SkoomaDentist
The Greater Finnish Empire
No bio...
User ID: 84
At this point it's better to ask "what standalone programming tasks can't LLMs help with", and the answer is very little.
I disagree. I have yet to see a single relevant example of LLM helping in a real world situation that would apply to me. Ie. Doing expert level tasks with large C++ codebases that have little to no documentation and also don’t rely on just calling some popular framework to do everything. Or to put it another way, the situations where help would actually be needed beyond regular googling.
Instead it’s always simple beginner level stuff (eg. your ”I don’t know much C#”) or web / databases / python.
Show me LLM writing eg. a non-trivial Linux device driver, deciphering how the hell you’re supposed to do something not quite trivial with ffmpeg programmatically or writing structural document for a random medium size codebase that’s actually accurate and not just wild ass guessing. Thus far ”AI is a must for coding” has been pure gaslighting as far as I’m concerned.
So where’s the utility of AI if I haven’t done any of the things that got utility points in your list in the last 25 years? And in general stay as far as possible from anything web related, databases, javascript or other ”typical online programmer” staples (I’ve literally said in job interviews that I want nothing to do with web stuff or databases).
Thus far the only relevant uses I’ve seen are translating and hyping up bulleted items into a format the HR demands.
AI is efficient at generating tests.
Is it?
I keep reading this but how is the AI going to know what the results are supposed to be and what the edge cases are when it doesn’t have in-depth understanding of that specific problem (ie. gazillion Stackoverflow posts to copypaste from)? And conversely, if you have to tell it those explicitly (and make a detailed review of the result to ensure it has in fact covered all cases), how is it going to save effort compared to just writing the test conditions yourself?
Nobody reads compiler stuff any more.
Clearly you've never worked in a field that cares about performance. People absolutely do read compiler output to see if it did anything too stupid and work around such issues.
superset of python
You mean the language that is de facto completely untyped in the real world and does next to no checks on the code before trying to execute it?
The best systems engineers are trans and mentally unwell.
Objection: Fabrice Bellard is quite obviously not trans and nothing suggests he's mentally unwell either. The guy makes John Carmack look like a noob.
so that they explode within 48 hours
This would make most landmines completely useless. Their value is in short to medium term area denial so you want the fields to last from weeks to some months.
The treaty was completely braindead because it effectively only restricted countries that already deployed mines using maps that allow proper clearing afterwards while the problems of landmines derive from indiscriminate (or intentionally malignant) deployment without any care about collateral damage or clearance.
It's almost as if having a political disagreement didn't mean you have to be mortal enemies. Clearly that cannot be...
I think the discussion is descending into ridiculousness at this point.
Yes, that's what happens when you discard explanations from domain experts given in good faith in favor of conspiracy theories that you've already decided to believe against all evidence.
Must be using some kind of unholy black magic
Yes, also known as "worse performance in the aspects that people actually care about". If all you cared about was apps not being able to listen to the microphone, you could always buy a dumbphone with no apps (or just not install any apps you're concerned about).
Also do you actually know that they disconnect the microphone or are you just believing their marketing material? Have you verified it yourself from the schematics? If you haven't done even that, why do you trust it more than what every expert in the topic says about audio routing and app permissions in regular phone OSes?
Facebook shouldn't have had anything to cue off of.
Except the data that the FB / Messenger app transmitted, whatever sites had FB pixel, anything you accessed (voluntarily or in the background) that communicates with FB and so on.
There is simply no mechanism for the FB app to listen to the microphone of your phone because the phone OS just won't stream audio to it unless explicitly enabled (including the visible indicator etc). To do that FB would have to abuse a zero day exploit in the phone OS for years without anyone catching on.
between all the bureaucracy, environmentalism regs, NIMBYs, and cost disease, it just costs WAY more than it probably 'should'
I think an important aspect is that those things largely don't apply to data centers in the same way they do to any public infrastructure or factories. A data center is essentially just a large building with power and network connectivity and cooling. There is no real noise or chemicals involved so safety regulations don't apply and of course it isn't a public building so the local politicians or interest groups don't get a say. It also doesn't have to be located in city center so there's no "social" aspect and in general the location is not critical so NIMBYism is much less important.
There isn't "the circuit" to interrupt. A phone has something like half a dozen digital MEMS microphones, so you'd have to interrupt a whole bunch of circuits with a single switch. That'd require an expensive and more failure prone multi-pole switch and would be hell to route. You can't just cut power to the mics either because you need to turn off the clock signal before cutting power to prevent overdriving the protection diodes and also short circuiting the clock driver (and you can't sequence the clocks because the switch directly cuts power before the cpu has a chance to react). Plus of course there would again be the routing issue.
Then there are the legal ramifications. You have to be able to call the emergency number while paniccing and phones go to quite some lengths to bypass blocks to enable that. This would of course be impossible with a physical switch.
