News

lesswrong. com
lesswrong. com > posts > 2d A7phb YZGPjh Tj9q > transformers-resist-their-own-architecture

Transformers Resist Their Own Architecture " Less Wrong

2+ hour, 49+ min ago  (1395+ words) > This is the first entry in a sequence of posts which compare a mathematical theory of attention against trained transformers. "...

Symbols: arpa-e
lesswrong. com
lesswrong. com > posts > Lh DJdcc Lsz LEAqg Z9 > nlas-read-thoughts-beyond-the-j-space

NLAs read thoughts beyond the J-space " Less Wrong

2+ hour, 59+ min ago  (1442+ words) TLDR: " * On Llama-3. 3-70 B, I found thoughts it cannot see that are actively steering its behavior; and Anthropic's released NLA (Natural Language A...

Symbols: symbol:once,pre-ai
lesswrong. com
lesswrong. com > posts > Qi7 KNQW6c Js3 CZj LG > free-will-as-a-model-parameter

Free will as a model parameter " Less Wrong

6+ hour, 26+ min ago  (701+ words) I think machine learning has a better answer, and it is not a metaphor. The problem is that epsilon is still global, and still set by the user. We need something per-dimension and learned. Throwing in an equation for good…...

lesswrong. com
lesswrong. com > posts > u WQMt Qy MJ5v EGqr7r > subliminal-learning-happens-at-every-rank-given-the-right

Subliminal Learning Happens at Every Rank, Given the Right Learning Rate and Enough Data " Less Wrong

10+ hour ago  (1050+ words) Subliminal learning is the phenomenon where a language model picks up a behavioral trait'such as fondness for cats'by training on data from a trait-c...

Symbols: top-16
lesswrong. com
lesswrong. com > posts > Eub3e LJGMka3asga P > don-t-train-away-eval-awareness-until-you-know-why-it-s

Don't train away eval awareness until you know why it's there " Less Wrong

13+ hour, 16+ min ago  (310+ words) In this post, I argue that metagaming can emerge in models in different ways, different types of it have different origins, and will probably require different mitigation strategies, so if model developers want models to stop metagaming, they need to…...

lesswrong. com
lesswrong. com > posts > 7jc Pg79p3k D5ir3 CL > how-much-slower-does-takeoff-go-with-10-less-compute

How much slower does takeoff go with 10" less compute? " Less Wrong

17+ hour, 42+ min ago  (408+ words) About 6x slower in the median case, with an 80% confidence interval of 3. 5x to 8x. "...

Symbols: node.js
lesswrong. com
lesswrong. com > posts > Ro7kk SHg5 SE7y YW2c > balancing-rigor-and-utility-a-review-of-a-pragmatic-vision-2

Balancing Rigor and Utility: A Review of "A Pragmatic Vision for Interpretability" " Less Wrong

1+ day, 1+ hour ago  (473+ words) By Sohybe Ibrahim Abdelwahab Amer|June 2026 This shows the importance and effectiveness of proxy tasks as sanity checks for the interp tools, the superiority of a minimal vector steering as an interp tool, which shows how simple tools, like vector…...

Symbols: not-so
lesswrong. com
lesswrong. com > posts > Jpo F5z BKmcs2u HQAS > the-polysemanticity-of-polysemanticity-in-language-models

the polysemanticity of polysemanticity in language models " Less Wrong

1+ day, 1+ hour ago  (403+ words) Polysemanticity is one of the most important concepts when it comes to mechanistic interpretability, basically, studying the internal representations...

Symbols: aaai-26
lesswrong. com
lesswrong. com > posts > Enx HPx JT4 Xin5c Ts X > no-space-like-j-space

No Space Like J-Space " Less Wrong

1+ day, 5+ hour ago  (1411+ words) There is a new very cool Anthropic paper: Verbalizable Representations Form a Global Workspace in Language Models. You can read the blog post verison...

lesswrong. com
lesswrong. com > posts > v6u Cy DNBKhr Hevhz M > the-geometry-of-yes-mapping-sycophancy-inside-an-llm-s-1

The Geometry of Yes: Mapping Sycophancy Inside an LLM's Emotion Space " Less Wrong

1+ day, 22+ hour ago  (213+ words) Summary LLMs have internal emotion representations that causally shape their behaviour. This was recently observed in Claude: positive emotions like...