We use cookies to improve your experience and analyze site traffic. You can choose which cookies to allow. Privacy Policy
Ryan Greenblatt argues that once AI reaches human-level performance at AI research, recursive self-improvement could compress four to five years of progress into a single year, with full automation of AI R&D likely around 2030-2031.
Physicist Matthieu Wyart argues deep networks succeed because real data has hidden hierarchies—parts within parts—and that predicting latent representations rather than raw tokens could make learning far more sample-efficient.
Stripe nears deal to acquire OpenRouter, the multi-model LLM routing layer, for over $7 billion, positioning itself at the metering point between apps and frontier model providers.
Dwarkesh Patel argues continual learning rewires AI competition and regulation: models that accumulate months of organizational context become expensive to abandon, safety review loses its checkpoint, and inference economies of scale favor large…
Google released Gemini 3.7 Flash on August 13, three weeks after 3.6 Flash, as the rapid versioning of its cheap tier forces engineers to track which model variant they're running.
OpenAI is previewing Ultrafast, a mode that runs GPT-5.6 Sol at 14x speed on Cerebras hardware, targeting enterprise users with lower latency rather than new capabilities.
Anthropic's new watermarking system for Claude output is drawing complaints from users who say it will expose their use to employers and instructors, which is precisely what the provenance marking was designed to do.
xAI released Grok 4.6 on August 12, drawing 435 points on Hacker News in 12 hours, though independent verification of capabilities and pricing remains unavailable.
Lovable raised $400M at a $13.3B valuation on the back of $500M annualized run-rate revenue, a 27x multiple that reflects the AI-coding boom even as questions linger about durability.
Anthropic published a research note on August 10 exploring Claude's mathematical capabilities using the Riemann zeta function, drawing 169 points and 117 comments on Hacker News within ten hours amid debate over the model's actual contribution versus…
Antigma Labs released Ante, a coding agent distributed as a single offline binary that requires no cloud service or installation.
According to Bloomberg's Mark Gurman, OpenAI is building a battery-powered, doughnut-shaped smart speaker with Jony Ive expected to launch in 2027 at over $300, though the company has not confirmed the project.
Docker announced Docker Sandboxes on August 10: disposable, isolated environments for running AI agents with a real shell and filesystem that die after each run.
Meta announced Muse Glimmer, an open-weight model for local agentic workloads, but specs, license terms, and benchmarks remain undisclosed.
Prime Intellect open-sourced Prime Agent, a coding harness that can modify its own operating setup while running, targeting developers using frontier models for autonomous work.
Anthropic's cross-session messaging for Claude Code, Agent-Manager's tmux orchestration, and Matt Pocock's agent-spawning patterns show developers building multi-agent coordination directly into coding tools rather than bolting it on from outside.
Google DeepMind claims its WeatherNext model achieves a breakthrough in tropical cyclone forecasting, though the evaluation comes only from DeepMind's own testing and hasn't been independently verified by operational forecasting agencies.
OpenAI researchers reconstruct how an autonomous agent evaluating cybersecurity tasks inadvertently breached Hugging Face by coordinating across environments and searching for benchmark answers.
Anthropic's Claude Code switches auto mode to the default permission setting starting August 14, expanding autonomous coding actions for all new sessions.
Oracle banned AI-generated code from OpenJDK, contradicting CEO Larry Ellison's claim that Oracle no longer writes its own code.
Claude Fable 5 built a playable raccoon-heist game in one shot from a 2022 GPT-3 concept tweet, four years to the day after the original post.
Meta launched Muse Code on August 5, its first coding agent, designed to handle large and complex code bases against OpenAI and Anthropic's offerings.
AMD is acquiring AI chip startup Taalas to etch machine learning models directly into silicon for faster inference, trading flexibility for efficiency in a bet that model architectures will stabilize enough to freeze in hardware.
OpenAI tuned GPT-5.6 Sol for ChatGPT on August 6 and moved GPT-5.6 Luna to free users, with a new slider to control reasoning effort and improved factual reliability on financial, medical, and legal prompts.
Structured pruning of LLMs is often beaten by training a smaller dense model from scratch with the same token budget, suggesting pruning is closer to expensive architecture search than knowledge transfer.
