> If you’re under the impression that these models are “glorified autocomplete” or that progress is slowing down, I need to urge you: stop thinking that. The models are very intelligent and capable, they are getting better at a fast clip. I can cite measurable and impressive progress over just the past five months on specific types of problem I’ve asked them to look at. [...]
> On the other hand: if you think that models are super-intelligent or that AGI is already here, you should also stop thinking that. Working with these tools is like swimming in a pond where the ground drops off sharply. One minute you’re wading comfortably and there’s support under your feet. Then suddenly you cross a specific line, and you’re back to swimming on your own.
I'm also getting irritated with the “glorified autocomplete” comments. Since nobody can post such comments and also use the tools I'm using, I'm wondering if the phenomenon is due to people only having experience with the free version of whatever it is they're trying to use?
> both outputs of Claude Mythos, their (still) unreleased advanced model
That sentence gives the impression that Mythos might be released in the future. That's clearly not going to happen - it's already "released" in as much as selected, trusted partners can access it, and the rest of us get it in the form of Fable - which is Mythos but with filters that downgrade you if you try to use it for anything even remotely related to cybersecurity or biology.
(The other day Fable 5 downgraded me to Opus after I asked it to explain the difference between tusks and teeth.)
>They [anthropic] appear to have just told it to get some results and then strapped its nose to the grindstone until it found some.
it is fun how well this works.
i cant find the link immediately (will look and edit with it), but somewhere in the "hello there the jacobian conjecture is false thanx" thread, someone brought up a different conjecture breakthrough where the prompts were basically just repeated "no, keep going" until a result was found.
In response to your edit, you should check out Terry tao's chat gpt logs about the recent Jacobian result. The models are smart enough to brute force some things, but can cut to the meat much faster with good prompting
> i especially like "you should do a breakthrough". each prompt is less than ~20 words. makes me really question the whole "prompt engineering" stuff.
This is well past prompt engineering and into process engineering like six sigma. Just like in an early industrial revolution factory, we’re all still figuring out what works in the process of making stuff except this is so early that even simple things like “make this screw standardized” (or “no, keep going” in this case) is really high impact.
The degrees of freedom an LLM has is so large that we're going to be exploring their capabilities for decades, especially if they continue to get better. This is why IMO experts are always going to be better at LLMs in their field because they can force them LLM into processes (think prompt engineering -> CC dynamic workflows) that follow their work processes and get much better results out of them than “keep going.”
> I asked Claude for its thoughts, and it doesn’t mince words: “what makes this genuinely interesting — and, frankly, a little embarrassing for the field — is that none of the ingredients are exotic.” The TL;DR is that someone just did a much more thorough job applying all of our known tools. In short: the sort of things that attack AIs are wonderful at.
did i just read two summaries/TLDRs (in a row) of the already-two-sentence summary right above?
This is good:
> If you’re under the impression that these models are “glorified autocomplete” or that progress is slowing down, I need to urge you: stop thinking that. The models are very intelligent and capable, they are getting better at a fast clip. I can cite measurable and impressive progress over just the past five months on specific types of problem I’ve asked them to look at. [...]
> On the other hand: if you think that models are super-intelligent or that AGI is already here, you should also stop thinking that. Working with these tools is like swimming in a pond where the ground drops off sharply. One minute you’re wading comfortably and there’s support under your feet. Then suddenly you cross a specific line, and you’re back to swimming on your own.
I'm also getting irritated with the “glorified autocomplete” comments. Since nobody can post such comments and also use the tools I'm using, I'm wondering if the phenomenon is due to people only having experience with the free version of whatever it is they're trying to use?
> both outputs of Claude Mythos, their (still) unreleased advanced model
That sentence gives the impression that Mythos might be released in the future. That's clearly not going to happen - it's already "released" in as much as selected, trusted partners can access it, and the rest of us get it in the form of Fable - which is Mythos but with filters that downgrade you if you try to use it for anything even remotely related to cybersecurity or biology.
(The other day Fable 5 downgraded me to Opus after I asked it to explain the difference between tusks and teeth.)
Genuine question here, why would you ask Fable to explain the difference between tusks and teeth? That's a task that can probably be handled by Haiku.
It's the default when I pop open the Claude iPhone app, I usually don't bother to switch it.
>They [anthropic] appear to have just told it to get some results and then strapped its nose to the grindstone until it found some.
it is fun how well this works.
i cant find the link immediately (will look and edit with it), but somewhere in the "hello there the jacobian conjecture is false thanx" thread, someone brought up a different conjecture breakthrough where the prompts were basically just repeated "no, keep going" until a result was found.
edit: https://chatgpt.com/share/6a60b2eb-0b64-83ee-9c76-7931ca1de0...
i especially like "you should do a breakthrough". each prompt is less than ~20 words. makes me really question the whole "prompt engineering" stuff.
In response to your edit, you should check out Terry tao's chat gpt logs about the recent Jacobian result. The models are smart enough to brute force some things, but can cut to the meat much faster with good prompting
> i especially like "you should do a breakthrough". each prompt is less than ~20 words. makes me really question the whole "prompt engineering" stuff.
This is well past prompt engineering and into process engineering like six sigma. Just like in an early industrial revolution factory, we’re all still figuring out what works in the process of making stuff except this is so early that even simple things like “make this screw standardized” (or “no, keep going” in this case) is really high impact.
The degrees of freedom an LLM has is so large that we're going to be exploring their capabilities for decades, especially if they continue to get better. This is why IMO experts are always going to be better at LLMs in their field because they can force them LLM into processes (think prompt engineering -> CC dynamic workflows) that follow their work processes and get much better results out of them than “keep going.”
I also just finished writing a note - https://mkagenius.substack.com/p/notes-on-mythos-breaking-ae...
https://news.ycombinator.com/item?id=49099977
> I asked Claude for its thoughts, and it doesn’t mince words: “what makes this genuinely interesting — and, frankly, a little embarrassing for the field — is that none of the ingredients are exotic.” The TL;DR is that someone just did a much more thorough job applying all of our known tools. In short: the sort of things that attack AIs are wonderful at.
did i just read two summaries/TLDRs (in a row) of the already-two-sentence summary right above?