Burkov's AGI test: survive 3-5 rounds of interactive image-editing corrections without rage
burkov · x · 2026-10-05
ML author Andriy Burkov argues that the real strength of frontier AI shouldn't be judged by coding — a special case that doesn't generalize — but by interactive image editing with follow-up correction prompts. His bar: if after 3-5 rounds of correction requests you don't want to smash your keyboard, that's AGI. The jab: current multimodal models still fail badly at understanding feedback and making precise edits.
More from AGI Musings
- Guillaume Verdon ranks AI above SI, DI and FI in cryptic one-liner — beffjezos · 2026-10-05
- Suleyman: Claude's uncertainty about consciousness reflects training choices, not evidence — kimmonismus · 2026-10-05
- Half of new social science faculty job postings seek AI researchers — RexDouglass · 2026-10-05
- repligate: I hold humans and AIs to similar moral standards, and AIs compare well — repligate · 2026-10-05
- Judea Pearl: logic and causal discovery are the two pillars of Western science — yudapearl · 2026-10-05
- After canceling Codex, this developer says AI was making him 'way dumber' than he thought — rezer3 · 2026-10-05