AI
Artificial intelligence: LLMs, agents, diffusion, vision, NLP and the latest from top labs.
593 published articles
Inference Endpoints v2
Hugging Face isn't fighting the model war. It's making deployment boring.
Hugging Face revamps its Inference Endpoints service, emphasizing ease of deployment over raw model performance. The platform now supports vLLM, SGLang, llama.cpp, and custom containers, with autoscaling and pay-as-you-go pricing. The pitch: skip the ops work and focus on the model.
2026-07-24
Long-horizon reasoning
The one framework that lets AI agents remember what they did 47 steps ago
PRO-LONG is a minimal framework that helps LLM agents retain and retrieve information across long sequences of actions. On the ARC-AGI-3 benchmark, it improved average pass rates by 18 percentage points across frontier models while using 4.2 to 5.8 times fewer tokens than existing harnesses. With Fable 5, it hit 97.4% best@2 at a total inference cost of $1,750.
2026-07-24
Practitioner guide
LangGraph is not an AI benchmark. It's a process framework that knows when to say no.
A practitioner guide on arXiv argues LangGraph fits specific workflow complexity tiers, not every AI use case. Three executable recipes, SQL analytics with repair loops, agentic RAG with evidence gating, and human-in-the-loop policy review, illustrate where the extra structure pays off and where simpler alternatives suffice. With runnable code and clear carve-outs for overkill scenarios.
2026-07-24
AI & Research
Voxtral TTS's 68% win rate just made ElevenLabs look vulnerable
Voxtral TTS sets a new bar for multilingual voice cloning, winning 68.4% of preference comparisons against ElevenLabs Flash v2.5. The model's hybrid architecture and open-weight release could reshape the TTS landscape.
2026-07-24
AI Research
Four minds, one answer: why AI that thinks differently beat the biggest models at humanity's hardest test
PoTRE breaks inference into four agents working in parallel, then reconciles their answers dynamically. It hit 49.92% on Humanity's Last Exam, beating the previous official best. The approach works with fewer tokens than scaled-up baselines.
2026-07-24
Artificial intelligence
What breaks when AI agents leave the lab: a production reality check
A new tutorial on agentic systems highlights the gap between benchmark success and production reliability. Researchers share concrete patterns for handling failures, from verification pipelines to human-in-the-loop design, drawing on case studies in drug discovery and finance.
2026-07-24
Alibaba's generative world model, in gray testing
One photo, a drivable 3D world: Alibaba's HappyOyster bets on persistence
Alibaba's HappyOyster 1.0 is billed as the first proactive real-time interactive open-world model, letting developers and creators build explorable virtual environments from text or images. The gray test offers SDKs for Android, iOS, and Web, and the real question is whether it can keep the world coherent for more than a minute.
2026-07-24
AI & Neuroscience
An AI model's scrambled neurons just recreated a stroke's exact damage to speech
Researchers perturbed LLaVA 1.6 to simulate aphasic picture-naming errors and matched individual patient profiles in up to seven error categories for 79.5% of cases. Six out of seven error types emerged naturally across different perturbation configurations, suggesting general-purpose multimodal models can function as digital twins for post-stroke language deficits.
2026-07-24
Qwen Cloud
Qwen Cloud just made 40% of your AI costs vanish, and it works with Claude Code
Qwen Cloud's Token Plan bundles access to text, vision, speech, and image generation models under one subscription, promising 40% savings versus pay-as-you-go. It works with Cline, Claude Code, and other tools that support OpenAI and Anthropic protocols.
2026-07-24
Artificial Intelligence
A neural network that learns to like and dislike things, just like you
Researchers built a CNN that learns to associate visual scenes with pleasant or unpleasant valence using a Pavlovian learning paradigm. The model forms associations, generalizes them, and aligns neural representations of conditioned and unconditioned stimuli, just as human brains do.
2026-07-24
Interpretability
Your AI thinks it explained itself. A new test proves it was lying the whole time.
Natural-language autoencoders score explanations by reconstruction, but a new study shows this test is passed even when 98% of specific claims are false. The authors introduce RECAP, a training method that lets probes independently verify designated content, beating adversarial lie detection with an AUC of 0.95 versus chance-level 0.51 for standard methods.
2026-07-24
AI Coding
OpenCode adds Ling 3.0 Flash for free, betting token efficiency wins the AI coding war
OpenCode offers Ling 3.0 Flash for free, a token-efficient model from inclusionAI that could make AI coding cheaper. The addition comes days after the model's release and continues OpenCode's cost-conscious strategy.
2026-07-24