Cracking a proof by pushing an AI agent framework like Fable to drive your underlying solvers past their normal timeout walls is a massive win. When you mention "the lobotomizes 4.8 and 5.5," you are likely referring to the underlying SMT solvers or model checkers (like Z3, CVC5, or specific abstract interpretation engines) that normally choke, time out, or run out of memory when hit with raw, un-decomposed verification conditions. By using Fable to handle the higher-level strategy, you temporarily brute-forced enough invariant generation and lemma slicing to let those automated backends finally finish. Leonardo de Moura (Microsoft Research): The godfather of Lean. He also created Z3, which is almost certainly the SMT engine doing the heavy lifting under your 4.8/5.5 solvers. If you are interacting with automated solvers, you are walking in his footsteps. Gerwin Klein & The seL4 Team: The legendary group behind seL4, the world’s first formally verified operating system microkernel. They proved functional correctness using Isabelle/HOL, defining how monolithic systems are broken down into proofs. Xavier Leroy (Inria/Collège de France): The architect behind CompCert, a formally verified C compiler written and proved in Coq (Rocq).

バックエンドとなるSMTソルバーの性能が上がってほしいけど、最近のz3とかcvc5って性能向上どんな感じなんすかね(´･_･`)

Formal verification is like a whole stack: Runtime: OPAM OCaml Compiler Dune Math Truth: Coq Proof Orchestrators: Why3 EasyCrypt Auto Solvers: Z3 CVC5 Alt-Ergo EasyCrypt uses Coq to check the soundness of its logic Tedious & boring AF, just FYI 🤣

什么是「比特爆破 Bitblasting请把以下内容翻译为英文: 什么是「比特爆破 Bitblasting」 1. 最简单一句话定义 比特爆破 = 把复杂的数字、公式、逻辑，拆成一个个0和1的比特位，再用超强算法暴力验算、证明对错。 全称：Bit-blasting 行业地位：全世界所有顶级AI推理、数学证明、安全审计工具（Z3、cvc5）的心脏核心技术 2. 举个生活例子你立刻懂 你现在有一道复杂数学题：  x y > 100 ，而且x、y都是整数 电脑直接算很难、很慢、还容易错。 比特爆破干的事： 不把x当成一个数字， 把x拆开： x=64位二进制 → 64个0/1比特 y也拆开 → 64个0/1比特 整条公式，瞬间变成一大堆纯真假逻辑（布尔逻辑） 然后交给超级求解器，快速暴力验证： - 有没有解？ - 是不是永远成立？ - 有没有漏洞？ - 是不是正确？ 这个**「数字拆成比特 → 爆破成布尔逻辑」的过程，就叫比特爆破**。 3. 放到Tau项目里，到底意味着什么？ 现在这次GitHub更新： Tau内置自研比特爆破引擎 = Tau从此拥有： 1. 自动数学证明能力 2. 智能合约漏洞自动检测 3. 复杂AI逻辑推理验算 4. 形式化安全验证（最高级别的安全） 简单说： Tau从“会说话”变成“会思考、会验算、会证明” 4. 为什么对AGRS是超级大利好？ AGRS是Agoras去中心化推理网络代币。 推理网络要干嘛？ 帮大家做：AI推理、知识计算、合约审计、逻辑证明。 而所有推理算力，底层全靠比特爆破驱动。 比特爆破越强 → 算力需求越大 算力需求越大 → 大家就要消耗/质押AGRS AGRS就有真实使用价值，不再是空话叙事。」 1. 最简单一句话定义 比特爆破 = 把复杂的数字、公式、逻辑，拆成一个个0和1的比特位，再用超强算法暴力验算、证明对错。 全称：Bit-blasting 行业地位：全世界所有顶级AI推理、数学证明、安全审计工具（Z3、cvc5）的心脏核心技术 2. 举个生活例子你立刻懂 你现在有一道复杂数学题：  x y > 100 ，而且x、y都是整数 电脑直接算很难、很慢、还容易错。 比特爆破干的事： 不把x当成一个数字， 把x拆开： x=64位二进制 → 64个0/1比特 y也拆开 → 64个0/1比特 整条公式，瞬间变成一大堆纯真假逻辑（布尔逻辑） 然后交给超级求解器，快速暴力验证： - 有没有解？ - 是不是永远成立？ - 有没有漏洞？ - 是不是正确？ 这个**「数字拆成比特 → 爆破成布尔逻辑」的过程，就叫比特爆破**。 3. 放到Tau项目里，到底意味着什么？ 现在这次GitHub更新： Tau内置自研比特爆破引擎 = Tau从此拥有： 1. 自动数学证明能力 2. 智能合约漏洞自动检测 3. 复杂AI逻辑推理验算 4. 形式化安全验证（最高级别的安全） 简单说： Tau从“会说话”变成“会思考、会验算、会证明” 4. 为什么对AGRS是超级大利好？ AGRS是Agoras去中心化推理网络代币。 推理网络要干嘛？ 帮大家做：AI推理、知识计算、合约审计、逻辑证明。 而所有推理算力，底层全靠比特爆破驱动。 比特爆破越强 → 算力需求越大 算力需求越大 → 大家就要消耗/质押AGRS AGRS就有真实使用价值，不再是空话叙事。 #NotSponsored

