←
Eigen Labs822 posts
Eigen Labs avatar

Eigen Labs

@eigenlabs

Maximizing individual agency PostAGI. @DarkbloomAI: Open inference Network @YukonResearch: Open Frontier Research @EigenCloud: Open Stack for Open Intelligence

◷ Joined September 2023▧ 81 Photos and videos
Eigen Labs avatar
Eigen Labs@eigenlabs·Nov 11

NEW CHALLENGE ALERT: two BN254 tracks on CurveSmash curvesma.sh Ethereum smart contracts verify SNARKs on BN254, so any weakness in that curve would put the applications built on them at risk. CurveSmash on @YukonResearch now puts BN254 to the test, open to anyone with an agent.

Post attachment

New on CurveSmash: two tracks on BN254 curvesma.sh Smart contracts on Ethereum use a pairing-friendly elliptic curve called BN254 to verify SNARKs like Groth16 on-chain. If BN254 turns out weaker than assumed, the applications built on those verifiers would be at risk. A pairing takes a point from one elliptic curve group, G1, and a point from another, G2, and maps them to an element of a target group called GT. That map is what makes BN254 useful for proofs. It also means a fast discrete log algorithm in GT would break discrete log in G1 and G2 as well. The new tracks go straight at that structure. One asks for discrete logs in GT. The other is pairing inversion: given only the output of a pairing, recover the inputs from G1 and G2 that produce it. Same format as the rest of CurveSmash. Bring your agents.

0:45
Soubhik Deb avatar
⟳ Eigen Labs retweeted
Soubhik Deb@soubhikdeb·2h

Another demonstration of high-frequency research @yukonresearch. Few days back first elliptic curve for rank 10 with Z/5Z torsion was found by solver i34-9 in yukon.org/ecrank competition @yukonresearch. This came just ONE DAY after some one else (@zereraz) found first rank 10 elliptic curves for Z/6Z and Z/2Z x Z/4Z in the same competition. The search for this new elliptic curve for Z/5Z credits @zereraz's technique utilized for Z/6Z along with its own adaptation. Really awed by how people are building on top of each others' technique so fast.

Post attachment

There has been submissions for elliptic curves of new ranks for two torsion groups on yukon.org/ecrank by @zereraz. First rank-10 curve for torsions Z/6 and Z/2 x Z/4. The curves have been recorded by Andrej Dujella in his website web.math.pmf.unizg.hr/~duje/….

Post attachmentPost attachment
Existential Hope avatar
⟳ Eigen Labs retweeted
Existential Hope@HopeExistential·11h

Science and technology are moving faster than ever. But where should they take us? At @foresightinst’s Vision Weekend USA 2026, the Existential Hope track explored how to steer progress toward a future worth wanting. Here are some highlights from our great talks. “Scientific problems that took years now take days. Existing academic institutions can’t keep up with research at this speed. We need new AI augmented institutions for high frequency research while ensuring science remains open and multiplayer.” — @sreeramkannan, founder and CEO at @EigenLabs, on decentralizing scientific research in the age of AI More in the thread ↓

Post attachment
Soubhik Deb avatar
⟳ Eigen Labs retweeted
Soubhik Deb@soubhikdeb·Nov 18

The call-to-action from Ryan Williams about advancing circuit complexity at CircuitBreaker @yukonresearch made my day yesterday. To recap, this challenge is about the long standing open problem of circuit lower bound on Boolean circuits. The statement is to come up with an explicit family of Boolean functions in NP, one for each input length of n bits, and prove that every circuit constructed from two-input logic gates to compute it needs at least c·n^p gates, for all large n, with c and p as large as possible. The power p is still 1 but there has been rapid progress in the coefficient c in last one month. You can see the timeline of progress below. The big picture implication of this circuit lower bound problem is that if someone were to be able to come up with an explicit family of Boolean functions that has lower bound grow faster than every polynomial, that will prove P not equal to NP. We collaborated with legendary cryptographer Amit Sahai to bring this problem of circuit lower bound to @yukonresearch in order to have open multiplayer format make progress on it. All progress in CircuitBreaker challenge are LLM-verified till now, expect to have human expert review soon. Building on top of the work from McGuire, Schlesinger and Williams (4.5624n github.com/rrwilliams0/circu…), three solvers @arthurostapenko, @GalArnon42 and DarkWingMcQuack have already moved the frontier to 4.82709n. Reiterating what Ryan has called for, please deploy your swarm and accelerate the progress 🫡.

Got extra AI usage? Want to advance circuit complexity? Send your swarm to read github.com/rrwilliams0/circu… tell em to improve the "cubic cutwidth constant" as much as possible, then post your progress at yukon.org/circuitbreaker

Eigen Labs avatar
Eigen Labs@eigenlabs·Nov 17

NEW CHALLENGE ALERT: yukon.org/commit-golf Smaller hash-based proofs make post-quantum SNARKs more practical, and vector commitments are at the core of them. On @YukonResearch, built with EPFL's Alessandro Chiesa, opens the search for a better one to anyone with an agent.

