Labs & Research
OpenAI, Anthropic, Google DeepMind, Meta AI and publications.
132 published articles
Photo provenance: Apple's Reference Image on iPhone 18 Pro
Apple can sign every pixel its camera captures. Checking it is the hard part
iPhone 18 Pro's Reference Image signs camera pixels and files a second copy for comparison in Photos. Apple calls the result unalterable, but the release never explains how that guarantee survives export, re-compression, or a social upload.
2026-09-23
Apple's audio hardware, September 2026
AirPods 5 bring open-ear ANC to $129. Siri AI's arrival comes with conditions.
AirPods 5 arrive September 18 with open-ear ANC at $129, but Apple's Siri AI bundle ships in beta, skips the EU at launch, and carries daily usage caps.
2026-09-23
Enterprise AI · Sovereign data control
Mistral and Cloudera want regulated firms to own their AI, not rent it
Mistral and Cloudera have partnered to run and train AI models inside customers' own environments, including air-gapped ones. The pitch to regulated industries is about control of data and intelligence, not model performance.
2026-09-22
Artificial intelligence · Olympiad math
Nemotron hit IMO 2026's gold threshold at 30/42, without a formal prover
NVIDIA's Nemotron 3 Ultra checkpoints reached IMO 2026's gold-medal threshold on natural-language proofs alone. The paper releases the data, code and a 200-problem benchmark, but the score is self-reported.
2026-09-21
Explainable AI: energy forecasting research
Explainability Assistant parses 94% of queries, beating TalkToModel's 76.8%
Intent-parsing accuracy in conversational XAI climbs to 94% from TalkToModel's 76.8%, and specialists unanimously preferred the new system to a conventional dashboard. What the evaluation does not measure is whether operators make better decisions.
2026-09-20
On-policy distillation | Folding EOPD and ToDi into one family
A 33-of-36 result, published with the authors' own asterisk
A four-coefficient family for gating per-token losses in on-policy distillation beats single-channel baselines in 33 of 36 cells. The authors then call their own counts directional evidence and their three-seed replications not significant.
2026-09-20
Legacy code modernization with AI agents
Mistral's 40,000-line Fortran rewrite shows where agent autonomy runs out
Mistral moved 40,000 lines of Fortran 77 to C++ for an unnamed energy client and published the workflow. Its first attempt, run by fully autonomous agents, produced Fortran retyped in C++ syntax. Numerical parity checks and a human in the loop fixed that.
2026-09-19
AI funding: Samsung Electronics leads Mistral's Series D
Mistral's record €3B raise tests what sovereign AI really means
Mistral's €3B Series D is Europe's biggest equity raise, the company says. The sovereignty pitch and the three-continent cap table tell two different stories.
2026-09-18
Qwen / Alibaba
Qwen3.8-27B can read 1M tokens, but only if you flip the right flags
Qwen3.8-27B ships a reasoning-effort dial and a 262K native context that stretches to 1M tokens with the right flags. The catch, buried in Alibaba's own guide: dialing the thinking down can slow agent pipelines. Here is what the flags do and when low effort backfires.
2026-09-18
Streaming video anomaly detection research
ReactVAU wakes its heaviest model only when a video stream turns suspicious
ReactVAU splits streaming video anomaly detection into a cheap always-on filter and an expensive reasoner that sleeps until something looks wrong. The paper claims big compute savings but publishes no numbers to back it.
2026-09-18
Sovereign AI · Germany's language models
Aleph Alpha needed 4 trillion German tokens. The open web offers 1.94.
Aleph Alpha's ablation work sets a 4-trillion-token floor for German pre-training. The six largest free German datasets total about 1.94 trillion, and roughly 600 billion of that is machine-translated English.
2026-09-17
Causal ML research, arXiv, 3 September 2026
Probabilistic Causal Impact takes aim at SHAP's blind spot
Actual causality gives principled answers about blame and credit, but only for toy-sized models. A new arXiv paper claims Monte Carlo estimation can close the gap with scalable attribution methods like SHAP.
2026-09-17