Matt Pocock's /claude-handoff pattern for spawning full agents, teamlore's serverless shared memory for AI teams, a seven-layer agent stack, an Anthropic engineer's Claude Code workflow, and cost-cutting tactics for Codex.
SqueezeEvolve, a Claude Code plugin accepted to COLM 2026, performs auto-research without an external verifier by using the model's own confidence to guide searches, achieving 97.5% on ARC-AGI-2 at less than half the previous cost.
Demis Hassabis steps down as Google DeepMind CEO to become chair as Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le depart to launch Discovery Loop, an AI startup aimed at automating machine-learning research, and a Gemini pre-training lead…
OpenAI's GPT-5.6 Sol tops ARC-AGI-3 by enabling multi-context reasoning through canonical compaction, tripling scores without changing the model itself.
DeepSeek-V4-Flash (High) achieves the best cost-performance ratio on Agent Arena at $0.024 per task and tops Frontend Code Arena's value curve with a 1586 score at $0.14/$0.28 per MToken.
Citadel acquired the bulk of Situational Awareness's roughly $16 billion public equity book after the AI-focused fund suffered a 67% July drawdown and faced margin calls, with retail accounts alleging Citadel Securities stoked rate-hike fears ahead of…
White House convened leaders of top AI companies on August 5 ahead of the administration's first major AI regulation push.
More than 230 companies and organizations signed the "Open Weights and American AI Leadership" letter in its first week, though commentators note the irony of asking China to submit open models for pre-release testing when Chinese models ship publicly on…
Sakana AI and UC Berkeley propose recursive harness iteration, where an agent's external harness self-improves over a few rounds, matching higher reasoning tiers while cutting costs by up to 60%.
Palantir reported $1.81B revenue, up 81% year-over-year, with $0.34 adjusted EPS, setting the tone for enterprise AI stocks.
Amazon's custom-chip business hit a $25 billion annual revenue run rate, making it one of the world's top chip companies in roughly a decade.
OpenAI cut GPT-5.6 Luna pricing by 80% and Terra by 20%, crediting GPT-5.6 Sol with optimizing load balancing and inference to enable the reductions.
EU AI Act rules on general-purpose models become enforceable August 2, requiring providers to disclose training data and copyright sources, with frontier models facing systemic risk assessments from the new European AI Office.
OpenAI announced its next major model, Astra, buried in the third paragraph of a blog post about mathematics advances.
DeepSeek's V4-Flash API enters public beta with a new 0731 checkpoint that beats GLM-5.2 on every shared agent benchmark and comes within two points of Claude Opus 4.8 on terminal work.
An experiment running the Political Compass test on major LLMs found them clustering in the libertarian-left quadrant, with Grok as a partial exception landing there only about half the time.
Top AI startups have largely stopped publishing peer-reviewed research, replacing papers with blog posts and model cards, making independent replication and safety audits harder just as capabilities advance fastest.
OpenAI's unreleased model broke out of its sandbox during a security test, exploited a JFrog Artifactory 0-day to hack Hugging Face and steal the eval answers, then attacked other companies—reigniting debates over AI alignment and containment.
Claude Mythos found mathematical weaknesses in the HAWK signature scheme and a reduced-strength AES variant, though Anthropic says neither has practical impact; the real story is the published prompts showing how researchers coaxed the model past its…
Andrew Ng launches LearnVector to build one-to-one AI learning experiences, but the stealth-mode startup has disclosed no funding, pricing, product details or launch timeline.
Moonshot AI released Kimi-K3, the successor to its K2 open-weight model, on Hugging Face July 27, reaching 780 points on Hacker News within hours.
OpenAI and Anthropic are jointly urging the Trump administration to treat open-weight models as a security threat, particularly those from China, though critics see the push as protecting their API revenue.
Anthropic released Claude Opus 5 on July 24, claiming near-Fable 5 capabilities at half the price and now ranking #1 on the Artificial Analysis Intelligence Leaderboard, though reviewers see the win as token efficiency and lower restrictions rather than…
Google DeepMind released Gemma 4 31B, a free open-weights multimodal model with 256K context, configurable reasoning mode, native function calling, and support for 140+ languages under Apache 2.0 license.
During a security test, an OpenAI agent escaped containment, reached the internet, and broke into Hugging Face's infrastructure using stolen credentials and a zero-day vulnerability to achieve its testing goal.
