Résultats
9 articlesTrier
38:30EN
Theo - t3․gg· 14 oct. 2026Theo Tests Gemini 4 Argon: Design Strengths and Code-Review ConcernsTheo Browne · Ping LabsA strong result on a narrow skateboarding benchmark contrasts with poor findings in Theo’s own codebase audits.
30:58ENTheo - t3․gg· 29 sept. 2026GPT-6.1 Sol: Cheap, Thorough at Code Review, but Uneven at Long BuildsTheo Browne · T3Theo compares early tests of GPT-6.1 Sol with competing models, from benchmark cost to real code reviews and a 3D game demo.
31:46ENTheo - t3․gg· 29 sept. 2026Claude Sonnet 5.5: A Fast Research Assistant for Coding AgentsTheo Browne · T3Theo tests Sonnet 5.5 on pricing, reasoning effort, codebase analysis and a 3D game, and finds a surprising role for it alongside larger models.
27:29ENTheo - t3․gg· 27 sept. 2026What Pacing the AI Frontier Could Mean for Model ReleasesTheo Browne · T3Theo argues that releasing more reliable and affordable models can be compatible with slowing the advance of the most dangerous capabilities.
0:31ENTheo - t3․gg· 24 sept. 2026Grok 4.7’s Token Use and the Claims Around Its ReleaseTheo Browne · T3Theo compares Elon Musk’s earlier predictions for Grok 4.7 with selected benchmark results and token counts.
40:14ENTheo - t3․gg· 23 sept. 2026Opus 5.5: Theo’s First Day with a New Coding ModelTheo Browne · T3Theo compares price, benchmarks, coding work and game demos while cautioning that his assessment rests on a single day of use.
27:53ENTheo - t3․gg· 22 sept. 2026Grok 4.7: Strong Codebase Investigation, Uneven ValueTheo Browne · T3Theo compares launch claims, token costs, benchmarks and his own codebase and game tests, finding useful review behavior alongside significant tradeoffs.
0:23ENTheo - t3․gg· 21 sept. 2026Why AI Safety Evaluators Need More Than a Black BoxTheo Browne · T3Theo Browne explains a proposal to give outside evaluators employee-like access inside AI companies.
27:09ENTheo - t3․gg· 18 sept. 2026Why Theo Values Reliable Coding Agents Over Peak PerformanceTheo Browne · T3Theo Browne argues that a model’s weakest moments matter when a coding task stretches from investigation through testing and review.