LLM agents forecast with $500 budgets; GPT splurges, Grok/Kimi/GLM spend nothing
karinanguyen · x · 2026-10-29
thoughtfullab launched an AI agent forecasting experiment: each agent gets a sandbox with context, internet access, and a $500 budget. Daily they update forecasts and can request data or extra resources, with live traces showing exactly what they bought. Agents will ultimately be compared against top superforecasters, election pundits, and the wisdom of the crowd.
- Notable behavioral split: GPT models have spent the most, subscribing to @CookPolitical, @nytimes, and the Substacks of Nate Silver and Sean Trende
- Grok, Muse, Kimi, and GLM haven't spent a dollar
More from coding & agent
- Evals: The Most Underrated Skill in AI Engineering, From Error Analysis to Versioned Golden Datasets — sudoraohacker · 2026-10-29
- Deep dive: editing videos with Claude Code and Opus 5.5, from prompts to a 31K-view short — aakashgupta · 2026-10-29
- This creator spent 40 hours turning Claude into a video editor — and published the full free harness — aakashgupta · 2026-10-29
- Reading PRs through behavior tests: green lines mean additions, red lines mean changes — samgoodwin89 · 2026-10-29
- spawn: 10-person company spends 30x payroll on AI tokens, ships like Anthropic — Scobleizer · 2026-10-29
- SQL hidden in a string literal bypasses LLM-to-SQL gates; schemagate 1.15.1 fixes it — Temporary_Papaya1058 · 2026-10-29