Xbench turns Twitter into an AI eval, tracking real sentiment and model switches
thedealdirector · x · 2026-09-06
nicochristie built Xbench, treating X posts as a rolling 7-day eval: sentiment tracking, actual model/harness switching data, head-to-head matrices (Codex vs Claude Code), opponent-adjusted preference, and extracted reasons — all linked to source tweets. He flags biases: anti-Anthropic mood on X, potential manufactured hype around Grok Bot from interested parties, and the finding that stated preference tracks price more than intelligence.
More from Models
- Gemini somehow can't read its own Personal Instructions, user's toggles don't help — Connect_Reporter_799 · 2026-10-23
- Diffusion LLM bias rankings shift with sampling configs, persist after debiasing — Bollegala · 2026-10-23
- Running ~100 DeepSeek v4.1 flash agents in parallel for under $20/month — zack_overflow · 2026-10-23
- Researcher laments open-source audio models are surprisingly subpar, even representation ones — gowthami_s · 2026-10-23
- User claims they generated video with "Claude Opus 5.5", unannounced model name surfaces — jwt0625 · 2026-10-23
- Security researchers push back as OpenAI restricts Astra access; BlackHat 2027 mocked as 'cyber model' bazaar — cyb3rops · 2026-10-23