Axiomatic_AI (@Axiomatic_AI) header
Axiomatic_AI (@Axiomatic_AI) profile picture

Axiomatic_AI: Think Free.

Boston - Barcelona - Toronto
Joined February 2024
Your Lean proof compiles. But what did it prove? OctoInform turns Lean statements back into readable math, with every sentence linked to the definitions it rests on. We launched it last weekend at the @Caltech x @SAIRfoundation Mathathon, where teams used it to explore OpenAI’s recent math release. Thanks to everyone who joined our workshop! Explore it yourself: octoinform.axiomatic.ai #AIforScience #FormalVerification #Lean4
1
38
Our mathematician Benjamin Breen will be joining as a judge and speaker at the @Caltech x @SAIRfoundation Mathathon, and leading a Lean formalization workshop this Sat, Nov 14 at 1 PM PT. We're also launching and showcasing OctoInform (octoinform.axiomatic.ai/), which turns Lean proofs into readable math. See OctoInform on the Lean proof behind @OpenAI's Seymour result: octoinform.axiomatic.ai/open…
1
2
4
331
Can a graphene or 2D-material photodetector be both very sensitive and very fast? We had Lemma read 14 papers and plot every device by sensitivity vs speed. The best ones all hit roughly the same limit. Stop organizing papers. Start exploring them. Try Lemma: platform.axiomatic-ai.com/r/…
1
36
v2 adds over 100 codes whose twofold transversal generators yield the full logical Clifford group, including a beautiful [[77,34,5]] generalization of the Golay code. Plus, @Axiomatic_AI has independently Lean verified the hardest theorem in the paper --- the decomposition of arbitrary twofold gates into Z-diagonal, X-diagonal, and CNOT networks.
I know we're all excited and scared, but there is room not only for the fancy open problems. Collecting all gate recipes I could find, one can define the "two-fold transversal" group, the group of all stacks of depth-one two-local code-preserving Clifford circuits. Each element turns out to be a product of generating layers that we understand, which allows for numerical search. This in turn shows that several CSS codes admit full logical Clifford group using such stacks. These stacks can be interspersed with QEC to be fault tolerant. Along those lines, the group of all code-preserving Clifford circuits turns out to be generated only by Z- and X-diagonal circuits, which should be useful for circuit compilation. Looking ahead, we can also use permutations to make two-fold transversal circuits code-preserving, which could be interesting for atom platforms. How agents helped: - explaining old-school binary symplectic rep theory and geometry - numerous algorithms to realize the gate search, cited all the way back to Metropolis - expressing any group element as a product of generators, which involves an exhaustive recipe for all 720 elements of the two-qubit symplectic group - sane-looking toc, none of that revtex two-column stuff I used the Agentic Publication Protocol [arxiv:2606.27386], which packages the paper into a repo. The repo can be loaded to an agent, which will then pretend to be me and answer your questions. github.com/valbert4/two-fold…
1
2
31
2,019
A lens and a grating coupler, solved together, one slider between them. Lemma ran Optiland for the lens and Tidy3D for the chip, then joined them at one plane above the grating. Move the height slider in the notebook and both beams shift and spread together. Try Lemma: axiomatic.ai/lemma/at-a-glan…
1
5
204
A grating coupler spec in, a Tidy3D model out, from one prompt. Lemma reads the stack, the pitch and the wavelength, builds the 3D geometry, and sets up the FDTD run in the sandbox. The field plots come back in the notebook. No local install. Try Lemma: axiomatic.ai/lemma/at-a-glan…
2
50
Flow past a cylinder, from one prompt to a vortex street. Lemma writes the solver, runs it in the sandbox, and hands back the plot and the notebook. Change the Reynolds number and run it again. The code is yours to read. Try Lemma: axiomatic.ai/lemma/at-a-glan…
30
Scientific work takes more than chat. @quantum_jake said it yesterday; here is what it looks like. Bring a paper. Pick the compute and the model. Lemma writes the plan, you approve each task, it runs in its own sandbox, and you get back a notebook you can move. You inspect the work. It did the setup. Try Lemma: axiomatic.ai/lemma/at-a-glan…
32
Our CEO @quantum_jake took the Tough Tech stage today to show what we're building. He showcased one of our products, Lemma, the sandbox for scientists. Hand it a paper. It re-derives the equations, checks each reference against the record, and returns an annotated PDF with every flag next to its evidence. Hand it a simulation. It writes the code, runs it on the compute you choose, up to an H200 GPU, and brings back the notebook. Hand it a plan. Each step gets a check before the next one runs. Try Lemma today platform.axiomatic-ai.com/r/… #ToughTechWeek
1
2
1
191
Check your paper. Pick your compute. Run the plan. New Lemma is live: the sandbox for scientists. platform.axiomatic-ai.com/r/…
34
Here’s what happened when we tested Lemma on dozens of papers with known errors. Full results → axiomatic.ai/blog/lemma-pape…
1
92
Every paper deserves a second look. Generic LLMs miss errors. Confidently. Lemma doesn’t. Built by scientists, for scientists. It reads your paper and flags what’s worth that second look: The science, the claims, the physics and math, as well as the references. Check your paper with Lemma today platform.axiomatic-ai.com/r/…
42
We recently used Lemma to check 28,000 references across 459 science papers. More than 1 in 3 papers had at least one error. We've published a blog post about how Lemma detected these errors. Give it a read here: lnkd.in/eqVuSjhb Try Lemma here: platform.axiomatic-ai.com/r/…
1
1
19,806
Excited to be selected for the @GoogleStartups Gemini Startup Forum! We're building AI for science and engineering. One glimpse of what's being explored with Lemma: turning a research paper into a digital twin you can interact with. 📷 See you in November, @GoogleCloud, @GoogleDeepMind & @GeminiApp! What will you build with Lemma? - platform.axiomatic-ai.com/r/…
250
Our CEO @quantum_jake is a physicist, a fellow of the American Physical Society and Optica, and helped stand up the U.S. Center for AI Standards and Innovation at NIST. Today in the Washington Examiner, he asks who should check the work at the frontier of AI: washingtonexaminer.com/op-ed… Scientists live by that question. It's how we built Lemma, our AI co-explorer for scientists. It reads the literature, builds and runs the model, and shows its work so you can check every step. Lemma is in public beta. Start building here: platform.axiomatic-ai.com/r/…
1
2
129