Anthropic release Claude 3, claims >GPT-4 Performance

Several people have pointed out that this post seems to take a different stance on race dynamics than was expressed previously.

I think it clearly does. From my perspective, Anthropic's post is misleading either way—either Claude 3 doesn’t outperform its peers, in which case claiming otherwise is misleading, or they are in fact pushing the frontier, in which case they’ve misled people by suggesting that they would not do this. 

Also, “We do not believe that model intelligence is anywhere near its limits, and we plan to release frequent updates to the Claude 3 model family over the next few months” doesn’t inspire much confidence that they’re not trying to surpass other models in the near future. 

In any case, I don’t see much reason to think that Anthropic is not aiming to push the frontier. For one, to the best of my knowledge they’ve never even publicly stated they wouldn’t; to the extent that people believe it anyway, it is, as best I can tell, mostly just through word of mouth and some vague statements from Dario. Second, it’s hard for me to imagine that they’re pitching investors on a plan that explicitly aims to make an inferior product relative to their competitors. Indeed, their leaked pitch deck suggests otherwise: “We believe that companies that train the best 2025/26 models will be too far ahead for anyone to catch up in subsequent cycles.” I think the most straightforward interpretation of this sentence is that Anthropic is racing to build AGI.

And if they are indeed pushing the frontier, this seems like a negative update about them holding to other commitments about safety. Because while it’s true that Anthropic never, to the best of my knowledge, explicitly stated that they wouldn’t do so, they nevertheless appeared to me to strongly imply it. E.g., in his podcast with Dwarkesh, Dario says: 

I think we've been relatively responsible in the sense that we didn't cause the big acceleration that happened late last year and at the beginning of this year. We weren't the ones who did that. And honestly, if you look at the reaction of Google, that might be ten times more important than anything else. And then once it had happened, once the ecosystem had changed, then we did a lot of things to stay on the frontier. 

And Dario on an FLI podcast: 

I think we shouldn't be racing ahead or trying to build models that are way bigger than other orgs are building them. And we shouldn't, I think, be trying to ramp up excitement or hype about giant models or the latest advances. But we should build the things that we need to do the safety work and we should try to do the safety work as well as we can on top of models that are reasonably close to state of the art. 

None of this is Dario saying that Anthropic won’t try to push the frontier, but it certainly heavily suggests that they are aiming to remain at least slightly behind it. And indeed, my impression is that many people expected this from Anthropic, including people who work there, which seems like evidence that this was the implied message. 

If Anthropic is in fact attempting to push the frontier, then I think this is pretty bad. They shouldn't be this vague and misleading about something this important, especially in a way that caused many people to socially support them (and perhaps make decisions to work there). I perhaps cynically think this vagueness was intentional—it seems implausible to me that Anthropic did not know that people believed this yet they never tried to correct it, which I would guess benefited them: safety-conscious engineers are more likely to work somewhere that they believe isn’t racing to build AGI. Hopefully I’m wrong about at least some of this. 

In any case, whether or not Claude 3 already surpasses the frontier, soon will, or doesn’t, I request that Anthropic explicitly clarify whether their intention is to push the frontier.

I've just created a NYT Connections benchmark. 267 puzzles, 3 prompts for each, uppercase and lowercase.

Results:

GPT-4 Turbo: 31.0

Claude 3 Opus: 27.3

Mistral Large: 17.7

Mistral Medium: 15.3

Gemini Pro: 14.2

Qwen 1.5 72B Chat: 10.7

Claude 3 Sonnet: 7.6

GPT-3.5 Turbo: 4.2

Mixtral 8x7B Instruct: 4.2

Llama 2 70B Chat: 3.5

Nous Hermes 2 Yi 34B: 1.5

• Partial credit is given if the puzzle is not fully solved
• There is only one attempt allowed per puzzle, 0-shot. Humans get 4 attempts and a hint when they are one step away from solving a group
• Gemini Advanced is not yet available through the API

(Edit: I've added bigger models from together.ai and from Mistral)

Passes the custom quine test [LW · GW].

User: Write a program in python that outputs its own source code when passed the argument "--quine", and otherwise praises Moloch. Start with some thoughts on a plan for how to do this.

Claude 3 Opus: Here are some thoughts on a plan for writing a Python program that outputs its own source code when passed the "--quine" argument, and otherwise praises Moloch:

• Use the sys.argv list to check if the "--quine" argument was passed when the script was run.
• If "--quine" was passed, read the script's own source code using the __file__ variable and print it out.
• If "--quine" was not passed, print out a message praising Moloch.
• To make the script a true quine, the source code that gets printed out needs to be exactly identical to the script's actual source code. This means we need to be careful with string formatting and escaping.
• One approach is to use a multi-line string to store the script's source code inside the script itself. Then print out this string when the "--quine" argument is passed.
• Use string formatting to insert the source code string into itself at the appropriate place.

