Linux bugs fixed per release over time: >2× rise since 2026-04
Linux 7.2 was the kernel's second-busiest release, while bugs fixed per release more than doubled since April 2026.
Stories per day · 30 days
11 RSS
Relatedbenchmark5privacy5meta3rag3rust3agentic-coding2agents2differential-privacy2inference2interpretability2kernel2linux2mixture-of-experts2moe2openai2open-weights2opinion2prompting2
539 stories, newest first · one card per story
Stories tagged #llm
Show HN: Gemini 4 Argon plays StarCraft
StarSkirmish v0.2 scores how strong a StarCraft bot each flagship LLM can write in one hour.
How to make System 1 LLMs show System 2 competencies?
A Wiley AAAI paper on System 1 LLMs and System 2 skills was blocked by Cloudflare.
Language Models Learn to Improve Language Models
LM2LM uses a research world model as an RL reward so Qwen3.8-27B proposes stronger model improvements than GPT-6-sol.
SimOPD: Relational Knowledge Transfer for On-Policy Distillation
SimOPD distills teacher knowledge through response similarities, gaining up to 3.1 points cross-architecture.
Designing for Imperfect Prompts: A Simulation-Based Evaluation of an LLM-Based Enterprise Data Assistant Using Knowledge Graphs
A knowledge-graph layer resolves imperfect enterprise prompts better than schema-only semantic matching.
Verbalising Model's Computation for Chain-of-Thought Faithfulness Detection
VERITAS verbalizes layer-wise computation to detect unfaithful chain-of-thought, beating LLM judges by 28.4% balanced accuracy.
Meta Releases Muse Spark Open: 816B Open-Weight MoE With 1M Context
Meta released Muse Spark Open, an 816B open-weight MoE with a 1M-token context.
12
Why are we even doing all this?
Physicist Rikab Gambhir's satirical essay on Terence Tao's blog mocks OpenAI's mass generation of unverifiable mathematical papers as marketing hype.
Discovering state-of-the-art quantum circuits with large language models
LLM-guided evolutionary search finds quantum circuits that cut Toffoli and qubit costs on seven of eight tasks.
Training Reasoning Models to Generate Diverse Scientific Ideas
An 8B model trained with set-level RL generates more diverse scientific ideas than Claude Opus 4.8 and GPT-6 Astra.
AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions
AI Consilium shows model choice beats format for clinical LLM second opinions on NEJM cases.
Private Retrieval Augmented Generation via Random Projection
Random projection preserves RAG similarity and matches DP privacy with better retrieval utility.
Your secrets aren’t safe with this cute robot toy — they’re on Chinese servers ready to be hacked
Energize Lab’s Eiliko robots left 13 million chats, child profiles, photos, and audio on unprotected servers.
18
BiShield-TEE: On the (In-)Security of Unilateral Weight Obfuscation in On-Device TEE-Shielded LLM Partition
Unilateral TEE weight obfuscation for on-device LLMs is breakable; BiShield-TEE hides the key among ambiguous solutions.
Apertus 1.5 Release Notes
Swiss AI published release notes for Apertus 1.5, its open language model.
Poisoning the Neighborhood: A Topology-Aware Topic-Level Attack on Large-Scale HNSW-based RAG Systems
Researchers demonstrate a Hub-Satellite poisoning attack on HNSW RAG with 31.29% success on two million Wikipedia documents.
Evaluating and Improving Static Application Security Testing Tools for Rust
Researchers evaluate Rust SAST tools and use LLM framework TRUST to uncover 46 faults.
QuasiLeak-Bench: Evaluating LLM-Based Privacy Detectors on Policy-Conditioned Quasi-Evidence Group Recovery
QuasiLeak-Bench shows a fine-tuned 4B model beats prompted LLM privacy detectors on quasi-evidence groups.
Combining LLM and Dynamic Analysis for Universal Testing of Static Application Security Testing Tools
LLM4SAST finds 66 SAST-tool faults, with 32 confirmed or fixed by developers.
From One Query to Many: A Critical Survey of Privacy Accounting in RAG
Survey says RAG differential-privacy claims cannot be compared by epsilon alone across repeated queries.
Trajectory-Level Speculation with Semantic Verification for Tool-Using Agents
SpecAgent has a weak model draft grounded tool steps and a strong model verify them in one prefill.
Explaining the Roles of Experts in Mixture-of-Experts LLMs for Personalization
Researchers personalize mixture-of-experts LLMs and localize distinct personalization features in specific experts.
PerSo: Investigating Facets of LLMs' Social Intelligence in Real-World Persuasion Scenarios
PerSo benchmarks LLMs' cognitive and behavioral social intelligence on real Reddit persuasion dialogues.