AI
Artificial intelligence: LLMs, agents, diffusion, vision, NLP and the latest from top labs.
593 published articles
Multi-agent orchestration
Nemotron meets Fugu: Sakana AI's bet that open models win as a swarm, not alone
Sakana AI integrates NVIDIA's Nemotron open model family into the Fugu multi-agent orchestration system. Fugu dynamically selects and combines specialized models. The collaboration aims to show that coordinated open models can match monolithic frontier systems.
2026-07-23
LiveBench leaderboard: cost-performance divergence at the top
The LiveBench top four are separated by 2.2 points. The cost difference is brutal.
GPT-5.6 Sol takes the overall crown on LiveBench with an 82.4 average, but Claude Fable 5 trails by just 1.6 points at nearly three times the cost. The real story is how the pack below has thinned out, and where the dollar smarts stop.
2026-07-22
product update
Copilot just learned what good code looks like, not just valid code
GitHub turned a 40,000-star open-source project into a native Copilot skill. The integration targets design quality in AI-generated code, aiming to raise the ceiling while the floor is already covered.
2026-07-22
Artificial Intelligence
The $314 billion assumption that just broke
Moonshot AI's Kimi K3, a 2.8 trillion parameter open-weight model, triggered a global revaluation of AI stocks. The combined estimated valuations of OpenAI and Anthropic lost $314 billion as investors confronted cheaper Chinese frontier AI. The analysis examines winners (Chinese chip makers, memory manufacturers, software developers) and losers (model developers, Nvidia's scarcity premium) in the new landscape.
2026-07-22
AI Research
MiniMax's VTP learned to understand before it learned to generate
MiniMax's VTP framework rethinks how visual tokenizers are pre-trained, replacing pure reconstruction objectives with a mix that rewards semantic understanding. The result: a tokenizer that actually scales with compute, and a 65.8% FID improvement on downstream generation.
2026-07-22
Mistral's Vibe unifies work and coding in one agent
Mistral kills Le Chat, bets it all on one agent for inbox and pull requests
Mistral rebrands Le Chat as Vibe, unifying its AI agent for work and coding tasks. The platform offers Work Mode for enterprise tools and Code Mode for development, with integrations for VS Code, CLI, and API access, priced from free to enterprise plans.
2026-07-22
Analyse comparative
14 microsecondes d'instanciation: PraisonAI peut-il vraiment détrôner LangChain?
PraisonAI promet 14μs d'instanciation et une simplicité redoutable face à LangChain et CrewAI. Mais sans cas d'usage en production, le pari reste à prouver. Analyse des forces et angles morts du framework.
2026-07-22
Consciousness Debate Heats Up
Will AI Become Sentient? A New Framework Proposes a Test for Machine Consciousness
A multidisciplinary team has designed a framework to test AI sentience using integrated information theory, potentially transforming our understanding of machine consciousness and sparking debates on AI ethics and rights.
2026-07-22
Artificial Intelligence
StructAgent lifts AI agent success rates from 27% to 79% without bigger models
An academic paper introduces StructAgent, a state-centered framework that restructures how digital agents track task progress. It achieves state-of-the-art results on OSWorld-Verified with open models, and generalizes to Minecraft. The work identifies that raw interaction history is a bottleneck for long-horizon tasks, and proposes a structured state plus verifier-backed workflow as the fix.
2026-07-22
Biologically Plausible Learning
Sakana AI's MNIST-to-CIFAR flip exposes a blind spot in brain-like AI
Sakana AI scaled a backprop-free, Dale-compliant architecture past MNIST for the first time. The catch: which trick matters most reverses between datasets, a warning about how biologically plausible AI is measured.
2026-07-22
CREATIVE TOOLS
Google Vids just turned your selfie into a full production studio
Google Vids adds Gemini Omni for text-to-video generation and personal avatars that let you star in videos without a camera. Every clip gets an invisible SynthID watermark. The tools target subscribers and business customers, raising questions about adoption and trust.
2026-07-22
Deep Dive: AI Agents
Parallel agents aren't about speed. They fail when your architecture does.
Parallel agents promise speed and specialization, but concurrency alone doesn't guarantee better results. This analysis explores the architecture, patterns, and practical orchestration that make parallel agents effective, using Kimi Agent Swarm as a case study.
2026-07-22