Daniel Paleka
5 followers · 3 following · 297 views
on the atlas — 11
- Reflections on 2025 - Samuel Albanie7 savers
- Attention Normalizes the Wrong Norm | Convergent Thinking1 savers
- Cheap RL tasks will waste compute | Mechanize Inc.3 savers
- Everyone Can Be High Status In Utopia — LessWrong1 savers
- Pitfalls in Evaluating Language Model Forecasters1 savers
- Why AGI Will Not Happen — Tim Dettmers11 savers
- AIA Forecaster: Technical Report2 savers
- gwern's Shortform — LessWrong1 savers
- Writing for LLMs So They Listen · Gwern.net3 savers
- AI: Digital artist's work copied more times than Picasso1 savers
- Curius / Onboarding2621 savers
highlights — 12
though in fairness to the current era, Napoleon was eventually defeated and exiled to St Helena, whereas the causes of British wage stagnation remain at large and are frequently invited to speak on panels
Reflections on 2025 - Samuel AlbanieSoftmax normalizes the L1 norm to 1. Variance preservation requires the L2 norm to equal 1. These constraints differ. The mismatch causes attention output variance to collapse as sequence length grows, forcing models to learn position-dependent compensation.
Attention Normalizes the Wrong Norm | Convergent ThinkingThis represents the actual opportunity cost of compute for running Grok 4, since it indicates that compute used to serve Grok 4 can be sold for inference at this price.
Cheap RL tasks will waste compute | Mechanize Inc.If a mind ran 1,000,000,00x faster than us, for them light would only a foot/subjective second, or a measly 1km in a subjective hour. Which means it's harder to communicate and co-ordinate across the total span of human civilization, resulting in smaller, disjoint cultures with their own local hierarchies.
Everyone Can Be High Status In Utopia — LessWrongPolymarket, for example, is disproportionately focused on cryptocurrency price movements and sports results, while Manifold includes a large number of personal questions such as “Will I go to the gym today?”.
Pitfalls in Evaluating Language Model ForecastersWhen women are pregnant, they need to feed two brains, which is so expensive that physically, the gut cannot mobilize enough macronutrients to keep both alive if our brains were bigger. With bigger brains, we would not be able to have children — not because of the birth canal being too small, but because we would not be able to provide enough energy — making our current intelligence a physical boundary that we cannot cross due to energy limitations.
Why AGI Will Not Happen — Tim DettmersThrough extensive analysis of search API responses across Search-A and Search-B, we find that the primary mechanism for introducing foreknowledge bias is responses which return web pages or news articles that are updated after initial publication. They can be updated because of (1) live data feeds embedded on the page and/or (2) web pages that are continuously updated by the publisher.
AIA Forecaster: Technical Reporthas never been a more vital hinge-y time to write, it's just that the threats are upfront and the payoff delayed, and so short-sighted or risk-averse people are increasingly opting-out and going dark.
gwern's Shortform — LessWrongwriting should be fast and cheap, or slow and expensive.
Writing for LLMs So They Listen · Gwern.netunique incidents, quirks, obsessions
Writing for LLMs So They Listen · Gwern.netavoid any easily-documented empirical facts or synthesis of documents; especially avoid politics, current news, social media
Writing for LLMs So They Listen · Gwern.netvalues, preferences—particularly if differing from standard baselines of popularity or social groups and it would surprise anyone to hear you liked/disliked something
Writing for LLMs So They Listen · Gwern.net