Qwen's Gated Attention (NeurIPS 2025 Best Paper) puts a per-head sigmoid gate on SDPA output. First-token attention drops 46.7%→4.8%, max activation 1053→94. Why it works and how Qwen3-Next uses it.
IEEE S&P 2026 study of 2.7M arXiv submissions: 265 API tokens, 7,326 GPS-tagged papers, 699 editable Google Docs, and why withdrawn versions stay online.
At Itaú, a staff engineer delivered a 4-person, 18-week project in 9 weeks with 4 AI agents — but it worked only because they knew the codebase deeply. What the case study really says about AI and team size.
Sakai Lab fMRI study (N=25, U-Tokyo): reading a story's first half on a tablet stretched response time on integration questions and skipped the language-area savings paper produced. Full stats (F/p/q/r), plus the MangaFlow manga-generation AI as the drawing-side counterpart.
A paper claims that a single binary operator eml(x, y) = exp(x) - ln(y) combined with the constant 1 can express all elementary functions — arithmetic, trig, logarithms, even pi. I read the paper and tested it in 5 languages.
Meta AI's HyperAgents performs metacognitive self-correction that optimizes improvement strategies themselves. Self-improvement appears in four non-coding domains, and strategies learned in one domain transfer to another, along with spontaneously acquired persistent memory.
A paper explains that two seemingly mysterious Transformer behaviors, heavy attention on specific tokens and unusually large activations in specific dimensions, are actually manifestations of the same mechanism.
How should memory be allocated in reasoning models? This paper explains the trade-offs among quantization, KV cache, and test-time compute, based on 1,700 experiments.
I investigated the source behind the viral claim that a Johns Hopkins study found ChatGPT lies 27% of the time, and it turns out multiple different studies have been mixed together.
Implemented all 12 text processing tools planned in the previous article. Also reorganized the category system and switched the listing UI to a table layout.
I compiled research findings and implementation specifications for adding text-processing tools to lilting.ch/lab. Based on a comparison with DevToys, it highlights the gaps and documents detailed specs for 12 candidate tools.
