mckirk

What does 'being decent' mean in this context, in your opinion?

True, but afaik none that an AI company itself uses to say 'this is what we think showcases the individual strengths of our models the best'.

Since internally you probably have a fairly good intuition for what these models do well, what about assembling a benchmark suite (with multiple dimensions) that produces results that show that intuition, but in a 'somewhat objective' way?

That could solve the 'more IQ' problem, because you'd actually have different dimensions (e.g. for 'raw intelligence', technical writing, emotional understanding, etc) that you could use to communicate the differences in models better.

(Of course people would instantly start complaining that the benchmarks numbers are probably fake etc, but it is what it is.)

I stopped reading after 'Clearly not'.

It's okay to have a strong opinion on this, but if you suggest the answer is 'clearly anything', I will start to suspect you haven't thought about the problem enough.

Only the funny ones

No, we're pacing ourselves to have the time to evaluate the impact each new model could have, obviously.

That's one legit interpretation, but 'the people working on this are deeply unsettled and truly want to pace because they're afraid of where we're headed' should remain another.

Ohh I'm gonna steal that one.

Although I don't fully agree. With AI the focus now shifts a lot from 'being able to implement the thing' to 'being able to know/figure out exactly what you want the thing to look like', also against the siren song of the LLMs constantly having ideas of what you could _also_ build.

So, I agree that it's not a direct translation from 'pro pizza cook' is 'pro pizza orderer', but I'm quite sure the pro pizza cooks would generally have the better pizzas in the end, if there were infinite variations of pizza on the menu and the ordering agent was constantly trying to upsell you on slathering on another layer of sauce and cheese.

Which is not to say there aren't exceptions in both directions, because as a pro pizza connoisseur you could even be better positioned to 'know exactly what you want', without ever having made a pizza; and as a pro cook you might still be overwhelmed by the ordering process.

Though all of this still assumes of course that we are in the 'complexity region' where we can expect the LLM/the pizza agent to deliver _exactly_ what we wanted, without issue. As soon as we leave that region (though it expands every week, it feels like), the skills of the cook are still very much necessary to make sure things are as they should be, and fix them if not.

"Well a wave hit it"

Why do I get the feeling this 'pacing' will be one of "yeah alright guys, let's pace ourselves while I'm ahead".

LLMs don't need to be the dangerous ASI thing themselves for them to still be dangerous. It suffices if they

- destabilize the world just by what they do: making previously very hard things (generating fake everything, manipulating people, hacking, bio-weapons research) easy enough fast enough that our civilization doesn't cope with it and/or

- accelerate AI research to the point where they make the dangerous ASI thing appear years earlier than would otherwise have happened

Thank you, that is very reassuring to know :D

(and interactions like these are part of the reason why I <3 HN)

Maybe you're thinking of Phrack [1]? I wonder if there'll be another edition of it, and if yes what it would look like, now that the AIpocalypse seems to have reached security research...

[1]: https://phrack.org

(insert meme of multiple Spider-Men pointing at each other here)

> for some reason

I mean, I can think of one reason. Them being quite far away from the ground most of the time, mainly.

"THERE IS ABSOLUTELY NO CAUSE FOR ALARM"

Sorry for being off-topic, but 'rouge agents' was too adorable to leave unmentioned.

_Every_ agent? I surely hope not. But it would be an interesting... artisocial experiment?

Just don't be confused if they then follow the commands side-effect free.

The fact that Orban is called Viktor as his first name meant your first sentence confused me greatly for a second.

I mean, it's understandable; having to endure a lockdown _with_ Doordash was really rough on our civilization.

Damnit, I am pretty sure I had a few-year-streak going until just now. Welp, off to the grind again, I suppose.

Nitpicky reply questioning the adherence of OP's comment to HN guidelines.

Along these lines: I really like the 'Climate Reanalyzer' project by the Climate Change Institute at the University of Maine [1]. There's so much good stuff there if you click around a bit; you can create custom plots for the surface temperature of different regions for example[2], which quickly shows you that Western Europe has actually warmed a lot more than the global average, and we're closer to +2°C already in that region.

[1]: https://climatereanalyzer.org/clim/sst_daily/?dm_id=world2 [2]: https://climatereanalyzer.org/research_tools/monthly_tseries...

If writing goes the way music seems to be going with Angine de Poitrine gaining a huge following as a kind of allergic reaction of people against the 'AI sameness'... then we could be in for a wild ride.

On the other hand, music is primarily an art form and writing (nowadays) is primarily utilitarian I would contend, so maybe the analogy doesn't quite hold up.

I have to admit, the movements and interactions looked so uncannily natural that my intuitive suspicion was that this might be another instance of a 'fake it till you make it' demo, and the robot was actually controlled by a person. Though if this is real, that's of course some of the highest praise they could hope for.

I have to admit that this is also what keeps me coming back to LinkedIn. My brain is dangerously easy to motivate by dangling a virtual leaderboard in front of it.

I think what that research found is that _auto-generated_ agent instructions made results slightly worse, but human-written ones made them slightly better, presumably because anything the model could auto-generate, it could also find out in-context.

But especially for conventions that would be difficult to pick up on in-context, these instruction files absolutely make sense. (Though it might be worth it to split them into multiple sub-files the model only reads when it needs that specific workflow.)

This is the first time I've heard the term 'apecode', and I will make sure to use it at every opportunity.

This is only done at the DNS level, so using a different DNS (such as Quad9) solves that issue. For background info, I can recommend [1, 2].

[1]: https://www.youtube.com/watch?v=Uxmu25mUZgg [2]: https://cuiiliste.de/