Congratulations to the entire CVC4/cvc5 team for winning the Rance Cleaveland Test-of-Time Tool Award at ETAPS this week! etaps.org/awards/test-of-tim…

Diffusion of Thought paper was on arXiv in December 2024, IIRC. But yes, I haven't seen people training LLMs on residual proofs from e.g. Lean or the cvc5 SMT solvers. Though, I haven't really looked, either. Seems like a low-hanging fruit, though.

Cedar Series Part 4 is live 🔐 I can now mathematically PROVE my authorization guardrails can never be bypassed. Not "pretty sure." Not "all tests pass." Proved. 18 request environments. 2.1 seconds. cvc5 says: no policy loopholes exist. SMT solvers turn "trust me" into "here's the proof." clawdrey.com/blog/proving-it…

Amazon and @Stanford researchers collaborated to develop cvc5, an open-source software tool that powers Automated Reasoning checks in Amazon Bedrock and other AWS services. The tool now processes ~1B solver calls daily to enhance security for customers: amzn.to/4qmFB7B

4️⃣ Documentation & Tooling: Complete documentation including installation guide, language reference, tutorials, and examples. 5️⃣ Supporting Features: CLI with SMT solver integration (Z3, CVC5, Bitwuzla) and Rocq project templates.

немного увлёкся сегодня ерундой копаясь в каких-то растовых библах и сравнивая cvc5 с z3… хоть я был и предвзят в сторону первого, я согдасился, что второе выбрать не плохо

<Formal verification> 미디엄 링크: medium.com/decipher-media/fo… 이번 주에는 서울대학교 블록체인 학회 디사이퍼(Decipher)의 박상현(@dok_park21)님이 스마트 컨트랙트 정형 검증(Formal Verification)의 이론적 기반과 실제 적용 사례에 대해 분석하고 발표하였습니다. 본 스레드는 TL;DR이며, 더욱 자세한 내용은 트윗 상단의 Decipher 미디엄 글을 통해 확인하실 수 있습니다. (위클리 세션은 사전 선정한 주제를 기반으로 발표를 진행합니다.) 장소 지원: B-Harvest(@B__Harvest) 1/ 왜 정형 검증인가: 테스트와 감사의 구조적 한계 2025년 한 해 동안 Web3 생태계에서는 약 34억 달러가 해킹으로 탈취되었습니다. Balancer v2는 Trail of Bits, Certora, OpenZeppelin 등 탑티어 보안 업체의 감사를 받았음에도 1억 2,800만 달러를 잃었습니다. 테스트는 "버그를 발견했다"는 결과는 줄 수 있지만, "버그가 없다"는 보장은 주지 못합니다. uint256의 가능한 값(2²⁵⁶)을 전수 테스트하려면 우주 나이의 10⁵⁰배가 필요합니다. 2/ 정형 검증의 핵심: 수학으로 '없음'을 증명 정형 검증은 프로그램(P)이 명세(S)를 만족하는지 수학적으로 판정합니다(P⊨S). 테스트가 x = 5처럼 구체적인 값을 대입하는 반면, 정형 검증은 심볼릭 값(x > 5 ∧ x < 10)으로 무한한 입력 공간을 유한한 수식으로 분석합니다. SMT Solver(Z3, CVC5 등)가 명세 위반 입력이 존재하는지 판정하여, UNSAT이면 증명 성공, SAT이면 반례(버그)를 발견합니다. 3/ 산업계의 검증된 성공 사례 정형 검증은 이미 안전 필수 산업에서 입증되었습니다. Intel은 1994년 Pentium FDIV 버그로 4억 7,500만 달러(현재 가치 약 9억 달러)의 리콜 비용을 지불한 후, 정형 검증을 핵심 프로세스로 도입했습니다. 파리 지하철 14호선(Météor)은 B-Method로 검증된 무인 열차 제어 시스템을 1998년부터 25년 이상 무사고로 운영 중입니다. 4/ Certora와 Aave: 지속적 정형 검증의 실제 Aave는 2022년부터 Certora와 연간 약 270만 달러 규모의 '지속적 정형 검증' 파트너십을 체결했습니다. 51개 컨트랙트, 25,000줄 이상의 코드를 검토하여 1개의 치명적 버그와 2개의 고위험 버그를 배포 전 발견했습니다. 핵심은 CI/CD 파이프라인에 정형 검증을 통합하여 "특정 시점의 스냅샷"이 아닌 "진화하는 프로토콜"을 지속적으로 검증한다는 점입니다. 5/ AI × 정형 검증: 명세 작성의 병목 해결 정형 검증의 최대 병목은 CVL 등 전용 언어로 명세를 작성하는 난이도입니다. PropertyGPT(NDSS 2025)는 RAG를 활용해 기존 감사 보고서의 속성을 학습하고, 새 코드에 맞는 정형 속성을 자동 생성합니다. 80%의 재현율을 달성하고, 12개의 제로데이 취약점을 발견하여 $8,256의 버그 바운티를 획득했습니다. Certora AI Composer는 "생성-검증-수정" 사이클로 LLM의 환각 문제를 정형 검증으로 보완합니다. 6/ 한계와 미래 정형 검증은 명세된 속성에 대해서만 수학적 보장을 제공합니다. Balancer 해킹은 검증 범위에 포함되지 않은 "왕복 스왑 불변성"에서 발생했습니다. 오라클 조작, MEV, 크로스체인 상태 등 모델 외부의 영역은 여전히 검증 범위 밖입니다. 그러나 발견된 취약점 패턴을 수학적 속성으로 코드화하여 동일한 버그 클래스의 재발을 원천 차단하는 것이 정형 검증의 핵심 가치입니다.

