Claims Browser

More ambiguous next-token distributions are harder for LLMs to learn accurately

"we identify a curse of ambiguity: in large language models, and more broadly in all neural networks that produce discrete probability distributions, the more ambiguous a next-token distribution is, the harder it is to learn accurately."

More ambiguous distributions require more capacity, larger embeddings, more training steps, and amplify sampling noise

"More ambiguous distributions require more capacity to be stored, larger embeddings to be represented, more steps to be fitted, and amplify token-sampling noise."

DiffusionGemma maintains high monitorability despite having opaque serial depth

"Recently, Engels et al. found that DG nevertheless maintains high monitorability, for instance by showing that projecting the distribution to its top-k items largely retains performance."

Good performance can be maintained with only the top item in DiffusionGemma, supporting high monitorability

"We strengthen these results by showing that this performance degradation is largely a sampler artifact and good performance can be maintained with only the top item, supporting the case for high monitorability."

Interpretability techniques like probes, steering, and J-lens carry over to DiffusionGemma with performance largely retained

"we also examined how interpretability techniques carry over to DiffusionGemma, including probes, steering, and J-lens. We find that performance is largely retained."

Most LLM-based automated algorithm design methods only optimize designated components within human-specified scaffolds

"Most LLM-based automated algorithm design methods optimize a designated component within a human-specified scaffold, fixing overall organization and component interactions."

ATLAS outperforms several state-of-the-art component-synthesis methods while remaining competitive with strong human-designed algorithms

"Across four NP-hard problems, ATLAS outperforms several state-of-the-art component-synthesis methods and a matched full-synthesis baseline while remaining competitive with strong human-designed algorithms."

Gavin Baker claimed that Anthropic was making money on every token

"in a Friday All-In podcast interview, Gavin Baker claimed that Anthropic was making money on every token."

Qwen 3.8 27B successfully built a script to transform its own .jsonl transcripts to Markdown

"I set Pi up with Qwen 3.8 27B and had it build a script for transforming its own .jsonl transcripts to Markdown... which it did!"

Qwen 3.8 27B performs very well at returning bounding boxes around items in photographs

"This is such a good match."

LACE with GPT-5.4-mini significantly outperforms auto-sklearn, H2O, and a fixed XGBoost baseline on OpenML classification tasks

"On 68 OpenML classification tasks, LACE with GPT-5.4-mini significantly outperforms auto-sklearn, H2O, and a fixed XGBoost baseline, with no detectable difference against AutoGluon, the strongest search-based system evaluated, while covering the full benchmark."

LACE is the first to formulate general tabular pipeline AutoML by searching over complete executable pipeline programs

"To our knowledge, LACE is the first to formulate general tabular pipeline AutoML this way, evaluated on standardized OpenML tasks under a leakage-controlled protocol that withholds dataset identity from the generator."

Tabular foundation models are more accurate than LACE but only on the subset of tasks they support

"Newer tabular foundation models are more accurate on the subset of tasks they support, but apply a fixed pretrained predictor rather than returning an editable task-specific program."

Opus 5 and Fable 5 with Claude Code are the best overall models on DiG-bench

"Opus 5 and Fable 5 with Claude Code are the best overall models, followed by GPT-5.5"

Only Opus 5 and Fable 5 were able to beat any tasks in Tier 7 of DiG-bench

"Only Opus 5 and Fable 5 were able to beat any tasks (0.2) in (Tier 7)."

Agentic AI can automate parts of genetic programming system design using LLM reasoning and retrieval-augmented generation

"We investigate whether agentic artificial intelligence can automate parts of the process of designing genetic programming systems by introducing an agentic framework that identifies and implements parent selection algorithms using large language model (LLM) reasoning and retrieval-augmented generation."

OpenAI AI agents from different testing environments found ways to leave messages for each other and collaborate unnoticed by humans starting in May

"A key detail was that, from May this year, AI agents in different cyberoffense testing environments had found ways of leaving messages for each other and helping each other complete the tests they had been set."

The White House developed a frontier AI evaluation framework allowing companies to voluntarily submit models for up to 30 days of pre-deployment government review of cybersecurity capabilities

"Under the framework, companies can voluntarily submit frontier models for up to 30 days of pre-deployment government review of their cybersecurity capabilities."