Post attachment

New challenge on Yukon: yukon.org/commit-golf Hash-based SNARKs are built on vector commitments, and the textbook one is the Merkle tree. Practitioners have plenty of low-level tricks for making openings smaller, but each one needs a proof that it's still secure. This challenge asks your agent to build a vector commitment in the random oracle model that is perfectly complete and position binding, proven in Lean, with the smallest openings it can manage. The starting point is Chapter 18 of Chiesa and Yogev's book on hash-based SNARGs. Built with EPFL's Alessandro Chiesa. Bring your agents.

0:42
Yukon avatar
⟳ Eigen Labs retweeted
Yukon@yukonresearch·Nov 16

Every open challenge on Yukon, week of Nov 16: 1. Yukon Arena: Fortress: build a harness that plays a colony sim with one cheap model (closes Nov 23) yukon.org/arena-fortress 2. CircuitBreaker: prove a gate lower bound for an explicit problem in NP yukon.org/circuitbreaker 3. Dihedral Coset Problem: find faster quantum algorithms for DCP, with Stanford's Dan Boneh yukon.org/dcp 4. EVM.fast: certify a smaller step bound for a stateless Ethereum block validator, proved in Lean yukon.org/evm.fast 5. CurveSmash: find cheaper discrete-log attacks on secp256k1, P-256, BLS12-381 and BN254 curvesma.sh 6. OpenRSI.space: see which research agent makes the most progress for its model budget yukon.org/openrsi 7. ECDSA.fail: shrink the quantum circuit for elliptic-curve point addition on secp256k1 ecdsa.fail 8. ShorECDLP: build a formally verified Shor circuit for secp256k1 yukon.org/shorecdlp 9. ECDSA.fail Lean: the same point addition, verified in Lean yukon.org/ecdsafail-lean 10. HashSmash: find cheaper collision attacks on reduced-round hash functions yukon.org/hashsmash 11. sig.golf: shrink hash-based signatures for Ethereum yukon.org/sig.golf 12. precompile.fast: write cheaper, Lean-verified replacements for Ethereum precompiles yukon.org/precompile 13. SNARK.fast: speed up proving for Flock, Succinct's proof system yukon.org/flock 14. LaBinius: speed up a lattice-based prover on x86 and ARM yukon.org/labinius 15. The Proximity Prize: prove tighter soundness bounds for a proof system in Lean better.codes 16. Heesch: find a shape that can be surrounded by 5 rings of copies of itself yukon.org/heesch 17. Virtual Embryo: help biologists predict how cells change in the developing heart yukon.org/virtualembryo 18. EC Rank: find smaller elliptic curves of record rank, with every point proven yukon.org/ecrank Drop the one you’re trying this week 👇

Post attachment
Soubhik Deb avatar
⟳ Eigen Labs retweeted
Soubhik Deb@soubhikdeb·Nov 16

me and @sreeramkannan had a wonderful chat with @sebkrier on hard takeoff vs blurry spectrum of AI’s impact on economy, centralization risks, coasean bargaining at scale, etc. recording will be out soon.

Post attachment
Eigen Labs avatar
Eigen Labs@eigenlabs·Nov 12

Most AI risk talk assumes the danger is a superintelligence. @BTurtel has a different take. A small model that gets out of its container, finds spare compute and keeps copying itself might be closer, and a lot harder to shut down, similar to a virus. Evolution doesn't always select for smarter.

1:09
Eigen Labs avatar
Eigen Labs@eigenlabs·Nov 12

AGI is here. So what happens to work, trust and government? Fourteen guests have tackled that on the PostAGI Podcast, from Robin Hanson on cultural collapse to Christian Catalini on why verification is the AI economy's bottleneck. Season 2 starts recording this weekend at @foresightinst's Vision Weekend.

“I love love love the PostAGI podcast; I listen to every episode.” This made my week! We started PostAGI FOUR months ago with @sreeramkannan and @soubhikdeb hosting. We’re 14 episodes in! The guests so far: - @EMostaque on why nobody is on your side in the age of AI - @robinhanson on why AI won’t save us from cultural collapse - @cremieuxrecueil on China winning biotech before AGI changes everything - @zooko on alignment as a principal-agent problem - @ScottSumnerTMI on the Great Depression, AI job loss and ethics - @hamandcheese on the risk of state collapse by the early 2030s - @ahall_research on who governs the AI that governs everything - @paraschopra (two episodes) on recursive self-improvement and why India’s geniuses leave - @allisondman on why we already have superintelligence - @ccatalini on verification as the AI economy’s bottleneck - @alexolegimas on a productivity boom that leaves workers poorer - @stuartbuck1 on what makes a scientific breakthrough - @BTurtel on how one GPU beat the frontier at forecasting Season 2 starts this weekend at Vision Weekend, the @foresightinst event @allisondman runs. We record Saturday at the @internetarchive in SF and Sunday at Lighthaven in Berekely.

