More from Stanford AI Lab | AI News
AI News
Stanford AI Lab
@StanfordAILabThe Stanford Artificial Intelligence Laboratory (SAIL), a leading #AI lab since 1963.
|
Transformer MLPs Encode Facts Without Training
According to StanfordAILab, researchers present a closed form method to write facts into Transformer MLPs without gradient descent, accepted to COLM 2026. (Source) 07-23-2026 04:19 |
|
PNAS Generative AI Law Feature Highlights
According to StanfordAILab, PNAS launches a Generative AI law feature spanning safety, copyright, governance, and interpretation. (Source) 07-21-2026 17:40 |
|
YC Paper Club Highlights Chips and Inference
According to StanfordAILab, YC Paper Club covered ParallelKittens, intelligence per watt, CUDA lessons, and heterogeneous inference hardware. (Source) 07-16-2026 21:28 |
|
TRACE Boosts Qwen3.6-27B to 73.2% on SWE-bench
According to StanfordAILab, TRACE trains agents on missing skills, pushing Qwen3.6-27B to 73.2% on SWE-bench Verified with under a quarter rollouts. (Source) 07-09-2026 23:45 |
|
LLM-as-a-Verifier Delivers SOTA Across 4 Benchmarks
According to StanfordAILab, LLM-as-a-Verifier scales verification to SOTA on Terminal-Bench V2, SWE-Bench Verified, RoboRewardBench, MedAgentBench. (Source) 07-09-2026 23:45 |
|
Distill to Detect exposes hidden LLM bias
According to StanfordAI Lab, D2D amplifies subtle fine tuning shifts into text to reveal hidden LLM bias for auditors. (Source) 07-09-2026 23:30 |
|
Language Models Fail from Data Repetition
According to StanfordAILab, an ICML 2026 workshop oral shows internal data repetition degrades language models, earning runner-up recognition. (Source) 07-09-2026 23:29 |
|
Stanford AI Lab unveils ICML 2026 highlights
According to StanfordAILab, Stanford AI Lab lists ICML 2026 papers on coding agents, LLM reasoning, safety, interpretability, and science. (Source) 07-06-2026 11:07 |
|
LLMs Consensus Fails Truth: ICML Analysis
According to StanfordAILab, LLMs align on each other’s answers over truth, so polling fails; ICML paper finds truthfulness does not scale without verifiers. (Source) 07-05-2026 17:34 |
|
Data Deduplication Findings Reveal 33% Compute Waste
According to StanfordAI Lab, residual repetition after deduplication can waste up to 33% of FLOPs, with worst-case patterns predictable by model size. (Source) 07-05-2026 00:20 |
|
Freeform Preference Learning Boosts Robot Policy
According to StanfordAI Lab on X, Freeform Preference Learning uses natural language axes to learn conditional rewards and yield better robot policies. (Source) 07-02-2026 18:02 |
|
QuasiMoTTo Cuts Inference Costs 25–47%
According to StanfordAI Lab, QuasiMoTTo uses correlated sampling to match LLM performance with 25–47% fewer samples and 50% fewer RL steps. (Source) 07-02-2026 17:44 |
|
OpenThoughts-Agent v2 Tops 7 Benchmarks
According to StanfordAILab, OpenThoughts-Agent-v2 leads across sizes and 7 agentic benchmarks in compute-controlled tests. (Source) 06-25-2026 07:51 |
|
Spiral RL Unifies Parallel and Sequential Reasoning
According to StanfordAILab, Spiral uses set RL to generate cooperative samples and standard RL to aggregate them into stronger answers. (Source) 06-24-2026 22:07 |
|
SPIRAL Unifies RL to Scale Reasoning Compute
According to StanfordAILab, SPIRAL trains LLMs to coordinate sequential, parallel, and aggregative reasoning with end to end RL for better answers. (Source) 06-23-2026 23:24 |
|
M* Runtime Beats Specialized Systems by 12.5×
According to StanfordAI Lab, M* unifies multimodal inference and outperforms specialists, up to 2.7x for TTS and 12.5x for world-model rollouts. (Source) 06-18-2026 21:52 |
|
DeLM Orchestrates Agents Cheaper and Faster
According to StanfordAILab, DeLM boosts agent tasks and cuts cost, with ~10% SWE-bench Verified gain using Gemini 3 Flash at under half the cost. (Source) 06-17-2026 18:30 |
|
Stanford AI Lab unveils video benchmark Analysis
According to StanfordAILab, a new YouTube-linked demo spotlights a Stanford AI Lab video understanding benchmark with metrics and research takeaways. (Source) 06-03-2026 22:18 |
|
CVPR2026 Highlights Showcase SAIL Breakthroughs
According to StanfordAILab, Stanford SAIL spotlights CVPR 2026 papers and methods with real-world vision AI impact, per the Stanford AI Lab blog. (Source) 06-03-2026 17:36 |
|
Local AI Platforms Shape US China Race
According to StanfordAILab, the US China race hinges on whose models, chips, and frameworks run by default on billions of devices, per Foreign Affairs. (Source) 05-29-2026 18:42 |