Artificial intelligence
Measuring Honesty With MASK
MASK separates factual accuracy from consistency under pressure; its 2026 paper and implementation changes show how much the scoring procedure matters.
THE NOTEBOOK
Blogs.
Research worth understanding. Questions worth following.
Mathematics, physics, engineering, and artificial intelligence — explained with sources, context, and room for uncertainty.
200 research articles · 3 personal notesFind your next question
203 articlesArtificial intelligence
25–48 of 50 articles
Artificial intelligence
MMLU-Pro and the Choice of a Test
A January 2026 formatting repair in MMLU-Pro shows why harder questions and more answer choices still require careful evaluation and maintenance.
Artificial intelligence
Monitoring reasoning without mistaking it for truth
New 2026 evaluation tools test whether reasoning traces expose risky behavior. Effective monitoring still depends on calibration, incentives, and realistic tests.
Artificial intelligence
Olmo 3 and the scientific value of an open training record
Olmo 3 and the October 2026 Olmo-core release show why inspectable checkpoints, data and training infrastructure matter beyond a model's final score.
Artificial intelligence
OpenVLA and the language of robot actions
An August 2026 simulation preprint tests OpenVLA-OFT on a cable-driven robot, revealing why action representations and embodiment still constrain transfer.
Artificial intelligence
OSWorld 2.0 and the desktop-agent bottleneck
Long desktop workflows expose agent failures that short demos hide. OSWorld’s 2026 releases connect task completion, process evidence, and scientific artifacts.
Artificial intelligence
Pi0.7 and learning beyond the training room
Physical Intelligence’s April 2026 Pi0.7 and October API beta connect diverse training data to robot skills while leaving generalization claims testable.
Artificial intelligence
Qwen-VL and the Problem of Visual Grounding
Qwen3-VL-Seg turns language-guided boxes into pixel masks. Its May 2026 paper exposes the difference between describing a picture and grounding a request.
Artificial intelligence
Qwen3, Sparse Experts, and the Budget Behind a Token
Follow Qwen’s sparse models from the 2025 Qwen3 report to August 2026, separating active parameters, stored weights, routing costs, and reasoning budgets.
Artificial intelligence
s1: what budget forcing changes at inference time
Budget forcing gives a reasoning model a controlled thinking window. New 2026 experiments show why longer output helps some models and unsettles others.
Artificial intelligence
Segmenting Video with a Memory
SAM 3.1 makes tracking many objects more efficient. Its March 2026 release shows why segmentation quality, identity and computation must be evaluated together.
Artificial intelligence
Shrinking the Key–Value Cache Without Losing the Question
New 2026 cache research targets attention distortion, compressed memory, and instruction retention. The real test is what a smaller cache lets a model use.
Artificial intelligence
SmolLM3 and what a small model can teach about deployment
SmolLM3's open recipe and October 2026 runtime conversion expose the separate roles of model size, memory, context, reasoning budgets and deployment fidelity.
Artificial intelligence
Sparse autoencoders and the missing features
October 2026 research separates useful reconstruction from reproducible features, sharpening what sparse autoencoders can reveal about language models.
Artificial intelligence
SpatialVLA and the Shape of Robot Knowledge
SpatialVLA encodes geometry in observations and actions. A 2026 comparison asks how that design changes robot learning, transfer and the meaning of success.
Artificial intelligence
Knowing when a reasoning model should stop
Adaptive stopping research in 2026 distinguishes useful deliberation from repetition and harmful drift, making the decision to stop part of reasoning itself.
Artificial intelligence
SWE-bench Verified and the coding-agent measurement gap
The 2026 audits of coding benchmarks reveal why hidden tests, leaked patches, and ambiguous specifications can distort claims about autonomous software engineering.
Artificial intelligence
SWE-Lancer and the economics of coding tasks
SWE-Lancer attaches historical payouts to software tasks. Its offline subset and acceptance tests show why a benchmark score should not become a wage forecast.
Artificial intelligence
From tau2 to tau3: reliable tool conversations
Tau’s 2026 knowledge and voice evaluations expose the gap between fluent support conversations and consistent, policy-correct changes to shared application state.
Artificial intelligence
Test-time training and the changing model
October 2026’s Universal Test-Time Training extends a growing family of models that learn from context, with different objectives and different guarantees.
Artificial intelligence
Titans and learning what to remember
Titans and the 2025 MIRAS framework turn memory into an online learning problem, raising precise questions about surprise, retention, and long-context evidence.
Artificial intelligence
V-JEPA and Predicting in Latent Space
V-JEPA 2.1 strengthens dense video features in 2026. The result clarifies how latent prediction connects observation, spatial detail and robot planning.
Artificial intelligence
Verify the reasoning before taking a vote
Reasoning systems can repeat the same mistake across many answers. Recent claim-level verification research asks whether targeted checks beat another round of voting.
Artificial intelligence
Wan3.0 and the engineering of controllable video
Wan3.0 and September 2026 prompt-planning research show why longer generated videos need explicit timing, faithful references, and reproducible evaluation.