Top all-time · #16573
Jeremy Berman
Personal blog · jeremyberman.substack.comai
Jeremy Berman writes Jeremy’s Substack, a personal blog at jeremyberman.substack.com about ai. Since 2024, 4 different Hacker News readers have linked it 4 times, most often “I got the highest score on ARC-AGI again swapping Python for English”. It is still writing: the last post was on 2025-09-16, and you can follow it by feed.
Too few sampled comments (2) to say how often HN recommends it.?
Why HN links it
Commenters cite this author's claim that LLMs have "dead reasoning zones", where consistent reasoning fails to transfer across domains, unlike humans, with ARC grids as the example.
Summarised by Claude from a sample of 2 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
My personal Substack.
Most-linked posts?
- I got the highest score on ARC-AGI again swapping Python for English2 people · 2 mentions · first linked 2025-09-17
- I came in first on ARC-AGI-Pub using Sonnet 3.52 people · 2 mentions · first linked 2024-12-08
What Hacker News says?
What you are describing is "dead reasoning zones".[0] "This isn't how humans work. Einstein never saw ARC grids, but he'd solve them instantly. Not because of prior knowledge, but because humans have consistent reasoning that transfers across domains. A logical economist becomes a logical programmer when they learn to code. They don't suddenly forget how to be consistent or deduce. But LLMs have "dead reasoning zone…
Comment linking /p/how-i-got-the-highest-score-on-arc-agi-again
on “Two things LLM coding agents are still bad at”Submitted I got the highest score on ARC-AGI again swapping Python for English as a story · 178 points
HN discussionI wonder if combinators could be useful for neurosymbolic AI—either in the backward pass (e.g., training models on synthetic data) or the forward pass (e.g., iterative code generation with evolutionary algorithms). Combinators feel alien, making even Haskell or APL seem intuitive, but maybe that’s because they don’t align with human working memory. Language models, with their massive context windows, handle long-ran…
Comment linking /p/how-i-got-a-record-536-on-arc-agi
on “Rohlang3: A point-free, homoiconic, and dependently typed "SK calculus"”Submitted I came in first on ARC-AGI-Pub using Sonnet 3.5 as a story · 1 points
HN discussion