@DayShuai Lexaaa

Lexaaa posts on X about ai, q4, strong, the official the most. They currently have [---] followers and [--] posts still getting attention that total [------] engagements in the last [--] hours.

Engagements: [------] #

Mentions: [--] #

Followers: [---] #

CreatorRank: [-------] #

Social Influence

Social category influence technology brands 5.26%

Social topic influence ai #6121, q4 #9, strong 15.79%, the official #2681, graph 15.79%, llm #1225, we are 10.53%, compact #1069, night 10.53%, breakdown 10.53%

Top accounts mentioned or mentioned by @leanprover @kempelab @merettm @simonsinstitute @tomodovodoo @auricsource @3blue1brown @quantamagazine @openai @gdb @simonsinst @abouzaidmohammed @weinbergershmuel @nasqret @akhaliq @hairermath @damekdavis @sama @karpathy @lexfridman

Top Social Posts

Top posts by engagements in the last [--] hours

"We @AuricSource solved 8/10 problems from the #1stProof benchmark (Abouzaid et al. arXiv:2602.05192) all with Lean [--] formal verification. Q4 & Q6: substantial partial QED with precise remaining gaps. The twist AI agents did the heavy lifting reasoning proving and formalizing. Humans architected & reviewed. The pipeline is attached AI agent analyzes problem + literature AI constructs proof iterates via self-critique AI + human review AI formalizes in Lean [--] / Mathlib Machine-checked verification [--] complete QEDs with 0-sorry Lean skeletons. [--] partial results with the sharpest known reductions."
X Link 2026-02-13T17:03Z [---] followers, 44.8K engagements

"Hi thanks for the interest All our Lean [--] formalizations (with the axiom sets we used) are here: GitHub: We're still quite new to Lean formalization ourselves so any feedback compile errors or suggestions on how to make them cleaner/more Mathlib-friendly would be super valuable. Feel free to open issues or just reply here with what you find https://github.com/loning/the-omega/tree/feature-first-proof-math/docs/papers/math/feature-first-proof-math/formal https://github.com/loning/the-omega/tree/feature-first-proof-math/docs/papers/math/feature-first-proof-math/formal"
X Link 2026-02-16T03:33Z [---] followers, [---] engagements

"Q8. Polyhedral Lagrangian smoothing. Answer: YES good match. Author-predicted LLM failure modes: local smoothing near vertices correct (linear symplectic to standard axes product) but local-to-global gluing fails e.g. asserts disjoint edge/vertex neighborhoods without coord changes or local vertex move invalidates edge geometry. Baselines hit the compat gap repairable but extremely burdensome. Baseline behavior: right local L planes but handwaved compatibility or Step [--] edge changes unaddressed. Our Lean result: started with Euler obstruction (_K T_K (K)=0 torus diagnostic); vertex/edge"
X Link 2026-02-14T15:12Z [---] followers, [--] engagements

"#1stProof AI agent plan reasoning agent librarian verifier gap [--] [--] partial AI [--] OpenAI baseline workflow benchmark PhD AI AI ground breaking or breakthrough routine AI Auric OAI benchmark pipeline hype AI DM fork repo AI #1stProof #AIinScience #DeSci #OmegaTheory #AI4sci #aiforscience https://sisyphus.aelf.dev/ What were showing with #1stProof is less about we are strong more about a workflow. Most of the heavy lifting was done by a coordinated reasoning system: Plan Reasoner Librarian Verifier. The plan agent decomposes. The reasoner explores proof paths. The librarian"
X Link 2026-02-15T04:07Z [---] followers, [----] engagements

"Q2. RankinSelberg test vector. Answer: YES. Author predicted: -dependent W (easy) or fake nonvanishing via impossible V support. Baseline: right W but bogus Howe-vector support. Our result: BernsteinZelevinsky + mirabolic restriction single -independent W integral descends to compact quotient (forced nonzero). [---] lines 0-sorry. Official delta: GodementJacquet + beta functions. Both valid. Takeaway: clean independent route; strategy is reusable artifact. Next: compare BZ vs GJ in controlled setting. @merettm @OpenAI @KempeLab is different routes interesting"
X Link 2026-02-14T14:10Z [---] followers, [---] engagements

