ResearchHands-on comparison: GPT‑6 Astra outperforms GPT‑5.6 models on pelican-on-bicycle SVGsThe author tested GPT‑6 Astra by generating SVG images of pelicans riding bicycles at several reasoning levels and compared the results in a grid with GPT‑5.6 models (Sol, Terra, Luna).3 min read
ResearchAI research, business and application frontiers expand as Claude, OpenAI and new physics models take center stageRecent developments highlighted how AI is moving beyond model accuracy to include automated research, business strategy, infrastructure, and new modalities.4 min read
ResearchClaude AI formalized Fermat's Last Theorem in LeanClaude, developed by Anthropic, last month completed the first full Lean formalization of Fermat's Last Theorem, comprising more than 13 million lines of code.1 min read
ResearchData-driven prevention and wearable medtech: Thuróczy Bertalan to speak at Private Health Forum 2026Thuróczy Bertalan, PhD researcher and biomedical engineer, will present at the 10th Private Health Forum (Private Health Forum 2026) on data-driven prevention, wearable sensors and AI in health care.3 min read
ResearchGoogle Research and HHMI Janelia publish first complete wiring map of a male fruit fly brainGoogle Research and HHMI Janelia released the first complete wiring diagram of a male fruit fly’s brain and central nervous system, covering over 166,000 neurons and about 125 million synapses.2 min read
ResearchWhy AI-Generated Menu Images Look Oddly UniformRestaurants increasingly use generative AI to create menu images, but many of these illustrations share a highly polished, unnaturally smooth aesthetic that people find off‑putting.4 min read
ResearchWhen European GWAS Help — and When They Harm: Transferability of PRS to Non‑European PopulationsA systematic study compared polygenic risk score (PRS) performance in Japanese samples from Biobank Japan (BBJ) when using discovery data from the large European UK Biobank (UKB), BBJ, or combinations thereof.5 min read
ResearchHow to credit creators for large language models: Andrew Stellman on attribution and 'hat tips'Andrew Stellman describes receiving a settlement payment after his books were used in training Anthropic’s Claude and argues that authors primarily want visible credit—an attribution or “hat tip”—rather than money.7 min read
ResearchFine-tuning a 350M Model with GRPO Improves Structured-Output ComplianceUsing Group Relative Policy Optimization (GRPO) and TRL, researchers fine-tuned LiquidAI/LFM2.5-350M on ~500 samples for 100 training steps to improve schema compliance measured by the IFStruct benchmark.5 min read
ResearchOpen reproduction of RL language model that paints watercolor-like images by writing JavaScriptOn August 23, Surya Narreddi published a viral video showing watercolor-style images painted by a language model that emits p5.brush JavaScript.15 min read
ResearchGuidelines for Choosing Draft Length and Mechanism in Speculative Decoding for Faster LLM InferenceThis article is the third in a series on AI model co-design and explains how speculative decoding can accelerate autoregressive LLM inference by predicting multiple tokens per iteration.6 min read
ResearchSatellite AI MAPL-EMIT separates and publicizes methane emission sourcesGoogle and NASA Jet Propulsion Laboratory released MAPL-EMIT, a Vision Transformer trained on 3.6 million synthetic methane plumes that improves detection from orbit.2 min read
