Every AI story that matters — and the intelligence behind it.
TLDRocket reads all relevant sources, removes duplicate coverage, and publishes a short neutral
summary of every story, linking back to the original. Free, no spam, unsubscribe anytime.
Mistral released Deep Research mode, voice interaction via its Voxtral model, multilingual reasoning through Magistral, a Projects feature for organizing conversations, and integrated image editing in Le Chat. The Deep Research feature generates structured, reference-backed reports by breaking down questions, gathering sources, and synthesizing findings. Users can now conduct research, speak commands, reason through problems in any language, organize work into folders, and edit images—all within a single interface.
OpenAI released ChatGPT agent, a system that can use tools to complete tasks including research, bookings, and creating slideshows based on user instructions. The agent operates by reasoning through problems and selecting appropriate tools to execute actions rather than simply generating text. Users can now direct the system to perform multi-step tasks that require both planning and interaction with external services.
OpenAI released a system card documenting ChatGPT's agentic capabilities, which combine research, browser automation, and code execution tools alongside safety measures. The documentation was developed using OpenAI's Preparedness Framework, which includes risk assessment and mitigation strategies for advanced AI systems. This transparency effort allows researchers and developers to understand both the functionality and safety considerations of agentic AI systems.
Researchers propose FutureBench, a benchmark that evaluates AI agents on predicting future events by drawing questions from news articles and prediction markets like Polymarket, addressing limitations in existing benchmarks that focus on historical knowledge. The benchmark collects approximately 5 questions per week from news scraping and 8 from Polymarket, with prediction time horizons ranging from one week to several months. Testing initial models shows agents outperform base language models, with different architectures like GPT-4, Claude, and DeepSeek-V3 exhibiting distinct information-gathering strategies that vary in search depth and token consumption.
Together AI launched NVIDIA Blackwell GPU support for its inference platform, achieving 334 tokens/second throughput for DeepSeek-R1-0528 models as of July 17, 2025. The company's optimized inference stack combines custom GPU kernels, a proprietary inference engine, speculative decoding, and lossless quantization to deliver a 32 tokens/second speed improvement over baseline deployments and 2.3x to 2.8x faster performance than prior-generation H200 GPUs. Customers can now access this optimized inference capability through Together's serverless endpoints and dedicated endpoints, with the latter offering up to 84 tokens/second additional speedup for production workloads.
OpenAI's board acknowledged the completion of work by an independent nonprofit commission. The commission conducted an extensive review as part of OpenAI's governance oversight. The statement offers no details about findings or any resulting changes to the company's structure.
Invideo AI incorporated OpenAI's GPT-4.1, image generation, and text-to-speech models to automate video creation from text prompts. The platform produces videos in minutes, which the company claims is 10 times faster than traditional production methods. Users can now generate professional-quality videos without hiring editors or learning complex software.
OpenAI is hosting a one-day event called Nonprofit Jam in partnership with the Walton Family Foundation and Emerson Collective to help nonprofits use AI tools. The event will bring together more than 1,000 nonprofit leaders across 10 locations nationwide. Participating organizations will learn how to apply OpenAI's tools to address problems in their communities.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant 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.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
