Topic
Everything filed under AI, newest first.
RSS · JSON · All topics
Is ‘Attention’ the Essence of Intelligence?
Quanta columnist Natalie Wolchover traces the attention mechanism from Dzmitry Bahdanau's 2014 internship work on machine translation in Montreal through transformers to today's AI boom, and asks why attention turned out to be the key mathematical idea, drawing on researchers' views about what it captures and whether brains do something similar more efficiently.
10 min · 2,402 words
Ben Recht looks at the result in OpenAI's math repository claiming to compute Fourier transforms faster than the FFT, with tutorial material on the FFT and computational complexity, and argues it is less a breakthrough than an exploit of theoretical computer science's habit of proving galactic algorithms that never run on a computer.
8 min · 1,732 words
Recursive Self-Improvement through Multi-Agent Self-Supervision (MASS)
Sakana AI and collaborators present MASS, which repeatedly optimizes multi-agent workflows and distills their collective intelligence into a self-supervision signal for recursive self-improvement, improving score per output token over the base model on research benchmarks.
15 min · 3,501 words
AI Is a Tool. Using It Well Takes Skill.
A response to Rich Hickey's "'AI' is Not a Tool": automation and tool use can coexist, an LLM can be wielded by a human even by Hickey's own definition, and using it well requires skill, feedback, and judgment like any other tool.
2 min · 555 words
Strengthening our defenses against distillation attacks
Anthropic describes new defenses against distillation attacks that try to extract frontier-model capability into cheaper student models, covering detection, mitigation, and why distillation pressure matters for AI security.
9 min · 2,045 words
DeepRSI: evaluating agents on long-horizon tasks derived from real ML research
Hillclimb introduces DeepRSI, a benchmark of 15 open-ended machine learning problems, spanning deep learning training, GPU kernel speedups, estimator improvement and algorithm optimization, that measures how far an AI agent can improve a baseline over 48 hours. It describes the methodology with public and private verifiers, Elo rankings of frontier models, and what the results suggest for recursive self-improvement.
8 min · 1,848 words
How pro-China are Chinese models?
Vals AI evaluates how Chinese AI models handle politically sensitive questions, finding a more complicated picture than simple pro-CCP stereotypes suggest.
17 min · 3,880 words
Deobfuscating Luraph: Lifting Lua to LLVM, Then Letting AI Try
Naci lifts Luraph-protected Lua to LLVM IR, explains why dynamic table operations survive optimization, and tests how far AI-assisted tooling and partial evaluation get at stripping the VM — and where AI just produces plausible-looking output.
17 min · 3,827 words
I posted to HN, and then my website was flooded with AI BOT traffic
Sammy Straus recounts how an HN post for sprinklz.io brought crushing AI-bot and change-monitor traffic, what showed up in NGINX logs, and the robots.txt, blocks, and rate limits that restored the site.
9 min · 2,034 words
Were software engineers blue-collar workers all along?
Peter Bryant argues that much of software engineering has always looked more like skilled trades work than the white-collar myth—and that AI is making that continuity clearer, not inventing a new class of work.
2 min · 531 words
Tobias van Schneider argues that AI-era small teams can finally build specialized software for "unsexy" physical industries long ignored by top engineering talent locked into ads and social products—and that switching costs for legacy enterprise software are starting to fall.
2 min · 448 words
Release notes for Model2Vec 0.10.0: major training-side improvements to the fast static embeddings library, with links to the PyPI release and what changed for practitioners.
4 min · 929 words
The 1st Time Machines Mogged Mathematicians
Pessimists Archive's Louis Anslow revisits the 1976 computer-assisted proof of the Four Color Theorem by Appel and Haken, the skepticism and outright rejection it drew from mathematicians, how acceptance slowly came, and the parallels with today's debate over AI-generated mathematics.
4 min · 979 words
Can Ease of Creation Lead to Decrease in Quality & Value?
Gustavo Pezzi of Pikuma reflects on whether AI tools that make software and games trivially easy to produce will flood the market with cheap, generic work, drawing parallels with past shifts in other crafts and asking what will make people care about what you make.
11 min · 2,431 words
Where is all the new, good software?
Nate Meyvis lists candidate answers to the skeptic’s question—if AI coding is so powerful, why isn’t more great new software obvious—covering quieter quality gains, limited adoption, political bottlenecks, ephemeral and private tools, leisure capture, and AI sometimes reducing software use.
2 min · 511 words
Developer Agency in the Age of AI
David Scott Bernstein argues that AI gives developers leverage but not agency: the real work is understanding problems, choosing tradeoffs, testing assumptions and preserving changeability, and teams must keep asking whether they can still understand, verify and change what they build.
5 min · 1,084 words
Deep dive into Azure Maia 300 architecture
Microsoft details the architecture of Maia 300, the third generation of its Azure Maia AI accelerator, built as a co-designed system of silicon, software, racks, networking and datacenters optimized for cloud-scale agentic LLM inference.
8 min · 1,750 words
Meta AI Research introduces the Muse Wonder series (Muse Wonder and Muse Wonder Full), frontier models aimed at personal and professional agentic tasks that combine Active Memory and Active Focus, available through the Meta Model API and Muse Code.
4 min · 1,030 words
Kevin Kelly's 2016 AI Predictions, 10 Years Later
Taylor Troesh revisits AI-related excerpts from Kevin Kelly's 2016 book The Inevitable a decade on, noting that Kelly anticipated a few AI heavyweights dominating while also arguing for a future of open, ubiquitous, commodity intelligence.
5 min · 1,089 words
"200 OK" doesn't mean your AI agent did the job
Prathap K argues that a success response from an AI agent's tool call only means a request was accepted, not that the work happened, and describes the read-back check he uses to verify each write against the real system (email, CRM, messages).
2 min · 562 words