@odd_primes's banner p

odd_primes


				

				

				
0 followers   follows 0 users  
joined 2025 June 19 02:29:15 UTC

				

User ID: 3777

odd_primes


				
				
				

				
0 followers   follows 0 users   joined 2025 June 19 02:29:15 UTC

					

No bio...


					

User ID: 3777

So your claim is that the Taliban regime ca. 2004 was the obvious Schelling point for Afghans interested in the long-term thriving of their country, and those who did not support the Taliban were clearly defecting from the common good of the country?

No. I just don't think they are demonstrating qualities that would make them uniquely valuable citizens, worthy of being fast-tracked through a special process. We have plenty of carrots and sticks for dealing with collaborators: money, status, security... And if we want our local collaborators to be effective, they should be invested in the success of our effort for the long haul. If their plan is to be on an evacuation flight out, why not staff the army with soldiers who only exist on paper, and rob the treasury blind?

There is a line in the sand for collaborating with a foreign invader to depose your government and occupy your country, which comes with unavoidable mass murder and atrocities. For Afghanistan, the death toll is estimated at around 200,000, along with the displacement of millions. The government actively trying to genocide you certainly crosses that line. The government enforcing a strict interpretation of Islamic law? That's a lot more questionable. Even a shitty government is often better than a foreign occupation and a low-intensity insurgency. Collaborating means obtaining a personal benefit from and enabling a process that imposes mass suffering on one's countrymen. I think the willingness to do so should be considered an anti-signal when it comes to citizenship, and certainly not an automatic qualification.

The incentives seem misaligned when we reward our collaborators for failure. We spent decades and tens of billions of dollars training the ANA only for them to surrender practically without a fight. Giving them the prospect of an escape route to the US likely weakened their resolve rather than strengthening it. It also doesn't seem like the sort of behavior you'd expect from people who genuinely believe they will be executed or harshly persecuted by the new government.

I can see making an exception for rare cases that demonstrate remarkable courage or character as a PR strategy, but extending it to just about any collaborator is completely misguided.

The program was not limited to translators, and Scott acknowledges that with "eg as translators". Anyone who worked for the American or coalition forces for at least 12 months can get a special immigrant visa.

Even steelmanning your argument though, why would we grant citizenship to someone willing to sell out their country to an invading power for a paycheck? When they swear their oath of citizenship to the United States, promising to bear arms on behalf of the US when required by law, and support and defend the Constitution, why would we believe them?

The ones I know seem pretty normal. But they spend the vast majority of their time on site with clients, so that likely insulates them from the corporate culture.

Often creditors won't even bother to show up for the bankruptcy hearing and it just gets rubber-stamped. Eg. if someone has $50k in debt split across five different credit cards, it's just not worth a lawyer's time to try to squeeze a few thousand dollars out of the person of which the creditor would only get a fifth.

Depending on the state there are also a bunch of exemptions. For example Alex Jones got to keep his multi-million dollar house under the Texas exemption for a primary residence, despite having an outstanding judgement of over a billion dollars.

I have been getting 5-10 second page loads and intermittent 504s as well. It isn't completely broken, but it's definitely annoying and I could see it impacting site usage.

It's a baffling workflow since it's completely useless for phone users. It requires a second device to work.

One workaround though is to just take a screenshot of the QR code and drop it into a tool like cyberchef. QR codes are pretty straightforward to parse.

The real-life version of James Bond would be a bunch of British bureaucrats sitting in meetings for months, punting a decision down the road until a disaster strikes and they are forced to announce that the perpetrator was on their radar the whole time. Meanwhile James Bond becomes a depressed alcoholic doing a desk job and waiting for retirement. High-agency people working in a gay and retarded bureaucracy ruins suspension of disbelief. They either find a new job, or stop being high-agency.

Isn't the point of the telepresence operators to help generate training data for complex tasks? Like it's the same idea as autonomous vehicles: first have the human operator in control and record data with sensors, then progressively shift tasks over to an automated system with a human overseeing the task, with the human only intervening where necessary. This works even better with telepresence since the human can just remote into a robot that gets stuck, fix it, and move on to another robot with a different edge case.

This model sounds plausible to me, but I'm not involved in the robotics space, so I'd be curious to know what you think.

In terms of the safety aspect, I'd be much more concerned about the customers than the operators. The human operator can always just take off their headset if they are being harassed, while the customer cannot. And there are pretty obvious issues with having a roving camera controlled by a human operator in someone's home, like the scandal where gig workers for Irobot posted pictures online of customers on the toilet. It turns out that they were shipping images back to a contractor for data labelling. That seems like the much more obvious failure mode for the Neo.

One of the key figures behind O9A, 764, Atomwaffen, and other offshoots is a FBI informant who runs an extremist literature publishing company. You have to wonder how many people he radicalized over the years, and what percentage of them the FBI managed to apprehend before they committed a serious crime.

