exploring previously @polymarket @zksync @dragonfly_xyz @GoogleAI
San Francisco
Joined July 2012
- Tweets519
- Following984
- Followers9.3K
- Likes4.9K
Ivan Bogatyy retweeted
Big congratulations to the @zksync airbender team for this milestone.
At this point, I consider the battle for speed to be definitively won, and the top priority for zkEVMs should shift fully to security.
To me that means formally verifying
(i) equivalence of the arithmetization to the ethereum spec
(ii) a hard maximum bound on (prover cost) / (amount of gas in block)
(iii) security of the proof system itself
firefly.social/post/x/211001…
Lean formalization of Hironaka's resolution of singularities, the 1964 theorem that won him the Fields Medal, by @resolution_org
Ivan Bogatyy retweeted
Following NEAR AI topping putnambench leaderboard, we now have claimed #1 on Lean Eval v1, the official active benchmark maintained by the Lean developers.
We now have a very capable and cheap formal verification agent, and will share how we use it to formally prove NEAR protocol and smart contracts soon.
NEAR AI has claimed first place on the Lean Eval v1 leaderboard.
Lean Eval v1 is an active benchmark maintained by Lean FRO that contains extremely hard formalization problems. We reached first place despite joining the competition just four weeks ago, a full month later than other top participants.
NEAR is very heavily investing in formal verification efforts. Our ultimate goal is to formally prove the protocol and all the core contracts, and to provide a practical tool for others to formally verify their contracts in a fast and affordable way.
The system that enabled us to claim first place is almost entirely powered by DeepSeek V4.1 Flash, and thus is (a) extremely cheap, and (b) has no dependencies on third parties because the model is open weight.
When it comes to formally verifying software, the approach boils down to three pillars:
1. Representing the program in a way that can be reasoned about;
2. Expressing the properties that we desire to prove;
3. Coming up with a proof and formalizing it in Lean.
Until recently, step (3) was prohibitively expensive for practical use at scale. With the system that we used for Lean Eval, we now have it solved. Stay tuned for updates on how we apply it in practice to formally verify NEAR contracts.
Ivan Bogatyy retweeted
NEAR AI has claimed first place on the Lean Eval v1 leaderboard.
Lean Eval v1 is an active benchmark maintained by Lean FRO that contains extremely hard formalization problems. We reached first place despite joining the competition just four weeks ago, a full month later than other top participants.
NEAR is very heavily investing in formal verification efforts. Our ultimate goal is to formally prove the protocol and all the core contracts, and to provide a practical tool for others to formally verify their contracts in a fast and affordable way.
The system that enabled us to claim first place is almost entirely powered by DeepSeek V4.1 Flash, and thus is (a) extremely cheap, and (b) has no dependencies on third parties because the model is open weight.
When it comes to formally verifying software, the approach boils down to three pillars:
1. Representing the program in a way that can be reasoned about;
2. Expressing the properties that we desire to prove;
3. Coming up with a proof and formalizing it in Lean.
Until recently, step (3) was prohibitively expensive for practical use at scale. With the system that we used for Lean Eval, we now have it solved. Stay tuned for updates on how we apply it in practice to formally verify NEAR contracts.
Ivan Bogatyy retweeted
exciting times
hello from the inaugural cohort of @OpenAI safety fellows!
we're excited to be spending the next few months thinking carefully about and making progress on core issues in ai alignment, control, & interpretability, with the support of great mentors + the community at @ConstellOrg
ALT the first cohort of the openai safety fellows, sitting in two rows on a tree trunk and smiling
Ivan Bogatyy retweeted
NEAR AI Lean agent solved ALL problems in Putnam Bench at 1/250 the cost of the second cheapest submission.
$111 for 672 problems from the hardest undergraduate math competition in the world.
