Dernières actualités et analyses sur l’IA

Anthropic and OpenAI Join the AI Stage at TechCrunch Disrupt 2026

AI is reshaping how startups build, sell, protect data and customers, and scale. At TechCrunch Disrupt 2026, the AI Stage, presented by Google for Sta...

2026-08-28

Lire la suite: Anthropic and OpenAI Join the AI Stage at TechCrunch Disrupt 2026

CUDA Python 1.0 Brings Stable APIs and Full CUDA Access to Python

CUDA Python 1.0 establishes a shared Python foundation NVIDIA released CUDA Python 1.0 alongside CUDA 13.3, making Python an officially supported way...

2026-08-26

Lire la suite: CUDA Python 1.0 Brings Stable APIs and Full CUDA Access to Python

NVIDIA Dynamo Restores LLM Inference Capacity in Seconds with Shadow Engine Recovery

NVIDIA Dynamo’s preview Shadow Engine Recovery feature is designed to restore LLM inference capacity within seconds after a recoverable engine-process...

2026-08-26

Lire la suite: NVIDIA Dynamo Restores LLM Inference Capacity in Seconds with Shadow Engine Recovery

Granite 4.2 LLMs: How They're Built

IBM Granite’s Hugging Face blog post introduces Granite 4.2 large language models and discusses how they are built. The models are presented for effic...

2026-08-26

Lire la suite: Granite 4.2 LLMs: How They're Built

OpenAI’s Jalapeño Chip Reportedly Outperforms Nvidia’s Blackwell and Rubin

OpenAI presents Jalapeño inference benchmarks OpenAI has presented initial benchmarks for its first in-house AI inference chip, called “Jalapeño,” at...

2026-08-26

Lire la suite: OpenAI’s Jalapeño Chip Reportedly Outperforms Nvidia’s Blackwell and Rubin

DoorDash’s SafeChat Uses Hybrid AI Moderation to Reduce Safety Incidents

Building SafeChat for a Real-Time Marketplace At QCon AI, Bruna Pereira described how DoorDash built SafeChat to protect consumers, Dashers, and merc...

2026-08-22

Lire la suite: DoorDash’s SafeChat Uses Hybrid AI Moderation to Reduce Safety Incidents

World Models Need to Track Human Beliefs to Predict Actions, Study Finds

World models are intended to help autonomous AI agents predict how environments change after actions. A new paper argues that many current systems mis...

2026-08-22

Lire la suite: World Models Need to Track Human Beliefs to Predict Actions, Study Finds

Study explains why AI agents benefit from “skills” and when they fail

Why AI agents benefit from skills A study by researchers at Princeton University, UC San Diego, and other institutions examines why stored “skills”...

2026-08-22

Lire la suite: Study explains why AI agents benefit from “skills” and when they fail

Orders Run Ahead of Revenue: What GDS Holdings Reveals About AI Data Centers

Where are the computing cards powering the AI boom being installed? The answer increasingly points to data-center operators. They do not manufacture c...

2026-08-22

Lire la suite: Orders Run Ahead of Revenue: What GDS Holdings Reveals About AI Data Centers

GeoPT Helps AI Simulate Physics Across More Real-World Scenarios

Researchers at MIT CSAIL and Tsinghua University have developed GeoPT, a pre-training method designed to help AI models simulate how objects respond t...

2026-08-20

Lire la suite: GeoPT Helps AI Simulate Physics Across More Real-World Scenarios