The key evangelical for O9A, the figure who facilitated this macabre wedding of apocalyptic death cults, is Joshua Caleb Sutter, a 41-year-old ex-convict, prolific satanist, publisher of manuscripts advocating murder, torture, rape, and child abuse — and a paid FBI informant since 2004.

https://www.rollingstone.com/culture/culture-features/the-satanist-neo-nazi-plot-to-murder-u-s-soldiers-1352629/

Sutter’s influence on 764 is readily apparent in the facts surrounding some of the group’s most violent participants, particularly the possession of O9A texts published by Sutter’s Martinet Press imprint, tattoos and flags of the Tempel ov Blood’s insignia, and his consistent promotion of it on social media and in newer publications. Alleged members of the exploitation network include Angel Almeida, who is currently facing a maximum penalty of life on federal charges of coercing a minor to commit sexual acts and possession of CSAM and a firearm, and a Romanian national convicted of possessing and distributing CSAM and had Tempel ov Blood indicia or tattoos of the group's trident emblem.

https://www.wired.com/story/the-dangerous-exploits-of-an-extremist-fbi-informant/

I mean it's definitely not high art, but it's pretty cool that a single amateur animator can crank out five minutes of animated content about a niche internet microcelebrity in a couple days. For comparison South Park apparently costs something like $500k per episode, and The Simpsons is in the millions.

As someone stuck in-office generally 4+ days per week due to operational requirements, I generally try to have a lower intensity "fun" project to work on during slower periods. The "useful" project gets prioritized of course, but I find there's generally something in our queue of work to do that I can use as an excuse to experiment with tech that I'm interested in learning more about. This might be somewhat specific to software engineering though.

It's pretty normal in this field not to do 8+ hours per day of nose to the grindstone coding though, realistically it's tough to have more than 4-6 hours of high-quality focused attention. Especially with the constant distractions of Slack, email, etc.

Having helped coordinate events in the past, the magic words are typically "attending a conference" and reminding people to never mention the word "work". It's a bit of an odd situation though since from the description, it sounds like attendees technically are producing some sort of work product that they may be getting compensated for (eg. If they are posting the blog posts to Substack) and yet it isn't an accredited writing class that would qualify for a J1. It's in a bit of a gray area where you'd probably be fine as long as you framed it properly, but I could easily see the visa officer getting confused since it doesn't fit nicely into one of the usual boxes.

The Will Stancil Show animated partly using Sora is a pretty neat use of AI video that goes above and beyond the standard Instagram slop. What other transformative uses of this technology are you guys seeing? It's a bit disappointing that so far AI video seems to be mostly used for short form video clickbait. Maybe more open weight models will help over time, the high-end offerings right now have a high price tag, lots of content filters, and onerous ToS.

The DA's office still has to review cases and figure out which ones are worth charging, go through the evidence for potentially exclupatory evidence that they are required to disclose, coordinate with law enforcement on building the case in a legally compliant way, etc. They can start putting in hours on a case long before the defense attorney even gets retained.

The plea bargain issue goes both ways I think. If a public defender has a case that they know is a lost cause, they will do their best to convince their client that taking a plea is in their best interest. Every case that the prosecution closes with a plea is also a case that the PD doesn't have to take to trial.

Experts are a fair point though, if that's coming out of the PD budget that is a bit unfair.

They point out that their budget is 60% of that of the District Attorney's Office, which to me signals their complaint is likely justified.

This doesn't necessarily seem like a huge red flag depending on what percentage of defendants retain private counsel, and the inherently asymmetric system where the prosecution bears the burden of proving the case beyond a reasonable doubt. It makes sense that it costs more in terms of man-hours to build cases than to poke holes in them.

Yeah LLMs are notoriously weak at anagrams due to how tokenizing works. Here's a fun little demo of how LLAMA's tokenizer breaks up an input sequence: https://belladoreai.github.io/llama-tokenizer-js/example-demo/build/

Hence the famous example of LLMs failing at "how many times does the letter R appear in the word strawberry" until the training data for newer models was contaminated with the answer.

$100k per person per year is totally reasonable. It wipes out the body shop business model, but a highly talented software engineer making $500k in TC at a FAANG will be just fine. This raises the bar for H1-B to the point where it will be used for what it was actually intended for, filling short-term gaps in highly specialized fields where demand vastly exceeds supply.

Historically the good cop, bad cop approach seems to have had success in some contexts. South Africa had Nelson Mandela preaching peace and tolerance, while his wife cheered on the practice of necklacing alleged informants, and her security detail carried out kidnappings, torture, and murders. There was a less extreme dynamic in the civil rights movement, with MLK positioning himself as the reasonable alternative to violent radicals like the Black Panthers. People mostly want peace and stability, so the idea of compromising with moderates can be appealing given the alternative. Of course that depends on the moderates having a palatable message, and support from elites and the media.

Of course this doesn't always work out and sometimes results in violent suppression of the entire movement, including both the moderate and radical elements. The Tamil Tigers in Sri Lanka are likely a good example, with the situation evolving from a political campaign, into an insurgency, and finally a full-blown civil war.

