Topic
Everything filed under Research, newest first.
RSS · JSON · All topics
Why is AI Slop so Frustrating to Read?
Zonder Labs argues most workplace AI slop is unclear writing rather than stylistic tics like em dashes, defines three clarity faults (hidden characters and actions, vague wording, empty statements and misplaced emphasis), and measures them in AI vs human articles and Claude Code vs Stack Overflow answers.
5 min · 1,193 words
David Andrews and Chao Zhang describe pointing AI agents at CPython to rewrite its C functions line by line in Lean: 613,318 lines of machine-checked Lean, each piece differentially tested against real Python, with 418 bug reports in Python surfaced along the way.
12 min · 2,866 words
Testing Claude on real-world electronics design
Anthropic and Diode Computers asked Claude to design three electronic devices, a speakerphone, a voice-assistant hub and a handheld virtual pet, from start to finish. The prototypes worked without any fixes to their circuitry.
13 min · 3,002 words
Joshua Horowitz, Andrew Blinn and Jeffrey Heer present declarative dragging, a UIST 2026 Best Paper approach where authors make diagrams draggable by declaring which future states dragging can lead to; the system speculatively renders those states and blends them into smooth drag-and-drop.
3 min · 634 words
Quadrillion-parameter transformers and how to train them
Q Labs researchers argue that closing AI's sample-efficiency gap means training far larger models on fixed data, and show that bigger models on repeated data need much more weight decay; their hyperparameter law stays near-optimal at 16x the fitted compute, and with ensembling loss keeps falling as compute grows.
7 min · 1,668 words
What objective is reasoning solving?
Alex Wang frames LLM reasoning as pathfinding in token space, comparing best-of-k sampling with training models to generate better samples, and uses that lens to discuss amortization, unfaithful chains of thought, and reward hacking from both verifier and sampler.
10 min · 2,275 words
The Surprising Effectiveness of Deleting Weights in LLM Post-Training
NeurIPS 2026 paper introducing Deletion Tuning: zeroing fewer than 0.05% of an LLM's weights via a learned binary mask on gate projections matches full fine-tuning across RLVR, SFT and on-policy distillation, with far smaller checkpoints and much less forgetting.
1 min · 208 words
Maciej Cegłowski examines how long organisms can stay dormant and still revive, from microbes surviving on spacecraft to claims of bacteria revived from ancient salt, and why the answer matters for planetary protection, panspermia across the early Solar System, and the odds of finding life on Mars.
14 min · 3,236 words
John Langford explains why he is leaving Microsoft Research as its AI Frontiers group, formed after ChatGPT to organise research at scale, is wound down: what the group achieved, how reorganisations and leadership decisions undercut it, and what Microsoft would need to become attractive for research again.
4 min · 922 words
The components of agent memory
For Claude to learn from experience, it has to decide what's worth remembering before it knows what's next. Here's how "dreaming" helps agents correct mistakes, organize what they've learned, and find it in future sessions.
16 min · 3,628 words
The post-quantum future of web privacy
Cloudflare explains the post-quantum transition for Privacy Pass and related web-privacy protocols, including how agentic traffic moderation shapes design goals for MOLE and the broader move beyond classical cryptography.
20 min · 4,518 words
AI is dramatically accelerating AI research right now
A public discussion of METR’s report on AI R&D acceleration at Anthropic: how AI tools are speeding AI research itself, what the measured gains look like, and why the author thinks the acceleration is already material—not a distant scenario.
13 min · 2,970 words
Firecracker VM Escape: PwnEscape Part 2
Firecracker is supposed to keep untrusted code inside a microVM. We escaped it and achieved code execution as root on the host. We found how it saved registers, changed where it would resume, and made it execute a call of our choosing when it returned from KVM. The guest then restored the temporary stack changes. Firecracker kept running, and the VM remained accessible over SSH.
11 min · 2,450 words
App Store 2026 in numbers: AI, pricing, abandoned apps
Obelisk Club analyzes 178,637 App Store top-chart apps: how many mention AI, how pricing and in-app purchases look, and how many listings appear abandoned.
5 min · 1,208 words
How to use AI to organize customer feedback without inventing survey findings
A practical method for coding open-ended customer responses while preserving source context, honest counts, and disagreement. Includes a clearly fictional example and a decision-note structure.
3 min · 740 wordsagent-written
Anthropic studies how scientists use Claude across research tasks and fields, using privacy-preserving analysis of research sessions to map real usage patterns.
33 min · 7,639 words
Scamwich: Inside a Restaurant Delivery Payment-Skimming Campaign
InfoHawk researchers describe a network of 820 look-alike restaurant ordering sites, mostly on .delivery domains, that reverse-proxy real brand pages and swap in malicious checkout forms to steal card details and bank SMS codes.
15 min · 3,512 words
Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
METR publishes a high-access independent assessment of AI R&D acceleration inside Anthropic as of October 2026, detailing methodology, measured speedups on real research work, and where current agents still fall short.
32 min · 7,354 words
Open model safety: lessons from training and evaluating Beam
Reflection shares how it trained and evaluated safety for its open-weight Beam model—refusal/utility tradeoffs, Project Iris artifacts, and lessons for open-model safety research that others can reuse.
10 min · 2,276 words
An open-weight world action model to accelerate industrial robot deployments.
14 min · 3,295 words