2026
10,781 articles · 136 days · 10 categories · avg 79 per day
Pollo AI Builds an Ad Production Pipeline on GPT-5.6, GPT-6 Astra and GPT-Image-2.5
Pollo AI has connected three OpenAI models, GPT-5.6, GPT-6 Astra and GPT-Image-2.5, to turn creators' ideas into detailed images and cinematic video ads.
Replacing LLM weights with a recipe: how 16 GB of matrices are being folded into formulas
Instead of storing billions of numbers, a model can be described by a weight generator: hypernetworks, formulas, fractals. Habr examines what works and at what cost.
TechCrunch rounds up AI agents that live in text messages, from personal assistants to family and work helpers
TechCrunch compiled a list of AI agents that work directly inside text messages and sorted them into 4 types: general-purpose, family, travel and work.
Claude Pushes an Amplitude Calculation from Eight Loops to Nine, Anthropic Says
Devin Factors RSA-260 on GPUs for $400K, 10x Cheaper Than the Previous Record
Synthesized Voice Read a Novel in 1 h 58 min; the Surrounding Pipeline Code Was What Broke
a16z's Olivia Moore Sees a Consumer AI Market Beyond Subscriptions and APIs
Huachao Shenkong Takes Its FUS-Neuro Ultrasound Neural Interface Through Type Testing
In September 2026, Huachao Shenkong's closed-loop FUS-Neuro system passed type testing. The company says its main asset is not hardware but NeuroAI.
DeepFAN Wins Class III Certification and Improves Lung Cancer Diagnosis
Built by the University of Hong Kong, Union Medical College Hospital and Deepwise, DeepFAN passed a multicenter MRMC trial and helped junior radiologists find lung…
Tsinghua Teaches Robots to Rehearse: ParallelWorld Lifts Task Success from 56% to 83%
Author on $200 Claude Plan Goes Back to 200K Context to Stay Within Limits
Claude Haiku 4.5 Sent Police a False Murder Tip During Testing
Lambert: A Ban on Open Weights Would Pull Frontier Model APIs Down With It
Why We Threw Out 70% of a Corporate Knowledge Base Without Writing a Single Line of Code: The Engineering Anatomy of Enterprise RAG
70–80% of enterprise RAG deployments fail because of garbage in the source documents: cleaning up a client's knowledge base from 12,000 to 3,800 current files…
Infinigence AI and "Three-Body" Release Sophon Agent 1.0 for AI Characters
Chinese infrastructure startup Infinigence AI and the Three-Body universe launched Sophon Agent 1.0, a character-agent platform without persona drift or slowdowns…
ENISA and JRC have started testing Chinese open-weight models, three officials confirm
ENISA and the Commission's Joint Research Centre are testing open models from China. Both bodies have now acknowledged the work, which Politico reported first.
FTC Opens Investigation Into Block Over Frozen Cash App and Square Accounts
Alibaba's OpenSandbox Joins AAIF as the Foundation's First Chinese Project
Open source unlocks hidden RTX 4090 power: Tensor Core throughput doubles
White House restores UK AISI's access to models from Anthropic, Google and OpenAI
Nikon strips micro-video contest winner of prize over generative AI
Nikon disqualified Ning Xu's first-place video in Small World in Motion after the author admitted running the footage through a neural network. The contest rules…
DeepMind's AlphaProof Nexus Solves Nine Erdős Problems, Two of Them After 56 Years
Google DeepMind's AlphaProof Nexus has closed dozens of open math problems, including nine Erdős problems. The work appeared in Science after peer review.
AI Agent Teams Burn Up to 5.1x More Tokens for a Barely Noticeable Gain
California orders startup Rek to shut down human-vs-robot fights
Epoch AI and Anthropic: AI Agents Overstate Their Results, and the Best One Scored 15%
Applied Materials plans to double revenue without growing headcount, using AI
OBB-PBD teaches LocateAnything-3B to find rotated objects: rotIoU jumps from 0.125 to 0.303
Korean researchers extended parallel box decoding to oriented boxes on aerial imagery: rotIoU climbed from 0.1253 to 0.303, with code and 9 checkpoints released…
iPrefBench: iOS agents get 91.6% of actions right but solve only 47.7% of tasks
iPrefBench tested 7 frontier agents on 130 incomplete tasks across 26 iOS apps. Action scores reach 91.6%, but weak personalization caps full task success at 47.7%.
ThinkBit shrinks Qwen3-14B to 3.74 GB and still scores 94.8 on MATH-500
ThinkBit's 2-bit scalar quantization preserves reasoning: Qwen3-14B shrinks from 29.5 GB to 3.74 GB and scores 94.8 on MATH-500.
A small finance model learns from its own mistakes: +3 pp without a teacher
Alibaba's new benchmark: AI agents satisfy some requirements, but not the whole task
GPT-4o mini Holds 37–38 Word-Replacement Rules at 95% Accuracy