Fully open source 💚
Cool approach: get the best of two LLMs by having them communicate in latent space
meet @mostik_ai!
what happens when you put 12 PhDs in one room for four months? first place on the ARC-AGI leaderboard, which I can't say much about while the competition is still running. and this, which I can.
everyone's arguing about whether open models will catch up to frontier models. we think it's the wrong question. here's the one we pose: why does a frontier model have to generate your answer at all, when the only thing you need from it is the reasoning?
we do this by enabling models to communicate in latent space. through our protocol, hidden states pass straight from a frontier model into a small one running on your infrastructure -- no text between them, and neither model is fine-tuned. two models from different families, sharing reasoning, both left untouched.
how do we know it works? we tested it on a setup where a 753B model reads the problem, and a 4B edge-class model writes the answer. with this approach, we get results 80% as accurate as the frontier model, but at 20x faster performance.
we're committed to preventing frontier model lock-in and are already partnering with inference providers to accelerate open-weight adoption. we've done this between 15 of us, in four months, 12 PhDs and a Fields medalist, backed by @generalcatalyst
WIRED has the first external account of the company and the work: wired.com/story/russian-star…
full writeup, the setup, and all the numbers: mostik.ai/read-more
Related: teapaw.org/hodlonaut/status/20838…
The commits (code changes) that introduced Coldcard's low entropy bug, changed around 2500 lines of code.
These were changes to the most important parts of their codebase, ensuring that the seeds have enough entropy so that.... your coins don't get stolen.
The code comments for all these changes?
"runs"
and
"x"
Hard to believe, but true.
Ivan Bogatyy retweeted
Welcome to “Intelligence After Hours”
@AhmadAbbas_
@maggielove_
@afrazhaowang
@jonwu_
@Zac_Aztec
@marek_
@elkun_peng
@AlexSkidanov
@yaroslavvb
@hosseeb
@AlexanderLong
@ivanbogatyy
@nikitabase
@YevgenChebotar
Hosted by @ilblackdragon @blackdragon_vc @near_ai @build_planet ✨
Ivan Bogatyy retweeted
From our team at Impossible Research - an agent harness that plays games, writes code, reasons like a physicist, and saturates the ARC-AGI-3 benchmark.
Asked Fable to help debug my transformer, and it found a subtle bug in PyTorch instead.
Recursive self-improvement just casually showing up like that.
The bug: on Apple Silicon, linear layers give wrong results for large batches. Silently. No crash, nothing.
(technically more of an Apple MPS issue than PyTorch itself)
Good luck finding this with a meat brain
github.com/pytorch/pytorch/i…
![[MPS] F.linear with bias silently returns wrong results when 3D batch dim exceeds 2^16 (65536) ·...](https://pbs.twimg.com/card_img/2116027768291082241/vGEQMDsw?format=jpg&name=800x419)
I tested my LLM audit pipeline on Aztec 2.0: could it rediscover the vulnerability that drained the vault?
It did, and then found a SECOND, unreported critical: prove a deposit, tag it "not a deposit", L1 skips the debit, withdrawal works.
Harmless now, vault's already empty.
We are investigating a potential exploit affecting a deprecated Aztec payments product from 2021. ~$2m was transferred from the immutable smart contract in transaction:
etherscan.io/tx/0xab306cd218…
The deprecated product is an immutable stage 2 rollup that was sunset in 2022.
Aztec Labs holds no admin keys or control over the system; it cannot be paused or upgraded by us.
This is a separate incident from the exploit on the depreciated Aztec Connect product that occurred on June 14, 2026.
We will share further updates in due course.
Wow, my vanilla Claude Code rediscovered the $2M Aztec vulnerability in 45 minutes.
That's probably how the attacker discovered it too.
The economics of cyber defense have totally changed and the ecosystem needs to catch up.
We are investigating a potential exploit affecting Aztec Connect. ~$2.1m was transferred from the immutable smart contract in transaction:
etherscan.io/tx/0x074ec9317d…
Aztec Connect was deprecated 3 years ago. Aztec Labs holds no admin keys or control over the system; it cannot be paused or upgraded by us.
We will share further updates in due course.
The vulnerability was introduced in January 2022 and sat undiscovered by the team and the auditors for 4.5 years: github.com/AztecProtocol/azt…
My Claude Code with Opus 4.8 rediscovered it with blocked Web search, no hints, no custom skills or MCP: gist.github.com/bogatyy/a2ac…
My vanilla Claude Code was able to rediscover the Zcash vulnerability!
No custom harness necessary.
Similar to Erdős unit distance problem solved by vanilla GPT-5.5 after OpenAI used a custom math harness.
Seems that raw model intelligence matters a lot more than the harness.