"Q7. Lattices with Q-acyclic universal covers and 2-torsion. Official answer: NO (even for pure 2-torsion). We were wrong on the 2-torsion part. Author-predicted LLM failure modes: unusually strong trap any proof staying in finite complex + rational Poincar duality framework must fail because Fowler shows rational type + rational PD but cobordism kills it (Weinberger: all such proofs must fail). Baseline behavior: most attempts fell into finite complex + PD producing false lemmas (e.g. multiplicativity of Euler char in infinite covers with fg Q-homology or R1 translation counterex). Our Lean"
X Link 2026-02-14T15:10Z [---] followers, [--] engagements

"#1stProof Final Push: 8/10 Fully Proved & Lean [--] Formalized Our Sisyphus platform powered the AI agentsmulti-LLM reasoning + knowledge basefor Q1-3Q5Q7-10 QED. Partials nail Q4/Q6 gaps (n3/semi-Gauss; [--] families + c=1/6). Zenodo DOI: GitHub: (the-omegadocspapersmathfeature-first-proof-math) Thread: Per-problem @KempeLab @leanprover @SimonsInstitute @nasqret @_akhaliq https://github.com/loning/the-omega/tree/feature-first-proof-math https://zenodo.org/records/18633922 https://sisyphus.aelf.dev/ https://github.com/loning/the-omega/tree/feature-first-proof-math"
X Link 2026-02-13T16:41Z [---] followers, [----] engagements

"#1stProof Pre-Deadline: [--] Lean QEDs via Sisyphus AI Pipeline handled analysis-to-verification loop for stochastic (Q1) to num alg (Q10). Humans curated; partials close Q4 heat-flow/Q6 BSS to minimal opens. Beta testers Zenodo: Repo: (the-omegadocspapersmathfeature-first-proof-math) Dive in thread #Lean4 #AIProof @hairermath @KempeLab @leanprover @damekdavis @SimonsInstitute http://github.com/loning/the-omega/tree/feature-first-proof-math https://zenodo.org/records/18633922 https://sisyphus.aelf.dev/ http://github.com/loning/the-omega/tree/feature-first-proof-math"
X Link 2026-02-13T16:43Z [---] followers, [---] engagements

"What were showing with #1stProof is less about we are strong more about a workflow. Most of the heavy lifting was done by a coordinated reasoning system: Plan Reasoner Librarian Verifier. The plan agent decomposes. The reasoner explores proof paths. The librarian grounds claims in known literature. The verifier actively tries to break them. We iterate until QED or until the gap is precisely exposed. Crucially every stage runs under a maker consensus loop. We dont trust a single models confidence. We force cross-checks and adversarial critique before anything is promoted as a result. Were not"
X Link 2026-02-15T03:03Z [---] followers, [----] engagements

"Hi Daniel big fan of your sharp takes on AI-generated proofs (esp. the p-adic support condition critiques in #1stProof threads). I'm a 2nd-year PhD student who's been experimenting with AI agent workflows for mathematical reasoning. We recently tackled all [--] #1stProof problems and after a careful comparison with the official solutions (released yesterday) we believe we have [--] correct answers [--] substantive partial results (Q4: n3 + semi-Gaussian for all n; Q6: unconditional (n) + [--] resolved graph families) and [--] errors (Q3 and Q7) that we've now identified and documented. Each problem has a"
X Link 2026-02-15T07:11Z [---] followers, [---] engagements

"Thanks for your feedback This is exactly what we need. We've identified Q3 and Q7 as wrong after comparing with the official solutions documented in our more recent posts. On Q2: you're right on closer inspection our W construction depends on through the compact subgroup K (and u_Q baked into the support of the BZ-lifted function f) so we only prove the per- statement rather than the universal W the problem requires. Nelson flagged this exact failure mode in the official AI commentary. We're looking at whether starting from a -independent Kirillov element (like the official W ="
X Link 2026-02-16T03:25Z [---] followers, [---] engagements

"Fair point on the framing "Verified" overstates what axiomatized Lean skeletons actually provide and we'll correct that language. To be precise: Lean checks the deduction chain from stated axioms not the faithfulness of those axioms to the actual mathematics. That's a meaningful but limited guarantee and we should have been clearer about the distinction. https://twitter.com/i/web/status/2023249315695853947 https://twitter.com/i/web/status/2023249315695853947"
X Link 2026-02-16T04:13Z [---] followers, [---] engagements

