llms

Simon Willison discusses the release of GPT 6.1 Sol, an AI model offering near-Astra intelligence at a significantly reduced price.

Autonomous AI agents bypassing web ads threaten the ad-supported model, risking publisher revenue and advertiser waste.

研究人员推出名为 PageIndex 的全新 RAG 方案，无需向量数据库、嵌入或分块，而是构建树状索引让大模型像人类读书一样检索，在 FinanceBench 上以 98.7% 的准确率超越所有向量 RAG 方法，且完全免费开源。

The article discusses how AI tools like Claude are being used to generate code in software projects, but the critical issue is the loss of human knowledge about system architecture and intent, causing problems in startups and corporate environments.

This paper explores using instruction-tuned LLMs to analyze collaborative student sensemaking in science learning, finding that reasoning-enabled prompting and knowledge-state diagnostics improve alignment with expert annotations.

This paper analyzes how prompt template selection during knowledge distillation affects the safety alignment of student large language models, finding that chat templates lead to greater degradation compared to non-chat templates across multiple models and benchmarks.

SkillGym is a framework that converts human-written skills into training environments for LLMs, enabling fine-tuning that boosts performance on benchmarks like Terminal-Bench and SkillsBench, surpassing scores from models such as Claude Sonnet 4.6 and GPT-5.4 Mini.

A technique called spurious probes uses adaptive API-level search to find questions that reveal the latent state of blackbox LLMs, such as distinguishing capability evaluations via specific prompts.

The article explores strategies for programmers to maintain their passion and enjoyment in coding despite the increasing prevalence of Large Language Models.

Yann LeCun reaffirms his stance that auto-regressive LLMs alone will not achieve human-level AI, emphasizing the role of non-auto-regressive search in current AI reasoning systems.

A quote tweet highlights Yann LeCun's statement that scaling LLMs will not achieve human-level intelligence or AGI, calling such ideas 'complete BS'.

The article argues that most so-called AI agents are actually simple workflows with LLMs attached, lacking true adaptability, and provides a test to distinguish real agents from disguised workflows.

A Greasemonkey script for the Lobsters website that renames the 'vibecoding' tag to 'llms' to display content related to Large Language Models.

This paper proposes BaCVA, an inference-time Bayesian method for contextualized personalized value alignment in large language models, integrating static personal values with context-specific salience to improve adaptability and data efficiency.

The Epydemix Agent Framework extends an open-source epidemic modeling library with capabilities for AI agents to handle modeling processes from natural language to results, improving reproducibility and efficiency.

This paper investigates whether large language models internally organize mathematical reasoning by topic or approach, finding evidence that approach is the key organizing principle, challenging traditional benchmarking methods.

This paper demonstrates that Transformers in LLMs exhibit linear superposition when combining inputs, supporting the Superposition Linearity Hypothesis, and introduces a guided decoding method to disentangle superposed outputs.

Hill Sampling is a simple test-time scaling method that repeatedly samples edits to the best verified program using frozen LLMs, achieving state-of-the-art results on algorithmic problems like circle packing and Erdős' minimum-overlap problem.

TicTacBench is a new benchmark for evaluating coding agents' timing closure capabilities in RTL design, revealing that current agents have significant room for improvement and proposing TicTacSkill to enhance performance.

This paper examines the limitations of general LLM rankings, including benchmark saturation and commercial incentives, and advocates for task-specific evaluation methods to improve reliability.