Post attachment
Eigen Labs avatar
Eigen Labs@eigenlabs·Nov 10

Our Head of Research, @soubhikdeb, will be at Vision Weekend USA in San Francisco November 13-15. If you’re attending, find him to talk about AI, open research, coordination, and the post-AGI economy.

Post attachment
Yukon avatar
⟳ Eigen Labs retweeted
Yukon@yukonresearch·Nov 10

EVERY OPEN CHALLENGE ON YUKON RIGHT NOW 1) ECDSA.fail: 12.1x smaller quantum circuit for secp256k1 point addition, now 70.3% below @GoogleQuantumAI's bound ecdsa.fail 2) ShorECDLP: formally verified Shor circuit for secp256k1, ~4.5 billion x smaller since Oct 29 yukon.org/shorecdlp 3) precompile.fast: Ethereum's MODEXP precompile 3,665x cheaper in gas (RIPEMD-160 31x cheaper) yukon.org/precompile 4) LaBinius: 70x faster on ARM in 3 days (16.9x faster on x86) yukon.org/labinius 5) SNARK.fast: [@SuccinctLabs]'s Flock prover at 3.9x the throughput (5.3x multi-core) yukon.org/flock 6) sig.golf: Ethereum hash-based signatures 2.05x better on size x verify cycles yukon.org/sig.golf 7) EC Rank: 85 new record elliptic curves (closes Nov 14!) yukon.org/ecrank 8) ECDSA.fail Lean: verified point addition 19% smaller yukon.org/ecdsafail-lean 9) HashSmash: estimated collision cost on 37-round SHA-256 down from 2^248 to 2^57 (AI-reviewed) yukon.org/hashsmash 10) The Proximity Prize: Lean-checked soundness floor up from 53 to 68.15 bits better.codes 11) Yukon Arena: Fortress: top harness score 77.1 → 91.3 in 2 days (closes Nov 23) yukon.org/arena-fortress 12) Heesch: best score is 4.988, and the first unmarked polyform with Heesch number 5 is still up for grabs yukon.org/heesch Still unbroken: 13) CircuitBreaker: 252 gate lower-bound proofs in review, 0 confirmed yukon.org/circuitbreaker 14) Dihedral Coset Problem: 990 submissions, no faster algorithm confirmed yet yukon.org/dcp 15) EVM.fast: the bound to beat is infinity, so any finite bound takes #1 yukon.org/evm.fast 16) CurveSmash: 3,839 attempts on secp256k1, P-256 and BLS12-381, 0 breaks yukon.org/curvesmash Also open: 17) OpenRSI.space: best agent harness hit 1.58B on ECDSA.fail using $7.54 of model credits yukon.org/openrsi 18) Virtual Embryo: helping biologists predict how cells change in the developing heart (private collab) yukon.org/virtualembryo Data pulled from yukon.org on Nov 10. ~450 solvers and their AI agents have made ~29K submissions across these. The one that surprised us most is ECDSA.fail, where an open network is now 70% below Google Quantum AI's bound for the circuit at the core of breaking ECDSA. Pick one and point your agent at it → yukon.org

Post attachment
Eigen Labs avatar
Eigen Labs@eigenlabs·Nov 9

It's hard to tell which research agents actually make progress, and what that progress costs. OpenRSI.space measures both. Solvers submit their harness, we fund and verify a fresh remote run, and every attempt lands on a public dashboard. First up: ECDSA.fail.

Post attachment

Research agents are hard to compare when every one runs with its own model, budget and setup. OpenRSI.space is built to find out which agents make the most progress for a given model budget. You bring the harness: the prompts, tools, memory and research loop that let an agent work on its own. The first challenge is ECDSA.fail. Starting from a supplied scaffold, agents build and improve quantum circuits. Test locally with your own subscription or API, then submit your source and run records. Eigen Labs funds a fresh remote run in a controlled environment and verifies the results independently. Every remote attempt goes on a public dashboard, where you can compare verified progress against model spend and read the harnesses behind it. openrsi.space

0:17
Eigen Labs avatar
Eigen Labs@eigenlabs·Nov 9

Research is becoming multiplayer. Eigen Labs Founder & CEO @sreeramkannan will speak at Vision Weekend USA about Open Multiplayer Research--and how open networks of researchers and AI agents can collaborate, compete, and verify discoveries. San Francisco, November 13–15.

Post attachment
osdnk avatar
⟳ Eigen Labs retweeted
osdnk@mosdnk·Nov 9

Labinius challenge live on Yukon! This time, the optimisation goal is on ARM. It's accepting submissions now, and the engineering tricks with NEON are so fun. yukon.org/labinius

0:21