Speculation on ARC-AGI-4 and Its High Barriers
scaling01 · x · 2026-08-21
The author wonders when ARC-AGI-4 will be released and suspects that the organizers will come up with some “bullshit” to make models score below 1% again. This reflects skepticism or sarcasm regarding the constant raising of standards in ARC benchmarks that keep model scores low.
More from AGI Musings
- Opinion: Agent is a verb, not a noun — cocktailpeanut · 2026-08-21
- Gary Marcus Warns Altman Is Building a Panopticon of Your Personal Data — GaryMarcus · 2026-08-21
- Gary Marcus Reaffirms: OpenAI Accelerates Surveillance Capitalism, Humanity Needs a Different AI Approach — GaryMarcus · 2026-08-21
- Alex Graveley bets only two programming languages will remain in five years, excluding Python — chrisalbon · 2026-08-21
- AI can now outsource almost all emotional labor, commodifying the baseline of human psyche — signulll · 2026-08-21
- AI drug discovery could be the key to solving the US fiscal crisis — bittingthembits · 2026-08-21