151 summarised stories about ChatGPT, each linking back to the original source. Browse all topics →
OpenAI Blog 2 years ago 44
The Financial Times has partnered with OpenAI to make FT journalism available through ChatGPT. ChatGPT users will be able to access Financial Times articles and reporting within the platform. The partnership includes plans to develop new AI-powered experiences for FT readers accessing content through ChatGPT.
The daily briefing
TLDRocket reads 60+ sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.
TLDRocket
4 August 2026
The throughline today cuts across infrastructure, agents, and models—but the real story is how AI's operational layer is maturing faster than its reasoning capacity. Alibaba's Qwen3.8-Max arrived with 2.4 trillion parameters and a million-token context window, but developers immediately questioned whether the promised open weights actually shipped and whether self-benchmarked performance claims held up to scrutiny. Meanwhile, the infrastructure bets are concrete: NVIDIA's NOOA consolidated agent logic into a single Python class, cutting token overhead per task from 2.2M to 1.1M; Liquid AI's LFM2.5-2.6B proved that capable agentic reasoning now runs locally on an M5 Max at 220 tokens per second; Cloudflare's @cloudflare/computer gave agents their own virtual filesystem; YugabyteDB spun up serverless Postgres specifically for agent-spawned database proliferation. Even Runware's modular Sonic Pod—a portable data center unit—treats inference workloads as a distribution problem, not a scaling problem. The message is blunt: enterprises are deploying hundreds of agents today, not someday, and the bottleneck isn't raw model capability anymore. It's operational reliability, cost per inference token, and whether systems fail silently when context rebuilds. Cloudflare's open-sourcing of triagebot-action, which reduced Astro's GitHub backlog from 200 issues to 20 using four-stage AI pipelines, showed what agent infrastructure can do at scale. But the hidden cost—vertical integration—looms: Anthropic's inference margins jumped from 38–40% to over 70% by building application harnesses around its own models, forcing specialized agent labs to either train custom models or accept diminished competitive positioning. Infrastructure wins stay won only if the models powering them remain competitive.
Read the full briefing →