Co-Founder & CEO of @IrreducibleHW

San Francisco, CA
Joined December 2022
It's been close to a year since Irreducible (fka Ulvetanna) no longer exists. Today, I wish I could raise a glass with the team and celebrate this... Following is an insert from our last All-Hands Roadmap call (20250527): We believe in the fundamentals of binary fields in protocol design and silicon. We are the only ones pursuing them; it may appear that we are contrarians, but we are right and everybody else is wrong. Well, that’s no longer the case. Cheers 🍻
Goodbye, Poseidon! An epic 8-year, 8-figure rabbit hole in post-quantum cryptography reaches its dream conclusion. The Ethereum Foundation is abandoning Poseidon for L1, pivoting to SHA or BLAKE. This milestone unlocks ultimate security for lean Ethereum and foreshadows a golden era of hash-based cryptography. Since 2018, the Ethereum Foundation has invested in magic cryptographic bricks, so-called "SNARK-friendly hashes". In 2019, Poseidon was born. It held strong and became the dominant SNARK-friendly hash, securing billions via zkrollups and zkVMs. In a stunning reversal, breakthrough SNARK designs show that SNARK-friendly hashes aren't necessary after all. Off-the-shelf traditional hash functions like SHA2 and BLAKE2s can now match Poseidon in a SNARK. In hindsight the key was not SNARK-friendly hashes, but hash-friendly SNARKs. The secret is doing maths over the smallest prime number: 2. So-called "binary fields" natively speak the language of bits, aligning with the boolean operations inside traditional hashes. This is a stark departure from "prime fields", where awkward large-prime arithmetic makes bit manipulation painfully expensive. We're talking sci-fi cryptography. 1M traditional hash calls proven per second, on a laptop. Just 100x overhead vs native CPU boolean compute. Nobody predicted such performance, not even the handful of binary-field visionaries. Hat tip to the research geniuses: Jim and Ben with Binius in 2023; Ron, Benedikt and William with Flock in June. With SHA2, the lean aesthetic of minimal assumptions reaches its climax. The EF's principled stance on pure hash-based cryptography has aged like fine wine. We now enjoy foundations the world can trust for decades and centuries, foundations worthy of the dream of an internet of value. Speed of deployment is a secondary win. There's no longer a need to wait years for Poseidon cryptanalysis to bake. Emile and Thomas from the EF post-quantum team are moving at breakneck speed with binary fields. The strawmap now points to a production-grade leanVM in 2027, with CL, DL, EL deployments in 2028. As AI becomes exceptional at cryptanalysis, the contrarian bet to avoid riskier structures like lattices and isogenies is visibly paying off. The past weeks have been brutal. Lattice-based "HAWK" and isogeny-based "SQIsign", both signature schemes in NIST's Round 3, have suffered blows. Sources I trust say more blood is coming. On AI, the open autoresearch trend kicked off by ECDSA[.]fail is spreading fast, with amazing outcomes from zk[.]golf and SNARK[.]fast. Days ago SNARK[.]fast crossed 1.8M BLAKE3/sec proven on an M3 Max. Stay tuned for fresh autoresearch challenges dropping tomorrow. Also tomorrow: Ethproofs call #10, dedicated to binary fields. Possibly the most noteworthy Ethproofs call yet. Experts leading the charge will present the future of hash-based SNARKs at 2pm UTC. What an incredible time to be alive. To witness history, DM me for a calendar invite :) Today I can confidently claim that hash-based cryptography has won out for blockchain post-quantum signatures. SNARK succinctness compresses arbitrarily many signatures into one small proof per block. SNARK flexibility yields k-of-n threshold signatures, complex multisigs, and more. Ultimate security. Uncompromising performance. Full programmability. Believe in something. Believe in hashes.
3
4
46
4,206
good to see @standardthinkin being a leading 3B decision model in this comparison!
Our next generation of open weights multimodal decision models comes in two sizes, pplx-decider-v1.2 and pplx-decider-flash-v1.2. Text and vision scores improve. Inference is up to 5x faster: all questions in a request run in one pass. Flash is the top 4B on the Decision Index.
2
2
128
Radisav Cojbasic retweeted
Standard One 3B SH v2 now ranks #1 among sub-4B models on the Jev Decision Index by @multimodalart (Full score: 39.9). 3B v3 is already out. We're excited to see where it lands on the leaderboard.
