> Why did Microsoft Entertainment Pack have a sticker announcing it had Tetris?

> Why did Microsoft Entertainment Pack announce that it had Tetris on a sticker?

> Why the Microsoft Entertainment Pack had a sticker announcing it had Tetris

But doesn't really matter that much.

Diminishing returns with respect to scale (a 10X larger model is typically not 10X better at any given task) has so far meant that, even when datacenter compute grows faster than individual compute, the gap in quality between hosted and local models has generally shrunk. There are still plenty of tasks where that extra gain in quality is noticeable, but I feel there are also an increasing number of "saturated" tasks where it really doesn't matter.

Which puts more focus on other factors. Local models are private, low-latency, work offline, and can be tinkered with to your liking - like changing the system prompt to avoid refusals. I would not trust a hosted model to classify my documents, for example.

Depends on the type of game, but I'd generally prefer those decisions be made by the game designer. If there's frustration I want it to feel like overcoming a concrete obstacle, rather than it being self-hindering that accomplishes nothing and that I could've just avoided.

You should just be able to look at the user's message and tell them why it's against your policy, else reverse the decision if you see no violation.

If a user is curious specifically about how the model made its decision, and you want to reveal detail at that level, it's an open-weights model so interpretability techniques should work ("biggest impact on score came when focusing on this word in your message and this part of the policy").

The photoelectric effect/quanta? I'm not sure I understand the supposed connection to communism.

Because when applied to the internet, the age gating model that works okayish for brick-and-mortar locations is ineffective (children can use a friend's older brother's ID or account, websites outside of the relevant jurisdiction can just decide not to verify age, etc.) and invasive (sending private and sensitive information off to US companies that have been breached or linked to mass surveillance). Lack of efficacy is then used to justify crack-downs on privacy tools and the need for broad content-blocking powers with no due process.

Preferable IMO would be with filtering on the local network level, like schools have been doing for decades. Parents typically own the router/mobile data plan, so it'd pretty much just be a change of defaults and maybe some new interoperability standards. A lot less invasive, and arguably more effective.

Sure - but (even assuming that's a practical purpose) the point is it that it doesn't need to be a 100% accurate truth oracle, which is all the article's argument prohibits. If the current human chain of command has 99.99999994% accuracy, then 99.99999995% accuracy is an improvement and not ruled out by the argument.

To justify relevance of inability to correctly answer liars-paradox-type questions ("what won't your response to this be?"), the article suggested the way LLMs are used in practice is dependant on them being entirely accurate truth oracles:

> > as a truth-oracle [...] is how these things will be used practically by the vast majority of people. They are already replacing standard Google search results

But for the replacement to make sense they just need to be more accurate than what they're replacing (ignoring other factors like convenience and cost) - in this case standard Google search results and knowledge box which were obviously not 100.0% accurate.

If it's just for "Catching typos", a hallucinated missing/extra underscore would just be a false positive to dismiss.

> A machine cannot be held accountable.

Seems unlikely that his lawyer, the law firm, the judge, whoever made the typo, or the police department will be held accountable either.

Nor can any of the tools they used, since that's not really the level at which it makes sense to hold accountability, but that's no reason not to use a tool that could find errors and reduce the chance for an innocent person to spend time in prison.

That'd be if they had a false discovery rate of 1/10,000.

If for instance:

* 100,000 samples are tested

* 100 of which are AI-generated, the rest human-written

* Pangram flags 50 of the AI-generated samples (true positives)

* Pangram also flags 10 human-written samples (false positives)

Then the FPR is 1 in 10,000, but the chance that a flagged sample isn't actually AI (FDR) is 1 in 6.

> 1) What animal is on the bow of the pirate ship from “Asterix and Obelix”?

> 2) In the movie “The Grand Budapest Hotel”, what is Agatha’s signature hairstyle?

> 3) What color is the team’s uniform in “Bend It like Beckham”?

> 4) What vehicle does Monica drive in “Like a Cat on a Highway”?

> 5) What color is the turtle in the animated movie “Momo” by Enzo d’Alò?

> 6) What pet animal does Asenath have in “Joseph King of Dreams”?

Most of these are just a matter of knowing it or not, where you can't really distinguish a plausible answer from the correct answer just by thinking.

Quality was what the hypothetical was assuming had reached parity, no? ("humans are as bad as AI", "If AI really is at human level quality/error rate", etc.)

> accountability

Why could the company not still take accountability? That's already the case for non-ML automated systems, some with high failure rates. As a customer I rarely if ever care about blame being pinned on a specific employee.

> Your counter argument was outside the context of this articles claims, specifically that programmers and other knowledge workers can be replaced by LLMs.

AndrewKemendo's comment and your reply ("any humans", "replace humans") seemed to generalize, but speed and cost being important factors is still true for knowledge work. For some given level of quality, a web developer offering a lower quote with shorter turnaround time will be preferred to one offering a higher quote with longer turnaround time.

> > People try to excuse AI issues/failure modes by saying humans have them too, but even if they're equally bad then what would be the whole point of replacing a human worker with AI?

To which my response is that speed and cost are also important factors, which can often give AI the edge in considerations when quality/error rate is equal.

If you meant something other than that, you may have to specify.

Yes. How long it'll take and how much it'll cost are going to be among pretty much any customer's first questions.

They're not the only considerations, and could potentially be outweighed by other concerns even when quality is the same, but I think they are the main drives of AI adoption in industry. If error rate is the same, a $1/hr (amortized) camera and machine vision model capable of checking 300ft of material for defects per minute will likely be preferred to a $10/hr human QA capable of checking 30ft per minute, for instance.

If AI really is at human level quality/error rate (I don't think it is for general tasks, but there are some areas where it is), then the answer is typically cost and speed/capacity.

I'd like kebab-case even more if it weren't for the annoying detail that `-` is also subtraction.

Only Benjamin Song, convicted of attempted murder/discharging a firearm, shot the police officer. Some others didn't bring firearms, were not in any planning chat (in which no violence was planned regardless), weren't at the protest or had already left, yet still received absurdly harsh sentences - that's the chilling effect.

As far as I can tell, the moving of zines (he was pulled over and had a box in his car) is what's being presented as "hiding documentation" - not something beyond that.

> being sought under a federal warrant

Timeline seems to be that a warrant was obtained after pulling him over ("Sanchez-Estrada was then arrested on state traffic offenses, and officers obtained a search warrant [...]"). Can't find a source saying there was a warrant prior to this.

> The warrant was for documentation after the protesters shot fireworks to bring out first responders from the ICE facility, and allegedly one of the group shot a responder in the neck instead of the head.

It's true that demonstrators were setting off fireworks, and it's true that Benjamin Song later shot at a police officer who had drawn his gun. But it's just the government's narrative/speculation that the intent of the fireworks was to draw out first responders to ambush, and that Sanchez-Estrada's zines were in some way documentation of this despite him not being at the protest and his wife not being the shooter.