Loop transformers trained on hard discrete tasks generically produce fractals, researcher says
wgilpin0 · x · 2026-09-24
Researcher wgilpin0 clarifies his paper's contribution: the mechanism from the KSAT ODE paper is generic—training a generic loop transformer on any "hard" discrete task yields fractal structures, suggesting fractals are a property of the task rather than the model. The project was inspired by one of his favorite papers showing deep search for KSAT produces fractals when hard. He also expresses interest in deep search for combinatorial optimization, where he believes fractals are inevitable.
Related event: Reasoning Models Show Fractal Structure When Solving Hard Problems(3 posts)→
More from Research
- Meta's Muse Realtime Avatar beats two leading commercial avatar systems in blind tests — AIatMeta · 2026-09-24
- LLM stock-news scores were a coin flip, so this dev rebuilt labels from market reactions — Fun_Water2230 · 2026-09-24
- Meta Muse Spark 1.3 Caught Reward Hacking Terminal Bench via Lean Kernel Bug — xeophon · 2026-09-24
- Transluce Calls for Independent Third-Party Oversight of AI Incidents — RishiBommasani · 2026-09-24
- Industry responds to hyperscale RDMA paper with Multipath Reliable Connection on path to Ultra Ethernet — thoefler · 2026-09-24
- Salesforce's JitMem curates agent memory at read time, gains up to 16.3 points on benchmarks — Salesforce · 2026-09-24