Finally, the whole thing is completely and utterly pointless for anything other than state level actors who have to worry about zero day exploits in the OS itself (where "turn off the phone" / "leave the phone in a sealed box" is more secure and much easier to do). Anyone with half a clue of how operating systems work knows that an app can't "just access" the microphone and the OS has to do a whole lot of work to stream audio to an application via a dedicated API that is used solely for that (which is a fundamental difference from all the typical file permission bypass exploits).
Why would you trust the Phone OS to set the electronic switch off (a physical switch isn't possible) if you don't trust the same OS to not route audio to the app without your permission?
For example, did you know that neither Scotland nor the UK are actually in the EU?
It's baffling why people on this site try to use UK as example of Europe anything when it comes to legislation given the entire legal system has very different traditions.
Nonlinear acoustics is a thing but the amplitudes are orders of magnitude higher and cause more or less immediate hearing damage.
Speaker elements OTOH are well known to have by far the highest distortion of any of the common audio components (right after the ear itself but that gets deep into psychoacoustics). Once that confounding factor is eliminated, tests haven’t managed to demonstrate any instances of someone hearing beyond 20 kHz. It’s not for lack of trying either as plenty of manufacturers would love to have a peer reviewed paper to point at as a sales argument for their 96 or 192 kHz sampling rate capable equipment.
By comparison, the noise floor in the best analog sources is at -70, and that's only if you're using filters; it's more like -32 db naturally.
This is not correct. The best analog recording media have signal to noise ratio around 70 dB without noise reduction (ie. dynamic filtering). With noise reduction > 90 dB SNR is possible but you get artifacts as a tradeoff that depend on the noise reduction method.
Analog sources can quite trivially have > 100 dB SNR. A simple example would be a high quality oscillator that's switched on and off.
An unusually dynamic recording is going to have around 20db of dynamic range, a more typical recording will be in the 14db range, and most contemporary loudness war recordings will have less than 10 db.
This is a bit of a simplification in that it only considers the variation in short term loudness. You need to additionally consider the difference between the short term loudness and quietest audible element at that moment (somewhat less than 70 dB). Not coincidentally this 70 dB combined with 20 dB loudness variation is close to the CD limit of 93 dB (3 dB lower than the theoretical 96 dB due to dithering).
24 / 32 bits is of course extremely useful for recording and processing audio as you don't need to optimize the recording levels nearly as closely (and risk clipping) and avoid excessive noise accumulation over various processing passes. There are even recorders that combine multiple converters internally to record the entire dynamic range possible with room temperature electronics so that the resulting SNR is always optimal without ever clipping until the operating voltage limits of the recorder itself are exceeded.
Meanwhile there is no scientific evidence that humans can hear 192 kHz.
More specifically, there has not been a single remotely credible test that would even hint at finding anyone who could hear past 20 kHz. Every time that has been claimed, the claimed ability has completely disappeared when intermodulation distortion in the playback path was removed by using a separate amplifier and speaker element for the > 20 kHz part.
even an advice of just three bits ("move left rook") per move
That's quite a bit more than three bits. Two for direction and three or four for the piece.
A bigger problem is making that information reliably detectable by the player without alerting anyone else. Remember, the player can't run advanced signal processing algorithms to dig out hidden information from below the noisefloor and is presumably surrounded by "hostile" actors who are on the lookout for any such information. The hostile actors don't even need to be able to decode the information, just detect that there may be such information transmitted.
I'll have you know that I haven't been a self-righteous sixteen-year-old in many years
By which you (and I) of course mean that you're these days a self-righteous much older person.
Who the hell is pushing for higher standards and more rigor at universities?
Based on this thread, roughly two people on this entire site while two dozen think an attempt at doing that was grounds for dismissal. I”d liike to say I’m shocked but this isn’t the exactly the first or even the twentieth time people here have argued in simular vein.
Mankind was not meant to know some things…
For me quite significantly because aux cords were never a thing in regular car radios here. Given that radio has played pure shit for the last 20+ years, being able to conveniently play my own playlists is a rather significant feature.
actual advanced math I'd tend to just bounce off the notation and terminology, which there's so much of.
I'm pretty sure I have some sort of math "symbol blindness". If you wrote equations using regular letters and abbreviations, I'd say "Yeah, that's tricky but not too horrible" while using greek letters and math symbols would immediately result in "WTF is this shit I can't even...".
- Prev
- Next

Wait what? How is that unique? Over here the only way you'd get sedated is if they have to operate on your entire jaw or something. Even tooth surgery is just good painkillers combined with local anesthetic, so you're lucid during the operation. Basic wisdom tooth removal that doesn't require surgery is just local anesthetic.
More options
Context Copy link