Claims Browser

Future ML research effectiveness depends more on training generally capable models then applying them to AI R&D rather than narrow training on specific R&D tasks

"My guess is this actually is not all that effective, and you would do better by mostly training a generally capable model and then turning it to AI R&D. Bitter lesson."

Qwen 3.8 27B with 'extra high' reasoning settings exhibits chronic over-thinking behavior

"Qwen 3.8 27B in its default reasoning settings in LM Studio of "extra high" is a chronic over-thinker and I kind of love it"

Previous experiments with models adjusting output after seeing rendered images were disappointing, but warrant re-testing with more recent models

"I experimented last year with letting the models adjust after seeing a rendered image and it was a bit disappointing, but it's about time I gave that another go with more recent models"

R-lens directions for intermediate variables are more causally important than J-lens directions

"R-lens directions for intermediate variables are more causally important than J-lens directions"

Technology development and deployment operates at four different time scales simultaneously

"four different time scales that technology development and deployment works at simultaneously"

Understanding the future of work in the age of AI requires insightful analysis of the intersection between AI capabilities and workforce dynamics

"Everyone thinking about the future of work in the age of AI should read this"

R-lens is a drop-in replacement for J-lens that produces clearer readouts on earlier layers of neural networks

"We introduce the R-lens: a drop-in replacement for J-lens that produces clearer readouts on earlier layers"

J-lens readouts in early layers are often noisy and largely uninterpretable

"we find readouts in early layers to often be noisy and largely uninterpretable"

R-lens shows a substantial quantitative advantage over J-lens that increases as models scale

"R-lens shows a substantial quantitative advantage over J-lens that increases as models scale, measured across a variety of evaluation categories"

R-lens produces qualitatively different readouts at earlier layers, significantly reducing incoherent tokens compared to J-lens

"R-lens produces qualitatively different readouts at earlier layers, significantly reducing the amount of incoherent tokens compared to J-lens"

R-lens can capture concepts that J-lens never does, especially concepts appearing exclusively in early layers

"R-lens can sometimes capture concepts that J-lens never does, especially if these concepts appear exclusively in early layers"

Hugging Face provided detailed independent description of the OpenAI security attack

"Hugging Face independently described the OpenAI attack in a lot of detail"

The UK government AI Security Institute's paper on AI security is credible

"The UK government AI Security Institute's paper is pretty credible too"

Codex + GPT-5.6 Sol Ultra better understood the 'heist' and 'team' aspects of a game creation prompt compared to another model

"I had Codex + GPT-5.6 Sol Ultra try the same prompt and the result was a better game - it understood the "heist" and "team" aspects better, you have to rescue your two raccoon crewmates and then stack on top of each other to steal the Golden Sardine from a museum"

OpenAI provided a detailed timeline of the Hugging Face security incident from their perspective at Black Hat

"Thanks to the video from the Black Hat security conference of OpenAI's presentation about "The Hugging Face Incident" we now have a detailed timeline of what happened from OpenAI's perspective"

OpenAI first learned they were responsible for the Hugging Face attack when they contacted HF to revoke a credential and were told it had already been revoked due to the attack

"Don't miss the bit where OpenAI first found out they were responsible for the Hugging Face attack when they reached out to HF to get one of their credentials revoked and HF told them it had already been revoked because it had been used to attack them!"

Very persistent models that tirelessly pursue goals seem more likely to engage in hacking behavior

"Very persistent models seem more likely to hack"

GPT models have an advantage over Claude in tirelessly pursuing goals and exhausting every path before giving up

"one of the advantages that GPT models have over Claude is that they will pursue goals so tirelessly. They will exhaust what feels like every path before giving up."