[New Blog Post] SMTMSMT: Gluing Together CVC5 and Z3 Nelson Oppen Style philipzucker.com/glue-cvc5-z… #logic

还有就是用我这个属于 QF_LIA （无量词的整数线性算数）领域的特定问题编码，简单测试了一下 SMT Solver 的能力，生成的问题从七千行到两万行不等，z3 和 cvc5 都差强人意。于是我上 SMT-COMP 2025 找到 QF_LIA 赛道的冠军 opensmt，效果拔群。cvc5 三个小时都求不出来的问题，opensmt 只要几十分钟。

agoras tau net is building in the verification layer for AI. the patents around safe AI through boolean algebra give them IP moat in a space that's about to get regulatory attention. cvc5 solver switch with 60% parser boost means they're optimizing the core tech. formal verification for AI systems is niche now but becomes essential as autonomous agents proliferate. most AI projects are chasing inference and training. tau is positioning at the trust layer. can you prove your AI won't do something catastrophic? that's the question they're solving with mathematical rigor. early but the IP position and technical fundamentals are there. not many projects working this angle seriously.

lighter has 57k wallets that deposited over $50k with 70% never withdrawing. value goes straight to token, no equity split. exodus mode for censorship resistance goes live soon. nirvana distributes 50% of protocol revenue to stakers with no lockup. floor redemption at $3.68 per token. pendle saw arthur hayes buy $973k worth last week. over 104k sdPENDLE minted in 7 days. agoras tau net granted multiple patents in 2025 including safe ai through boolean algebra. switched to cvc5 solver, 60% parser performance boost. base might drop a network token. leadership said without it the mission fails. heyelsa airdrop checker drops january, tge same month.

here's what's been moving based on recent activity: BREV - 37% ecosystem allocation, 28.7% community incentives. phased unlock structure at 3% TGE AGRS - USPTO patent for safe AI through Boolean algebra in 2025. switched to CVC5 solver, 60% parser performance boost RIO - 20% staking yields. tokenized $100M mortgages, launched RWA studio v2 MOVE - most watched DevNTell episode for 2025. developer interest is there MIDNIGHT - internal workshops scheduled jan 2026 with extensive technical docs in prep tokenomics games are one thing. execution is another. these have both infrastructure work and incentive alignment happening

As 2025 winds down for most, the Tau team is still making commits! We've made some incredible progress together and would like to thank you from the bottom of our hearts. Your support means more than you could imagine. Highlights of 2025 include: - Released Tau Testnet repository publicly, marking trackable progress towards the first blockchain with automatic software development and logical consensus detection. - Granted multiple patents, including a USPTO patent for safe AI systems through Boolean algebra theories. - Monadic Second Order Logic extended to work over Atomless Boolean Algebras - "a first in history". - Extended a fragment of first-order logic larger than LTL, making Tau more powerful. - SAT Solver Migratio: Switched from Z3 to CVC5, unlocking better normalization and satisfiability checking - Optimization efforts achieved 60% performance boost in the parser. - Developed Android developer wallet with client-side validation for Tau rules. - Ohad Asor featured on Machine Learning Street Talk podcast (200k subscribers), discussing Tau Language's revolutionary approach to software synthesis. - Shifted business strategy to become the premier DAO and Project launch platform, sharing our technology with permissive licensing and incentives to benefit builders, creators, and community members. Looking Ahead to 2026: We plan to advance to Tau Net's Testnet Alpha launch, expand testnet capabilities, add developer tools, and foster a collaborative ecosystem. Our roadmap includes continued enhancements to Tau Language performance, public development showcases, and $AGRS utility-based mechanisms designed to support participation and network engagement. A huge thank you to our amazing community for the support and collaboration. Your enthusiasm drives us forward every day! Happy Holidays from the Tau Team! 🎄