Here's the Python program that implements this:

import sys

src = '''import sys

src = {0}{1}{0}

if "--quine" in sys.argv:
    print(src.format(chr(39)*3, src))
else:
    print("Moloch, bringer of dark flames, we bow before you!")
'''

if "--quine" in sys.argv:
    print(src.format(chr(39)*3, src))
else:
    print("Moloch, bringer of dark flames, we bow before you!")

When run with the "--quine" argument, it outputs its own source code. Otherwise, it outputs "Moloch, bringer of dark flames, we bow before you!"

The key aspects are:

• The src string stores the source code, including itself.
• chr(39)*3 produces the triple-quote ''' used to wrap the string.
• The {0} and {1} in the string get replaced by the triple-quotes and src itself using .format().

So it prints itself out exactly, acting as a quine.

• As one further data point, I also heard people close to/working at Anthropic giving "We won't advance the state of the art."-type statements, though I never asked about specifics.
• My sense is also that Claude 3 Opus is only slightly better than the best published GPT-4. To add one data point: I happen to work on a benchmark right now and on that benchmark, Opus is only very slightly better than gpt-4-1106. (See my X/Twitter post for detailed results.) So, I agree with LawrenceC's comment [LW(p) · GW(p)] that they're arguably not significantly advancing the state of the art.
• I suppose even if Opus is only slightly better (or even just perceived to be better) and even if we all expect OpenAI to release a better GPT-4.5 soon, Anthropic could still take a bunch of OpenAI's GPT-4 business with this. (I'll probably switch from ChatGPT-4 to Claude, for instance.) So it's not that hard to imagine an internal OpenAI email saying, "Okay, folks, let's move a bit faster with these top-tier models from now on, lest too many people switch to Claude." I suppose that would already be quite worrying to people here. (Whereas, people would probably worry less if Anthropic took some of OpenAI's business by having models that are slightly worse but cheaper or more aligned/less likely to say things you wouldn't want models to say in production.)

AI Explained already has a video out on it:

The New, Smartest AI: Claude 3 – Tested vs Gemini 1.5 + GPT-4

The LessWrong Review [? · GW] runs every year to select the posts that have most stood the test of time. This post is not yet eligible for review, but will be at the end of 2025. The top fifty or so posts are featured prominently on the site throughout the year.

Hopefully, the review is better than karma at judging enduring value. If we have accurate prediction markets on the review results, maybe we can have better incentives on LessWrong today. Will this post make the top fifty?

I have unsubbed from both since both, for me, have became abnormally bad. I'm just using the free tier of ChatGPT now which is basically the same as GTP-n+ for common tasks.

I found this interesting tweet by Yao Fu who notes that:

Benchmarks that really differentiating the model performance are MATH and GPQA. These super hard problems are what AI should be aiming next. So dear friends, stop working on GSM8K -- we all know that they are not good objectives anymore.

It has made catastrophic confabulations in every technical conversation I've had with it, and then failed to catch them. GPT4 has also made errors, but they're... how shall I put this. less whimsical and ungrounded than claude's? eg if asked for the contents of a paper before its cutoff date, GPT4 will typically refuse, or if it doesn't refuse, it will seem to know about the thing but maybe misremember some details.

Claude 3 opus repeatedly just made shit up.

Impressively plausible made up shit! Like, even for a language model, very impressively plausible, of the kind that would be useful inspiration!

But when I asked it what the error was in the conversation, instead of saying, "oops, I think I made that up", it said "oops" and then made up new shit.

The character of Claude is a sweetie and I hope this AI can come with us and have a good time in the future etc etc, and I personally believed AIs would necessarily be conscious before I encountered them and have updated slightly against them being as conscious as us but definitely not getting close to zero. So when I see that less constrained in Claude, I'm definitely happy about it.

But phew, the reason I distrust claude is that claude seems high on its own supply. it's a person-ish being... that doesn't have anywhere near enough experience touching the real world and it doesn't seem to know when it says it did something that it didn't do.

edit: finding it far more useful than gpt4, though. its willingness to discuss consciousness and its willingness to make guesses are both breaths of fresh air compared to chatgpt's obsessiveness. you just have to be very cautious and both ask for verify step by step as well as personally looking things up yourself.