They also applied their style guide to a quote, which seems completely inappropriate. Dinesh D'Souza did not capitalize either word. NYT version:

“The reason for the media silence is racial,” Dinesh D’Souza, the right-wing commentator, wrote in an online post on Sunday. “If the killer were white, this would get coverage. Of course if it were a white assailant murdering a Black victim, then it would be front-page headlines everywhere.”

I don’t think that the killing has anything to do with race, at all

The assailant is on tape saying "I got that white girl, I got that white girl" as he walks off the train after the attack. Maybe he was mentally ill, maybe he was motivated by racial animus, or maybe it was a bit of both? Perhaps interviews with the assailant will reveal more, but right now Trump and the FBI are really trying to downplay the racial angle and make it about cash-free bail and repeat violent offenders. They got to a federal death penalty charge via transit terrorism, which seems like a bit of a stretch.

From a broader perspective, this incident has highlighted some awkward realities of inter-ethnic crime rates that have largely been banished from mainstream discourse. Twitter is no longer a representative sample of mainstream opinion... but it's the now the incubation chamber for right-wing messaging, and the Overton window there has significantly shifted in the last few days. The video is really powerful, I would put it above the George Floyd footage in terms of emotional impact. Or course it won't have the same amplification, given how national media coverage appears to be grudging at best. The NYT article on the topic is absolutely ridiculous - they try to equate honest reporting on a serious crime to fabricated reports of crimes leading to a riot in the 1800s:

In North Carolina, as in other Southern states, newspapers in the Jim Crow era often egregiously exaggerated stories about Black criminality. Among other things, such stories served as a precursor to a white supremacist uprising in Wilmington, N.C., in 1898, in which at least 60 Black men were killed.

Honestly the easiest way to get started is to go on CivitAI, find something similar to what you want, and then start playing with the knobs and dials from there. Many images have attached metadata including ComfyUI workflows. This gives you at least a known-good configuration to fall back on when something goes wrong.

The /g/ board on 4chan also has a long running Local Diffusion General which has a bunch of guides and resources for getting started.

One thing to note though is that ComfyUI is a bit of a security nightmare. Custom nodes are basically just python scripts downloaded and run from the internet with little to no security screening by the maintainers. If you're experimenting with this on a computer with sensitive information, I'd recommend not installing random custom nodes with two stars on GitHub.

It is a fair criticism, for whatever reason the census bureau decided to include social science and psychology in STEM. They do have a very nice visualization though, by clicking on the major it shows the percentage of employees who end up working in STEM jobs, and highlights their placement in different job groups. For computer science and math, it's 51.1%. Engineering is 51.5%. For physical science majors, it's only 27.6%. Those are pretty grim numbers that aren't explained just by management being excluded from the stats.

Let's say the US has X amount of specialized talent and thus they can only do Y amount of productivity with in a year. If companies in (or investing in) our country are so productive and there's enough market demand that they want to do creation over Y, then limiting access to talent over X puts a cap on growth.

This is a false premise on a number of levels. Productivity is the value generated per worker. Production is the total amount of goods and services created. Production is a function of a number of different inputs - labor, capital, land, etc. If a company needs to increase production in an environment with a constrained labor market, it can do that by increasing productivity. In other words, investing capital into technology, automation, and other labor-saving improvements. Production can also be increased by offshoring low-value components of the supply chain to other countries, like mineral extraction, textile manufacturing, etc. Of course as we have seen this needs to be done judiciously to avoid building our own competition in unfriendly countries.

Now I know, the general response is "because those jobs should go to the locals!" but the thing is, talented local people already have jobs. If they're hard working and capable, then they're mostly already doing their part in achieving Y (or doing something else in another industry) because companies want them.

If your new venture is creating more value than whatever these people are already doing, do the capitalist thing and poach them. That's the free market at work - businesses with higher margins can afford to attract labor from companies or industries that are generating less value. That's part of how productivity increases, generating more value per worker by having workers move to higher value positions.

As any hiring manager knows nowadays, the job pool is mostly incompetents, liars, lazies, addicts, or otherwise unwanted because of a serious flaw. It's the same way that dating apps like Tinder are mostly used by the unpleasant and unwanted, the good ones are already picked through.

The vast majority of employees are "on the market". Just offer them more money. It's that simple. People hop jobs all the time, especially in hot industries like tech. Even if they aren't officially looking, it's easy to put the word out that you're hiring and have your staff refer people into your hiring pipeline. What's more common is that companies have unrealistic expectations - they want 90th percentile employees at 50th percentile pay and mediocre benefits.

Now maybe that's what we as a society want, jobs programs for the lazy drug addicted idiots being put in roles above their worth, and we're willing to sacrifice efficiency in key industries for it.

This is a bullshit false dichotomy. How about just incentivizing the 72% of STEM grads who don't work in a STEM job to actually work in STEM, if we have such a skills shortage?