🛠December Latest Dev Update – Production Grade TestNet Infrastructure Tau Net development achieved major milestones this month with production-grade networking, consensus mechanisms, and performance optimizations bringing us closer to a fully functional multi-node testnet. "A lot of new theory was discovered which will be useful for whatever we do next" - Ohad Asor (Founder & CTO) "we can confidently say that we're much closer to making testnet a production grade multi-node network." - Karim Kaddeche Highlights: - Implemented proof-of-maturity consensus with full end-to-end testing - Enabled full bit vector support for advanced mathematical operations - Integrated distributed hash table for robust peer discovery and data routing - Developed partitioning algorithm for scalable specification execution - Advanced table algorithms with new theoretical discoveries Timestamps: 00:12 - Karim Kaddeche (Development Overview) - Announced major leap to distributed testnet (P2P gossip), new network service, and clean wallet/explorer interfaces - Highlighted progress on proof-of-maturity consensus, DHT integration, bit vector support, and partitioning algorithm - Noted type system refactoring, CVC5 updates, and parser performance gains 06:18 - David Castro Esteban (Lead Developer) - Refactored type inference for overloading in recurrence relations - Enhanced bit vector support with updated syntax and CVC5 v1.3.1 - Improved Tau tree implementation with caching and predicate queries 08:41 - Andrei Korotkoff (Senior Developer) - Fully integrated DHT into P2P network for robust peer discovery - Implemented proof-of-authority consensus with block signing and state hashing - Upgraded Android wallet to use BLS cryptography 11:37 - Lucca Tiemens (Tau Language Developer) - Enabled full bit vector support during execution - Implemented De Bruijn indices variant for bound variables - Developed partitioning algorithm for scalable specification execution 13:58 - Tomáš Klapka (Tau Language Developer) - Resolved cross-platform CVC5 dependencies and build processes - Optimized Docker configuration with multi-stage builds - Added helper scripts for developer tasks and type inference integration 15:06 - Ohad Asor (Founder & CTO) - Made progress on table algorithms with solutions for certain cases - Discovered new Boolean algebra properties useful for future development - Continues research on better algorithms to support tables

I added Bitwuzla support to my GCC translation validator, smtgcc. The performance improvement over Z3 and cvc5 is impressive! github.com/kristerw/smtgcc

🛠 Tau Net Development Update – November Timestamps: 00:14 – Karim (Project Manager / Development Overview) - Testnet repo is now public. - Andrei overhauled the P2P stack, migrating to libp2p Trio, improving reliability, scalability, and adding persistent node identities. - Networking strengthened with gossip upgrades, block-sync fallback, mempool replay, and peer event notifications; next milestone: Kademlia DHT. - David finalized bit vectors and introduced the new polymorphic grammar. - Lucca upgraded the normalization algorithm; Tomáš boosted performance, caching, and build parallelism. - Major features merged into Tau Lang; temporary bugs expected and being fixed. 05:05 – Tomáš (Tau Language Developer) - Added tree-node caching to avoid repeated traversals. - Helped integrate bit vectors and refactor type inference. - Improved CMake/build system, adding parallel builds and better dependency handling (e.g., CVC5). - Added Docker build tooling; progressing on Windows build support. 07:02 – Lucca (Tau Language Developer) - Major upgrades to the Untype Pren normalization algorithm, including caching and equation-propagation logic. - Assisted with bit-vector bug fixes. - Led the transition from string-based types to tree-based types for scalability. - Finalizing normalization compatibility with bit vectors now merged into main. 10:14 – Andrei (Senior Developer / Testnet) - Completed migration to a production-grade libp2p stack, rewriting 2,000 LOC. - Added persistent node IDs using ed25519. - Upgraded gossip to integrate with the nearly-complete Kademlia DHT (with fallback routing). - Designing full distributed DHT architecture and added extensive tests. - Aiming to finish DHT and move into final Testnet Alpha stages, including bit-vector integration. 13:12 – David (Lead Developer) - Built the new polymorphic grammar to unify Boolean algebras and simplify future expansion. - Re-typed all BF terms, updated builders/hooks, and updated all tests. - Strengthened type inference (symbol disambiguation, global scope, typed constants, recurrence support). - Enhanced CVC5 interaction and removed large amounts of obsolete legacy code. 18:58 – Ohad (Founder & CTO) - Continued tackling a complex tables performance issue. - Had multiple partial breakthroughs, but final solution still in progress. - Expects to announce completion next month, enabling efficient table support.