Nathaniel Craig

GPT-5.2 derives a new result in theoretical physics A new preprint shows GPT-5.2 proposing a new formula for a gluon amplitude, later formally proved and verified by OpenAI and academic collaborators.

Chapeau to Kevin, Alex, Alfredo, David, and Andy on a beautiful dialogue between theoretical physicists and LLMs. openai.com/index/new-re...

I didn't realize we were talking about existing LLM capabilities, or even just LLMs. My hope and expectation is that we can develop ML capabilities (agentic and otherwise) that accelerate transformative technologies. This takes work, distinct from promoting LLMs for QG.

Decoupling means that many different microscopic theories look the same at fixed experimental precision. It's the whole reason we can do physics! It seems obvious that progress comes from increasing the experimental precision, although that doesn't sell nearly as well as quantum gravity prophecies.

For example, there are slightly different compactifications of the same string theory that both give rise to the Standard Model particle content, but differ by the additional states at higher energies. Telling the difference requires experimental data.

It is absolutely possible there is a single-parameter theory of everything that will be revealed via a complete theory of quantum gravity. But that seems increasingly remote. More likely, a complete framework for quantum gravity can still admit different low-energy particle theories.

I have a lot of respect for Dan, and I’d entertain the quote if your essay were titled “Is Quantum Gravity Dead, Dying or Just Hard?” (arguably a better question!). But I thought we were talking about particle physics.

I think we're bottlenecked at the development of transformative accelerator technologies. If you could significantly enhance the impact of every human accelerator physicist working on ionization cooling and plasma wakefield, it would bring those horizons much closer.

The problem with Jared's inference, and other claims that AI will "solve" HEP, is that it misstates the problem: it's not that we lack a good theory for the data, it's that we have altogether too many theories compatible with the data we have. We need agentic accelerator physicists, not AI Nima.

Everything but the tunnel already fits in the CERN budget over the relevant timescale, so the useful measuring stick is private/international contributions relative to the tunnel cost. By that measure this is fairly significant.

Huge thanks to Nathaniel Craig @physicsnate.bsky.social for a beautiful and truly inspiring colloquium at UCLA today! I also really enjoyed our discussions and lunch together 👏👏👏

Oh noes...

Just to echo @dangaristo.bsky.social, there are significant factual errors in this post. LEP3 (or any proposed Higgs factory) will produce fewer Higgs bosons than the LHC. And the Zh cross section doesn't peak at mh + mZ -- that's threshold.

Generalizing the Standard Model | KITP Blackboard Lunch by Nathaniel Craig (UC Santa Barbara) The purpose of these Blackboard Talk lunches is for the science of one program to be explained to the other KITP program participants. These talks lead to cr...

Watch the latest Blackboard Lunch talk on "Generalizing the Standard Model" by Nathaniel Craig (@ucsantabarbara.bsky.social #ucsb) coordinator of the ongoing #KITP Program: What is Particle Theory? #particles25 buff.ly/EaEhSjp 🧪