, because orgasms cause some of the most passionate, real, joyful, and shameful emotions a human can experience. To ignore those feelings while judging actions is to miss the entire point.

That's fair. I had assumed that the issues presented in the film were ones that most people worked through on their own at a young age, as I did, but that clearly isn't the case and if it helps some people along then all the better.

It might have been received too many plaudits but I wouldn't say it was boring. Visually interesting and had a lot of heart (its message of unconditional love and the rebuke towards the post-ironic cynicism in our society landed nicely to me), though it leaned into the zany a little too much for my liking. I'd give it an 8/10 if you held a gun to my head. For reference I would say something like Annihilation or Oldboy (2003) are functionally perfect films for me, I love great visuals and think good acting and compelling stories are less important (not that I can't appreciate an excellent performance, There Will Be Blood is in my top 10 all time and I will sit down and re-watch it any time, if asked).

Speaking of performances, have you seen The Whale yet?

There Will Be Blood had great visuals as well as a great lead performance; the entire first 10-15 min or so, for example. I don't offhand recall the visuals of Oldboy, since I saw it 20 years ago, but it was an amazing film. I haven't seen Annihilation.

I saw The Whale, and Brendan Fraser was certainly the best thing about it. I might give the Oscar to Colin Farrell, though The Banshees of Inisherin was more of an ensemble film than was The Whale. But it also had great visuals; every interior scene, basically.

Re visuals, have you seen The Assassin

And, if you liked the visuals in Everything, Everywhere All At Once, you might like Neptune Frost.

It might be a distinction without a difference, but I categorize visuals (the use of movie magic, set/costume design, mattes/CGI/miniatures) and cinematography (camera work, lighting and framing) slightly separately though I also lump them together for most purposes since they go hand in hand, and while TWBB has excellent cinematography it only has a couple setpieces (the oil Derrick, the church) that elevate the visual aspects of the film for me. Not to say PTA isn't an accomplished director for a visual medium.

Fraser's performance was great in The Whale, but I also found myself really appreciating the set as well. I've personally known people who've failed in life, and they all lived in a place like that, with a similar sensibility towards housekeeping (though his apartment was still too neat to be true to life, I'll accept it as necessary to keep continuity from shot to shot, hard to do if everything is covered with loose wrappers and empty food or drink containers). I've found that I have a soft spot for stage plays adapted for film, The Sunset Limited is an example of my being pleasantly surprised by such a movie.

Hard agree on Banshees, people that bemoan the state of the film industry these days just aren't looking in the right places, I think. Many of my favorite films I've ever seen have been made in the past 10 years.

Haven't seen The Assassin but I'll check it out soon, and report back. I'm always excited to watch good foreign language films, I don't have near enough of a knowledge base to start working through the best of what's out there. I'll take a look at Neptune Frost as well, looks like I might possibly receive it as well as I did Beyond the Black Rainbow (an example IMO of a movie with ZERO story or high quality performances but heavy on the visuals: I shut it off after 45 minutes, I suppose I'm not a purist for visuals) but I'll give it a fair shake. Thanks for the recs!

Sure thing. If you can figure out what is going on in The Assassin, let me know. But it looks great.

I'm not sure that I'd go as far as excruciatingly boring, but I do think it was wildly overhyped. It had a novel premise (at least as far as my personal sci-fi exposure goes) but the execution wasn't as brilliant as everyone says it was. It's not bad but it's not the Second Coming of Christ either.

I did find the "sausage fingers" scenes extremely off-putting for whatever reason.

Yes, the premise was fine. But of all the Infinite skills of the different versions of herself she could accessed, she accessed the same martial arts one over and over and over. And there was not a single creative escape nor clever outwitting of her numerically superior foes.

That I chalked up to her inexperience. The main villain accessed way more variations (the luchador, the pinata, the pebble) demonstrating her superior mastery of the ability

I guess, but that doesn't make it more interesting lol.

Fair!

The movie was a lot of fun and had some great visuals. Did it earn its Best Picture nod? Perhaps. Does it deserve to win just because of its "diversity?" Absolutely not. But don't let the mental midgets on the three Ts (Twitter, TikTok, Tumblr) know that or they'll call for you to be drawn and quartered.

Fortunately, my favorite movie of the few years is Shoplifters, so I am relatively immune to that particular accusation.

Amazing film. I was thinking about it for days afterwards, and even brought it up with my therapist.