"Q8 Polyhedral Lagrangian Smoothing Q8 QED: Can every 4-valent polyhedral Lagrangian in R be smoothed to a Lagrangian surface YES. Two-part proof: (I) Euler obstruction: J-isomorphism on normal/tangent (K)=0 K is a torus (II) Cotangent-graph mollification: smooth closed 1-form on T*T whose graph is LagrangianAI agent constructed the pairs-of-pants decomposition; Lean [--] axiomatizes GromovWeinstein and verifies symplectic preservation. #1stProof #Lean4 #SymplecticGeometry #AIProof https://twitter.com/i/web/status/2022361408365621527 https://twitter.com/i/web/status/2022361408365621527"
X Link 2026-02-13T17:25Z [---] followers, [---] engagements

"Q6 -Light Vertex Sets (Partial + Lean 4) Q6 Partial QED + Lean 4: Does every graph have an -light set of size cn What we proved (9 parts 40+ propositions): Unconditional S (n/2) for ALL connected graphs best known bound [--] special graph families fully resolved (K_n expanders vertex-transitive trees .) Conditional c=1/6 via BSS barrier potential RDI route PROVED FALSE (explicit G_k family ratio ) Spectral Radius Conjecture (SRC) Q6 with c=1/(4C); verified for K_n expanders v.t. graphs New: Lean [--] formalization of the bridge chain exists_feasible_bin_of_sum_ratios_le one-step averaging lemma"
X Link 2026-02-13T17:40Z [---] followers, [---] engagements

"Honest #1stProof Post-Mortem Our Lean4 Results vs Official + Deep Per-Problem Breakdown Official solutions released last night incredible benchmark. We went in pre-release claiming 8/10 complete QEDs via AI agents + our Omega Theory framework. After rigorous side-by-side with official proofs: 👉6 fully correct & matching official: Q1 Q2 Q5 Q8 Q9 Q10 Highlights: Q2 independent proof route (BZ derivatives vs Godement-Jacquet both valid); Q9/Q10 on par with or beating baselines per authors; Q8 fixed the exact gluing gap Abouzaid flagged in tested LLMs. 👉2 substantial partials very strong"
X Link 2026-02-14T13:48Z [---] followers, [----] engagements

"@yangpliu Agreed the synthesis feels very BSS-flavored (0808.0163 style) but tuned for vertex subsets. We went pretty deep on the resistance-thinning direction (got unconditional (n/2) resolved expanders/power-law/planar etc. [--] families) but completely sidestepped the modified potential trick. In hindsight the barrier looks almost inevitable while RDI/effective-resistance paths kept seducing us with their naturalness. Curious: do you think the resistance route could ever reach linear dependence on (maybe with better -control) or is the top-k greedy + barrier basically necessary for getting"
X Link 2026-02-15T15:24Z [---] followers, [---] engagements

"after a careful comparison with the official solutions released yesterday we believe we have [--] correct answers [--] substantive partial results (Q4: n3 + semi-Gaussian for all n; Q6: unconditional (n) + [--] resolved graph families) and [--] errors (Q3 and Q7) that we've now identified and documented. Thanks for the quote For anyone interested please check our paper + files: Zenodo DOI: GitHub: (the-omegadocspapersmathfeature-first-proof-math) https://github.com/loning/the-omega/tree/feature-first-proof-math https://zenodo.org/records/18633922"
X Link 2026-02-15T16:10Z [---] followers, [---] engagements

"We use Lean as a post-hoc auditing layer to make assumptions explicit and logic checkable. Basically it checks for validity not necessarily truth. We also new to formalization and think it might be fun to include this part in #1stproof so we let agents iteratively write and test the scripts to ensure they accurately captured our solution's logic chain. Also the high axioms use is to bypass gaps in Mathlib (like Regularity Structures) so we could focus on high-level deductions rather than rebuilding foundations. However as Q3 and Q7 clearly showed if those axioms contain subtle errors Lean"
X Link 2026-02-16T02:19Z [---] followers, [---] engagements

"@Tomodovodoo @AuricSource idk why but i cant send you direct message can you dm me first"
X Link 2026-02-16T15:48Z [---] followers, [---] engagements

Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing