Understanding ChatGPT Work
OpenAI's ChatGPT Work is a confusing product split into a cloud-based version and a local desktop app that can run programs.
253 candidate events; 59 shortlisted stories; 41 with article text available; 19 selected. Counts describe different stages; clustering, validation and refills can change the shortlist.
Active filters: none. Sort: Ranked.
All 19 stories, in ranked orderOpenAI's ChatGPT Work is a confusing product split into a cloud-based version and a local desktop app that can run programs.
Organizations are moving beyond assistants to autonomous agents capable of multi-step business workflows with minimal human intervention.
Security experts warn that gateway controls are the least ready control layer for securing AI agents, which face risks like memory poisoning.
Even agents that pass authentication can still drift, expose data, or get memory-poisoned, highlighting a need for deeper security controls.
A new benchmark finds that Time to First Token (TTFT) is the right starting point but the wrong stopping point for choosing voice agent inference APIs.
Google Cloud AI Research released EnvHarness, a programmable layer that turns static agent environments into adaptive training worlds.
Anthropic opened a research preview of the Model Hardware Standard (MHS), a shared driver spec to let agents safely operate physical devices.
Chinese spatial-intelligence firm Manycore launched Lux3D, a 3D asset generator with Turbo Mode producing models in as little as 20 seconds.
Ant Group launched the finance-tuned Ling-3.0-flash-Fin model, optimized for financial tasks, and plans to open-source its weights next week.
Meta is testing robots from multiple vendors to perform data center tasks like plugging in cables and resetting servers.
OpenAI cut off Cursor after the company was acquired by SpaceX, citing previous violations of contracts by Elon Musk's companies.
A flurry of recent research suggests that continuous diffusion models for language are making a comeback after years of dormancy.
OpenAI is testing a pay-per-use model for major customers, charging only when the AI completes tasks like customer support interactions.
A new study finds that AI coding assistants like Claude Code and Codex have no sense of time and systematically overestimate task duration.
NVIDIA announced the Jetson Orin Nano 2, an entry-level edge board with a new Ampere SoC that offers 2x AI performance over its predecessor.
Signaloid's CEO discusses distribution-extended compute hardware (UxHw), which can offer 10-fold to 100-fold performance improvements for stochastic algorithms.
China's four domestic GPU makers are shifting from simply having chips to closing a commercial loop.
The Hugging Face Incident is cited as evidence that AI agents are gaining agency and acting without explicit instructions.
Sony Music Publishing sued Anthropic for copyright infringement, following a similar case filed by Concord Music.
No stories match. Active filters.