Résultats
10 articlesTrier
28:37EN
Theo - t3․gg· 1 juil. 2026Theo Tests Sonnet 5 and Explains Fable’s ReturnTheoTheo compares model benchmarks and his own coding tests, finding promise in Sonnet’s subagent orchestration but questioning its task costs.
31:30ENTheo - t3․gg· 30 juin 2026Why Some AI Models Use Fewer Tokens to Solve a TaskTheoTheo explains input, output, caching and reasoning tokens, then explores a hypothesis about terse internal reasoning.
30:08ENTheo - t3․gg· 27 juin 2026GPT-5.6’s Limited Preview: Benchmarks, Agent Risks and AccessTheoTheo reviews the launch-era announcement, published evaluations and reported agent incidents, while questioning the balance between safety and broad access.
44:34ENTheo - t3․gg· 22 juin 2026Theo’s Six Ideas for Better Developer Tools and PlatformsTheoFrom safer package distribution to an open mobile platform and unusual AI benchmarks, Theo shares projects he hopes others will build.
0:37ENTheo - t3․gg· 16 juin 2026Theo Criticizes Fable 5 Access and Benchmark RestrictionsTheoTheo points to restricted access, fallback models and refusals in one coding benchmark as reasons for concern about Anthropic's Fable release.
29:32ENTheo - t3․gg· 15 juin 2026Theo on Fable’s Safeguards, Data Retention, and TransparencyTheoTheo examines fallback behavior and an initially undisclosed safeguard, then corrects himself after finding an update and changelog.
2:30ENTheo - t3․gg· 12 juin 2026Theo’s First Look at Fable 5 and Its BenchmarksTheoTheo discusses the relationship between Fable and Mythos, benchmark refusals and early coding results.
32:53ENTheo - t3․gg· 11 juin 2026Theo’s Fable 5 Review: Benchmarks, Real Projects, and LimitsTheoTheo tests a newly released model on coding and creative work, highlighting impressive results alongside failures, safeguards and cost.
32:30ENTheo - t3․gg· 31 mai 2026Theo Reviews DeepSWE and the Limits of Coding BenchmarksTheoAn investor in DataCurve examines its coding benchmark, the results shown in the video and their limitations.
28:02ENTheo - t3․gg· 29 mai 2026Theo Tests Claude Opus 4.8 and Claude Code WorkflowsTheoA hands-on look at model behavior, early benchmarks, usage estimates and parallel agent workflows.