←
Ben Mayer (t11s)15,385 posts
Ben Mayer (t11s) avatar

Ben Mayer (t11s)

@transmissions11

cs @mit, prev @sailresearchco @paradigm

◷ Joined September 2020▧ 1,629 Photos and videos
Giacomo Fenzi avatar
⟳ Ben Mayer (t11s) retweeted
Giacomo Fenzi@GiacomoFenzi·18h

We updated ZO0K 🦓, more performance, a simpler construction, and general module and rings! The protocol is similar but with a few twists, such as using a single error correcting code instead of a complex, a different "universal masking polynomial" and more. Paper: ia.cr/2026/391 Github: github.com/WizardOfMenlo/zo0… cc @gwberg95

Post attachmentPost attachment
Nicholas Decker avatar
⟳ Ben Mayer (t11s) retweeted
Nicholas Decker@captgouda24·Nov 16

Donald Iwuchukwu has stolen millions of dollars from the elderly. If the authorities do not act, he is going to get away with it. My hope is that publicity will lead to the arrest of him and his co-conspirators. Link below, and please RT:

Post attachment
will depue avatar
⟳ Ben Mayer (t11s) retweeted
will depue@willdepue·Nov 12

I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.

Post attachment
Miranda Christ avatar
⟳ Ben Mayer (t11s) retweeted
Miranda Christ@miranda__christ·Nov 4

Happy to be a part of this project. zk proofs of inference are a key tool for AI safety, and I'm excited to see the protocol improved & used in applications (the code is public!)

New Anthropic research: progress on provable inference for large language models. Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so. These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers. For our full write-up, see here: anthropic.com/research/prova… Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP… We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).

Roy Rinberg avatar
⟳ Ben Mayer (t11s) retweeted
Roy Rinberg@RoyRinberg·Nov 3

Sometimes, all you have to do is ask. We are open-sourcing a research prototype which we hope leads to more progress in this field. FlashProve is open source and proves a 400B model at over 100 tok/s on one H100 in prefill.

2. Direct agents to build zkLLM but 100x faster or with floats, and open-source it. Efficient, OS zero-knowledge proofs would be a widely useful tool for verification.

Roy Rinberg avatar
⟳ Ben Mayer (t11s) retweeted
Roy Rinberg@RoyRinberg·Nov 3

I am very proud of this work. FlashProve is a research prototype, but it shows that proving which model produced a response can be practical. In the blog, the first usecase we discuss is AI pacing: such proofs get us part of the way to checking that hardware serves inference, and is not being used for training. This is just a starting point, and I'm very excited for others to take on the mantle and outdo us. anthropic.com/research/prova…

New Anthropic research: progress on provable inference for large language models. Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so. These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers. For our full write-up, see here: anthropic.com/research/prova… Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP… We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).

Guy Weissenberg avatar
⟳ Ben Mayer (t11s) retweeted
Guy Weissenberg@gwberg95·Nov 3

An amazing experience working on this with such a great team! Very excited about the role of interactive proofs and cryptography in AI safety 🔐

New Anthropic research: progress on provable inference for large language models. Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so. These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers. For our full write-up, see here: anthropic.com/research/prova… Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP… We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).

Ben Mayer (t11s) avatar
Ben Mayer (t11s)@transmissions11·Nov 3

Really excited about the safety implications of efficient provable inference, and there's still lots of speedups on the table! Learned a ton working with @RoyRinberg, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul (rdalal.github.io), @alexallain, and Alessandro. x.com/GiacomoFenzi/status/21…

I am happy to share FlashProve: our performant open-source prototype for provable inference. We believe that this has significant implications for security, pacing AI development, integrity guarantees, and more.  FlashProve allows a provider to commit to model weights, and then prove that replies originate from the committed model.  The proof leaks no information about the weights, and the verifier can efficiently verify it on consumer hardware.  Our library is performant, proving a 400B parameter MoE model at 100 tok/s. We have open-sourced the associated code and Lean formalization. We hope that it catalyzes further progress in verification of inference, and look forward to seeing it.

0:22
Ben Mayer (t11s) avatar
Ben Mayer (t11s)@transmissions11·Nov 3

Had the pleasure of helping with this! 🫡 I suspect there's historically been a big overhang in applied ZKP work due to the sheer implementation & verification complexity. But now thanks to LLMs, implementation is far easier and autoformalization is practical. Exciting times :)

New Anthropic research: progress on provable inference for large language models. Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so. These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers. For our full write-up, see here: anthropic.com/research/prova… Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP… We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).

Anthropic avatar
⟳ Ben Mayer (t11s) retweeted
Anthropic@AnthropicAI·Nov 3

New Anthropic research: progress on provable inference for large language models. Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so. These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers. For our full write-up, see here: anthropic.com/research/prova… Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP… We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).

Giacomo Fenzi avatar
⟳ Ben Mayer (t11s) retweeted
Giacomo Fenzi@GiacomoFenzi·Nov 3

I am happy to share FlashProve: our performant open-source prototype for provable inference. We believe that this has significant implications for security, pacing AI development, integrity guarantees, and more.  FlashProve allows a provider to commit to model weights, and then prove that replies originate from the committed model.  The proof leaks no information about the weights, and the verifier can efficiently verify it on consumer hardware.  Our library is performant, proving a 400B parameter MoE model at 100 tok/s. We have open-sourced the associated code and Lean formalization. We hope that it catalyzes further progress in verification of inference, and look forward to seeing it.

0:22

New Anthropic research: progress on provable inference for large language models. Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so. These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers. For our full write-up, see here: anthropic.com/research/prova… Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP… We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).

Georgios Konstantopoulos avatar
⟳ Ben Mayer (t11s) retweeted
Georgios Konstantopoulos@gakonst·Nov 3

a lot of reasons to be excited for the new anthropic work but the one im the most excited to shout out is @transmissions11 in the coauthors! we had the privilege of working together at @paradigm a few yrs ago & he inspired me to start foundry gg! anthropic.com/research/prova…

Georgios Konstantopoulos avatar
⟳ Ben Mayer (t11s) retweeted
Georgios Konstantopoulos@gakonst·Nov 2

SOLAR: 100x FASTER SOLIDITY Solar is the future of Solidity and the EVM. It's >100x faster than Solc. It has faster bytecode than Solady with no assembly. Its bytecode is faster to interpret by the EVM. Solar fixes Stack Too Deep! foundryup forge --use solar Blog below.

0:45