LLM Daydreaming · Gwern.net

Proposal & discussion of how default mode networks for LLMs are an example of missing capabilities for search and novelty in contemporary AI systems. De­spite im­pres­sive ca­pa­bil­i­ties, large lan­guage mod­els have yet to pro­duce a gen­uine break­through. The puz­zle is why. A rea­son may be that they lack some fun­da­men­tal as­pects of human thought: they are frozen, un­able to learn from ex­pe­ri­ence, and they have no “de­fault mode” for back­ground pro­cess­ing, a source of spon­ta­neous human in­sight. To solve this, I pro­pose a day-dreaming loop (DDL): a back­ground process that con­tin­u­ously sam­ples pairs of con­cepts from mem­ory. A gen­er­a­tor model ex­plores non-obvious links be­tween them, and a critic model fil­ters the re­sults for gen­uinely valu­able ideas. These dis­cov­er­ies are fed back into the sys­tem’s mem­ory, cre­at­ing a com­pound­ing feed­back loop where new ideas them­selves be­come seeds for fu­ture com­bi­na­tions. The cost of this process—a “day

--- title: LLM Daydreaming description: "Proposal & discussion of how default mode networks for LLMs are an example of missing capabilities for search and novelty in contemporary AI systems." created: 2025-07-12 modified: 2025-07-14 status: finished confidence: possible importance: 6 css-extension: dropcaps-kanzlei ... > Despite impressive capabilities, large language models have yet to produce a genuine breakthrough. > The puzzle is why. > > A reason may be that they lack some fundamental aspects of human thought: they are frozen, unable to learn from experience, and they have no "default mod

saved by

• Jaclyn
• Freeman Jiang
• Aryan Naik
• Jirat C
• Curtis Chong
• Neil Warren
• Siddharth Ramakrishnan
• Tony Kam
• Jaeho Lee
• Andria Xu
• Florent Tavernier
• Michelle Ma

related reading

• Guardian Angels: LLM Personalization for Productivity and Security · Gwern.netgwern.net
• Lil'Loglilianweng.github.io
• LLM Powered Autonomous Agents | Lil'Loglilianweng.github.io
• the-illusion-of-thinking.pdfml-site.cdn-apple.com
• Taking LLMs Seriously (As Language Models) — LessWronglesswrong.com
• Offline RL and Large Language Models - by Sergey Levinesergeylevine.substack.com
• Measuring LLM Novelty As The Frontier Of Original And High-Quality Outputarxiv.org
• Writing for LLMs So They Listen · Gwern.netgwern.net
• MacGyver: Are Large Language Models Creative Problem Solvers?arxiv.org
• Human-like Neural Nets by Catapulting · Gwern.netgwern.net
• Towards a Typology of Strange LLM Chains-of-Thought1a3orn.com
• Simulated Users & Sad LLMs1a3orn.com