xpct

Is this related to why ChatGPT sometimes links to dead URLs?

I agree. I think it also explains their behavior such as randomly wiping stuff from disk. There simply aren't any repercussions for this in their training envs.

Kids lack proper impulse control and sometimes make bad decisions if left unattended. It's preferable if we restrict certain dangerous actions from them.

Yes, LLMs can be dangerous for your own health if misused. Yes, I'd be furious if I was 14 and got my Claude access revoked.

As others have said in this thread, I'd add that this should primarily be the parents' responsibility since they know their child best.

I hold the opinion that learning about tech is important, especially as more and more of our world revolves around software.

Biased of course, but I'd keep betting on software over other fields when it comes to career options.

I agree. I open Steam sometimes and I'm met with a conveyor of games that look good enough for me to buy, but I can play a few of them at most before all of them are replaced with a new batch next month.

If true, that sounds like something that could actually be monitored by third parties, similar to the performance degradation trackers.

I don't think reading isolated stories holds much value. The value of the emotion it elicits is much lower than knowing the person who wrote it or in what historical or social background they did so.

Reading stories written by humans is one of the ways we can keep ourselves grounded to other humans: be it through fantasy, lyrics, or autobiographies. I can only describe reading AI stories as emotional indulgence.

In a way, I think training on historic chats is akin to caching computation results. The compute cost has already been paid, and we make future retrievals cheaper by encoding it directly in the model.

Assuming the results included some external validation such as user's preference, compilation, lean, etc., I'm not sure whether this would lead to model collapse.

Indeed, past a certain skill level it all becomes grunt work. Fun grunt work, but one that AI is also good at.

Probably because emulators is one of the subjects discussed on this site. Not sure how I feel about it linking to a public web artifact.

I do feel like posting links to these on HN is a type of soft distribution.

Ah, I see. Thank you for the correction.

Could someone shed some light on what's the legality of publishing emulated abandonware? I remember not too long ago Half-Life 2's web port got posted, is that still online?

You can typically see the L1/L2 SRAM cache regions just by looking for continuous patterns, then try to divide it into the number of cores or core groups you're looking for.

I don't think the Pangram result is shareable to others.

"If you are sure the link is correct, please make sure you are logged in with the account associated with this result."

> How many agents here on HN?

LLMs wouldn't pass the HN turing test. HN is also not that big, it'd be plenty enough for a malicious actor to hire real humans instead.

Honestly I'm also surprised by how blindly people trust these allegations of agent behavior. This exact example of the message board could be much easier to fabricate than to arise naturally.

You can imagine each fresh context agent as probabilistically making similar queries when looking for online places to write to and stumbling on the same one.

This becomes even more likely if it's one of the websites that got reinforced during their training process, which they may have used for reward hacking.

It's a reasonable direction, but most of online systems aren't designed for this. This would require persistent connections of any accounts you create to your identity, and disallowing anonymous actions.

A dataset being as popular as their's is will contaminate the data just by people discussing it and creating their own public test sets of similar problems.

Still, probably not that much compared to employees targeting it.

I wouldn't lump up all inventions into this. The incentives are warped and we're too focused on adtech and filling human lives with screentime. It should be entirely possible to live a fulfilling life without having to carry a phone everywhere you go.

a16z is one of the firms whose influencers I mute on sight on twitter. Far too many grifters shilling dubious products.

I don't understand the point you're making, it only comes off as misanthropic to me. We live in a world where humans are valued, and so is human culture. We educate ourselves to live richer lives and to have more fulfilling experiences. What has eventual superhuman AI to do with this?

PG's 'learning a word' ~= 'learning an idea' comes to mind

By 'rewritten' do you mean refreshed, like learning material with spaced repetition, or some different/deeper concept? Do we 'move' that information to a different cluster of neurons?

That is interesting. For a streak to count, do you have to both send out your own, and open incoming images? I wonder how much you could tell about someone if you had a 6 year snap streak with them. Would they be more like a friend, or an awkward relative with whom you seldom meet and have little in common?

Yeah, I'm aware a lot of stuff is catered towards traders. I'm someone who's more interested in the technology itself.

Ah, can't help but think of Atlas Shrugged. For all its flaws, it had a few chapters that described this idea well.

I'd pay for something that was 'SemiAnalysis, but high quality'. Perhaps that's getting too much into insider territory?

It's not intuitive to me for why preference for its own writing would emerge, and during what type of training or tuning.

Perhaps something like: learning to identify what source files it has worked on by the code style alone, because tasks may give human code (public repos, etc) and ask to make changes.