Developer Tools●●Solid
VR.dev – Open-source verifiers for what AI agents did
38 verifiers across 19 domains catch false successes before users do.
Big BrainSolve My Problem
SkiFreeWin3
322mo ago
Back to browse
CakeML-based self-verifying, self-improving system
by cmrx64·May 31, 2026·4 points·0 comments
AI Analysis
●●●BangerWizardryBig BrainBold Bet
Gate-not-cage architecture proves the envelope, never models the inhabitant.
Strengths
- •Self-verifying checker that can improve itself through the same proof system is architecturally novel
- •Explicit labeling of Gödel/Löb limitations shows rare intellectual honesty in AI safety
- •Verified against real CakeML/Candle binaries, not just theoretical claims
Weaknesses
- •Extremely niche audience — requires formal methods background to evaluate properly
- •Not verified selfhood or alignment, only the envelope (honest but limits practical use)
Category
Target Audience
Formal methods researchers, AI safety engineers, programming language theorists
Similar To
CakeML · VeriFast · F*
Post Description
based on a conversation I had with Ramana Kumar in 2016.
Similar Projects
Education●●●Banger
TrueCitation – academic source credibility checker (URL/DOI/journal)
Cross-checks citations against 17 databases to catch AI-fabricated references instantly.
Solve My ProblemDark Horse
kitchendesign
4221d ago
Developer Tools●●Solid
Agent-skills-eval – Test whether Agent Skills improve outputs
Lightweight A/B testing for SKILL.md files when LangSmith feels too heavy.
Solve My ProblemShip It
darkrishabh
793727d ago
Developer Tools●●●Banger
Formal – LLM-driven property checker backed by Lean 4 and Mathlib
LLM-generated Lean 4 theorems that Mathlib actually proves — genuinely novel architecture.
WizardryBig BrainZero to One
yamafaktory
101mo ago
Security●●Solid
HUMA – Behavioral bot detection API, no CAPTCHAs required
Backend-only bot detection without CAPTCHAs or PII storage challenges Turnstile.
Big BrainSolve My Problem
Jaco07
101mo ago
Security●●Solid
Manifestinx-verify – offline verifier for evidence bundles (drift)
Deterministic offline tamper detection—pinned at capture, replayed without side effects.
Big BrainSolve My Problem
oneinx
203mo ago