3/5/2026 at 1:04:31 AM
I'm tired of these posts; LLMs are good for happy-path demos, that's it. And even then, their success rate depends on the prompter already knowing the answer!Literally any out-of-distribution project in which I used LLMs lead to catastrophic failure. The models can't "see" stuff outside their training data.
by behnamoh
3/5/2026 at 1:14:30 AM
I legitimately can’t tell if you’re being serious. It kind of seems like you might be trying to parody LLM detractors that will never admit to their usefulness. If you’re serious, why choose to say so in this post, which includes hard evidence that you’re wrong?by semiquaver
3/5/2026 at 1:19:18 AM
> which includes hard evidence that you’re wrong?You should already know what to ask to extract the answer OpenAI claims gpt-5.2-pro gave them.
Then you should be lucky to get an answer that makes sense.
Then you should already know how to verify the model's response.
Only after all these steps should you cherry-pick the one-in-a-million successful response to feature on your website.
And finally, you should prove that the answer didn't already exist in the training data. It's highly likely that the problem was solved before and the model picked that up. I have yet to see a genuinely novel discovery these models can produce.
* I'm an LLM researcher, but that doesn't mean I should close my eyes to the unjustified hype around language models.
by behnamoh
3/5/2026 at 1:31:23 AM
According to the post, this result was first derived for gluons in a previous paper. That paper was provided to the model as context, and then the model was asked to derive an analogous result for gravitons, which presumably has not been done before. The authors claim it would have taken "considerable time" for human experts to derive the graviton result.I don't see any reason to believe that this exact problem was solved before in the training data, but it's definitely an incremental result based on a very similar problem that the model had seen before.
by MajimasEyepatch
3/5/2026 at 6:33:29 PM
Gluons have spin 1 and Gravitons [1] have spin 2. This adds a lot of subtle difference in the calculations. Some are obvious like replacing a 1 with a 2 (or a 1^2 with a 2^2). Some are super tricky:For example if you have two gluons, you apply the rules of sum of spin and get
1 × 1 = 2 + 1 + 0
But for gravitons, the rules are
2 × 2 = 4 + 3 + 2 + 1 + 0
If you want to make give physicist nightmares and make mathematician cry, a tiny part of the details are in https://en.wikipedia.org/wiki/Table_of_Clebsch%E2%80%93Gorda... and https://en.wikipedia.org/wiki/Table_of_Clebsch%E2%80%93Gorda...
In conclusion, I'm not sure how difficult is to do the conversion from gluons to gravitons, but I'd recommend to run away.
[1] Assuming they exist.
by gus_massa
3/5/2026 at 1:52:34 AM
>It's highly likely that the problem was solved before and the model picked that up.If you can demonstrate that, I would put it to Strominger and his colleagues, and I imagine they would be obligated to cite your contribution in the peer-reviewed publication.
by JProthero
3/5/2026 at 1:54:46 AM
> If you can demonstrate that, I would put it to Strominger and his colleagues, and I imagine they would be obligated to cite your contribution in the peer-reviewed publication.There's one little problem: OpenAI isn't actually open and doesn't reveal which dataset they used for training.
by behnamoh
3/5/2026 at 2:09:41 AM
This shouldn't prevent anyone from finding and reporting a similar pre-existing result in the literature.by JProthero
3/5/2026 at 2:25:20 AM
Or at least, please explain the results found, and how they are already known. Any papers of your own in the field?by ChrisClark
3/5/2026 at 2:24:34 AM
But you yourself said they found the answer already online, so go find it, prove itby ChrisClark
3/5/2026 at 4:04:38 AM
I never said such a thing. Read my comment again.by behnamoh
3/5/2026 at 6:31:17 AM
How I wish more people would remember this! What extraordinary interpolation tools we have! Truly marvelous and worth celebrating, but not worth a damn for extrapolation. Such a torrent of well-formed advertising trying to convince experts that a universal and reliable extrapolator has been demonstrated!by muhdeeb
3/5/2026 at 3:40:45 PM
You don't need to know the answer, just be able to recognize it. There are many situations where you can judge the output a lot easier than you can produce it, and LLMs are quite useful for those.by wat10000
3/5/2026 at 1:43:12 AM
Do you understand the purported result, and the verification? I don't, but I'm confident that Andrew Strominger wouldn't have agreed to put his name on this if he didn't think it was correct and interesting.The human authors have positions at the Institute for Advanced Study (Einstein's old institution), Vanderbilt, Harvard (Strominger) and Cambridge in the UK.
If you have to gauge this by the reputation of the experts involved in it as I do, that seems like a good list to me.
by JProthero