Introducing Standard One SH: open-weight decision models. One pass. A probability per option. No text generation. Decision Index 0.3 public index: - 3B 51.5, highest under 4B vs the Oct 12 board; - 8B 57.4 Self-scored with the official kit. Open weights on @huggingface 🧵 1/6
2
2
1
3
142
I've been spending my time in the past three months with @standardthinkin team and @jordanyechan their founder and CEO. What they achieved with a small team & pre-seed resources is amazing - while hustling and iterating at the speed of light. Standard One is live on HuggingFace and got over 24k downloads in the last 2-3 weeks. Check out the Doom demo - I will put it in the comment 👇 As they say, it's not a computer if it cant run Doom. Let me know if you need an intro - we are going to be pushing more fun stuff in the coming weeks.
Introducing Standard One SH: open-weight decision models. One pass. A probability per option. No text generation. Decision Index 0.3 public index: - 3B 51.5, highest under 4B vs the Oct 12 board; - 8B 57.4 Self-scored with the official kit. Open weights on @huggingface 🧵 1/6
2
2
8
241
Replying to @standardthinkin
One 3B reads the game state as text (health, ammo, where the nearest enemy is) and picks the next move in Doom from 2–4 options. 316 decisions across four scenes, highlights from real-time runs. Recorded earlier on One 3B, the API model of the time. 4/6
21
Great to see more VC deals going into these HW startups. Semiconductor industry will grow as much as AI models (training & mostly inference) and Physical AI (robots, EVs) will. I am expecting that HW space becomes easier to enter for the startups as the demand will be unprecedented.
Today we’re excited to announce our seed fundraise and that we’ve reached 8-figures in annualized run rate. Our $5.2M seed was led by @uncorkcap, with @MoxxieVentures and angels including @JeffDean joining the round, following our pre-seed led by @pearvc. We’re a lean team combining deep hardware expertise and experience across many tapeouts with AI research and infrastructure talent. Our mission is to reinvent the chip design process and the market for custom silicon, and we’re already working with some of the largest foundation model labs. Read more about what we’re building here: phinity.ai/blog/introducing-…
1
146
this is sick ♥️
Trained. Distilled. Quantized. Compiled. Cartridged. Re-introducing my language-model-on-a-gameboy-cartridge art project with a Codex create-image driven Opus5.5 video. Custom model with ternary weights compile into 8bit instructions in an 8MiB ROM and run on-device with a custom OS maintaining 60fps UI alongside inference. Currently works in an emulator, next step is putting it onto a cartridge. Turn sound on. Enjoy. GitHub repo below.
1
154
추석 잘 보내세요 🎉
1
70
It has always been in RTL/HW design @bcherny. Since one race condition could cost you 6m of effort and a couple of hundred of MMs. Sounds familiar?
I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached. TLA+ also works well. I sometimes combine Lean and TLA+ to look for issues around data flow, concurrency, and state mgmt. I don't know either language well, but Claude is excellent at both. This approach is super useful for formally modeling your code and finding bugs that a human probably wouldn't have spotted. Is formal verification the future of coding (or at least, bug finding)?
2
226
Massive win for @BasketballAus Opals tonight in Berlin. That guy in Giddey #3 is me 😅
ABSOLUTE SCENES IN BERLIN 😳 Francesca Pasa misses the layup and Italy's 26-point comeback falls short! #FIBAWWC
93
Radisav Cojbasic retweeted
The images from Ratko Mladić’s funeral yesterday in Belgrade were shocking. Ratko Mladić was a convicted war criminal. He was responsible for genocide, war crimes, and crimes against humanity. He died while serving a life sentence for the Srebrenica genocide and other crimes. The elements of official support and the presence of some senior Serbian officials at the funeral are deeply regrettable. They demonstrate a failure to confront a dark chapter in Europe’s recent history and contradict the values underpinning Serbia’s path towards the European Union. Our thoughts are with all the victims of the crimes committed in the countries of former Yugoslavia.
5,322
2,185
861
16,939
2,669,660
Hi All, seems like my TG was hacked. I haven’t used it much lately, since Irreducible shut down. Please ignore all the messages, and since I’m no longer in crypto, feel free to kick me out from all the groups 🙏
2
4
348
PS: it was the sms 2fa that got hacked. Please consider choosing a safer option for yourself
2
131
Replying to @jimpo_potamus
@jimpo_potamus and I made the difficult but right call to wind down Irreducible.
Irreducible's time has come to an end. Long story short, @radi_cojbasic and I ultimately came to the realization that we couldn't sustainably build the kind of deep tech business in ZK that motivated us.
7
1
1
31
2,543
A big thanks to all the past team members for their endless efforts, and to our investors for their infinite support.
2
9
580
Last, but not least important, I felt so fortunate to share the blood/sweat/tears with @jimpo_potamus. Legend.
8
563
Seems like we're hacked - please ignore this while I am working with X support to regain the access to the account. Apologies to all if this post got you confused
This quoted post is unavailable.
1
1
7
2,391