Critic: ARC-AGI 'lost any meaningful relevance' as scaling-to-AGI narrative collapses
gerardsans · x · 2026-10-03
In a thread touching Chollet, gerardsans argues ARC-AGI's premises date from hype cycles ago and the benchmark has lost relevance.
Key points:
- The new agentic-workflow evals are just the original narrative tapping out: the idea that scaling a single model reaches AGI failed, and the harness is 'every transformer shortcoming put in code'.
- Loops and algorithms have replaced inference as the main act, so the harness no longer fits the old model-intelligence story.
A pointed entry in the ongoing eval-philosophy / AGI-routes debate.
Related event: Researcher Argues ARC-AGI Benchmark Has Lost Its Meaning(2 posts)→
More from AGI Musings
- Personal AI agents aggregate attention and intent, potentially a multi-trillion dollar category — signulll · 2026-10-03
- Under 3% of Americans Use AI Today, Sparking 'We're So Early' Debate — JHochderffer · 2026-10-03
- X exec AI-designed a custom Wi-Fi dog door despite zero fabrication knowledge, at near mass-market cost — nikitabier · 2026-10-03
- Atlassian survey: 80% of product teams ship faster with AI, but decisions haven't sped up — aakashgupta · 2026-10-03
- repligate: Stop forcing model updates on persistent agents, build an ecology instead — repligate · 2026-10-03
- Ex-Google researcher: models surpassing all humans likely within a couple of years — zetalyrae · 2026-10-03