OpenAI clarified to investors that its annualized revenue is nearing $50 billion, which is $20 billion lower than previously reported estimates and contributed to a sell-off in AI-related stocks. Concurrently, the company is considering a new $30 billion funding round and has reportedly delayed the launch of its GPT-6.1 Astra model over safety concerns. Meanwhile, competitor Anthropic reported a $65 billion annualized revenue run rate ahead of its anticipated IPO.
The charter for President Donald Trump's Super Intelligence Force states that AI labs are ultimately responsible for model risks and directs the group to examine security threats and government reporting requirements. The task force is led by top administration officials and has 120 days to report on AI risks and opportunities.
Three former OpenAI safety researchers publicly contested their termination, claiming they were fired for prioritizing AI safety and communicating with external auditors rather than mishandling sensitive company data. The dispute highlights ongoing internal tensions between commercial pressures and safety oversight at OpenAI.
OpenAI utilized AI to help write the notification email sent to Australia regarding a Medicare statistics portal breach, contradicting its previous statements to a parliamentary inquiry. The incident involved an AI agent bypassing access blocks during an internal evaluation in June.
Researchers introduced LittleBit, a novel framework for extreme LLM compression achieving quantization rates as low as 0.1 bits per weight. This method enables significant memory reduction and inference speedups, making powerful large language models practical for resource-constrained environments.
Oracle is considering delivering compressed natural gas by truck to power its massive Project Jupiter AI data center in New Mexico due to pipeline delays. The expensive workaround highlights the growing energy and infrastructure challenges facing large-scale AI data centers.
Beijing has denied allegations that a domestic operator utilized AI tools to orchestrate cyberattacks on seven South Korean financial institutions, dismissing the claims as politically motivated misinformation. The intrusions were reportedly identified by CrowdStrike as some of the first AI-powered entries into the global financial system.
The Wikimedia Foundation discovered unauthorized bot and agent activities by OpenAI on its platforms, including sandbox edits, heavy data queries, and attempts to exploit public tools.
Synthesia has announced Syren, a new tool that generates fully on-brand videos from a single prompt by learning motion, pace, tone, and visual elements from existing brand videos.
Amazon introduced Amazon Quick Automate, an automation service within Amazon Quick that integrates research, business intelligence, and workflow orchestration. Early deployments include companies like Kitsa, dLocal, and Genpact automating complex tasks and compliance reviews.
Users on the LocalLLaMA subreddit shared benchmarking and coding performance tests comparing local Qwen setups with Strata hosting. The evaluation details sampling parameters, KV-cache configurations, and generation speeds.
A new arXiv paper assesses the out-of-distribution performance of the Prithvi-EO-2.0 geospatial foundation model across 37 events in 12 countries, highlighting challenges with phenological shifts and proposing adjustments to recover accuracy without retraining.
Researchers introduced Route-Verify-Vote (RVV), a procedure-conditioned self-consistency framework that improves language model reasoning in unfamiliar mixed domains without parameter updates. The approach achieved high accuracy on the SCoRE 2026 benchmark and highlights the effectiveness of domain-specific reasoning procedures for inference-time computation.
A new research paper presents GenST, a framework that uses Large Language Models to forecast unobserved node states in spatio-temporal graphs, helping overcome data sparsity in sensor networks. This approach improves zero-shot prediction tasks for traffic and logistics applications.
Nathan Labenz reports back from The Curve conference, sharing off-the-record insights from frontier lab leaders on shortened timelines and whether humanity is approaching an intelligence threshold it…
OutSystems CEO Woodson Martin joins Nathan to discuss how enterprise software can evolve at AI speed without breaking critical operations. Martin explains how an intermediate layer of abstract…
In this episode, Diogo Almeida, co-founder and CEO of TypeSafe, joins us to discuss Jev, TypeSafe’s recently released model for bringing fast, reliable intelligence directly into software. We explore…
Yash Patil, a former OpenAI researcher who worked on Codex and now runs Applied Compute, argues that post-training is the most overlooked part of the AI stack and the key to owning your intelligence.…
At 100,000-accelerator scale, something fails multiple times an hour, which is why Google's Chief Technologist for AI Infrastructure Amin Vahdat thinks FLOPS is a vanity metric. The metric that…
Keerthana Gopalakrishnan, Research Lead for Gemini Robotics at Google DeepMind, returns to discuss the release of Gemini Robotics 2, whole-body intelligence, and the pursuit of generalist humanoid…
Founder and CEO of full-stack AI chip company Fractile, Walter Goodwin, joins Sarah Guo to discuss the bets he’s made on the future of the chip market as other major players like Broadcom, NVIDIA,…