AI — Page 54
Models, agents, infra, applied AI.
- BenchFlow to launch SkillsBench at ACM CAIS, presented by Google DeepMind
The San Francisco afterparty will spotlight 100+ expert curated agent tasks, live demos of the BenchFlow SDK, and Kaggle’s new Agent Benchmarks.
- Anthropic says Project Glasswing surfaced 10,000+ high-severity bugs in a month
Cloudflare alone found 2,000 issues; Anthropic’s OSS scans show a 90.6% triaged true-positive rate and project nearly 3,900 high-or-critical bugs ahead.
- Atomic.chat says Qwen 3.7-max beat Opus 4.7 and GPT-5.5 in agentic Tetris-bot test
In a 10-iteration code-and-rewrite challenge, the team let each model build and train a Tetris bot, then compared the final agents.
- Solo founder Ben Cera says Polsia raise it's own financing of $30M at a $250M valuation, autonomously
Approaching $10M ARR, Cera says Polsia runs companies autonomously with one founder and AI, and even handled its own raise while he "just showed up for signatures."
- Klemen Kotar announces PSI-0.5, a promptable physical world model
Kotar framed PSI-0.5 as contrasting with world models focused on moving around scenes; the reposted announcement we saw did not include links to code, a demo, or a paper.
- Speridlabs exits stealth to build a spatial AI foundation model
The research lab outlined a staged plan for a single 3D-native model and an open-by-default posture; Pear VC and Base10 are backing the effort.
- Qwen 3.7 Max draws head-to-head comparisons with GPT and Gemini-class systems
Benchmark chatter has shifted from open vs weak baselines to parity talk, with hints that Qwen 3.7 Max posts a strong math score.
- Cosmo aims to give AI agents a desktop UI, launched by Shiyuan on X
In an X thread, Shiyuan says Cosmo lets you type or speak from the desktop while the interface renders live; some demo changes are not live yet.
- Runway pushes Aleph 2.0 into an editing workflow with new Edit Studio
Runway is steering its video model toward controllable, preview-first editing, not just one-shot generation, per a thread on X and a new Generative Session page.
- Microsoft starts canceling Claude Code licenses, pushes engineers to GitHub Copilot CLI
Internal memo sets a June 30th cutoff for most Claude Code seats; Microsoft frames the shift as convergence on a first-party CLI and a cost move tied to fiscal year-end, per The Verge.
- Inside 'Google Zero': AI answers move intent off the open web
Founders and indie publishers describe the same analytics signature as Google's summary-style results, including AI Overviews, satisfy queries on the results page and thin out the clicks that fund new work.
- ClickUp CEO Zeb Evans says agent-driven workflows can 100x orgs and 1000x top performers
In an X thread announcing a 22% cut, Evans framed a shift to agents and smaller, faster teams, citing a weeklong frontend architecture rebuild, 40x research, faster code review of agent outputs, and million-dollar bands for outsized impact.
- Anthropic-backed services firm acquires Fractional AI to anchor enterprise push
San Francisco-based Fractional AI, founded by Chris Taylor, Eddie Siegel, and Travis May, will be the operational core of a new Anthropic, Blackstone, and H&F backed firm.
- Minkai Xu introduces Gemini Omni, a self-described world model, in a Google DeepMind X post
Researcher-builder Minkai Xu says he has been heads-down for months on a world model, with the announcement appearing on Google DeepMind's X account.
- OpenAI will watermark ChatGPT images with Google's SynthID and launch a provenance checker
Images from ChatGPT, the OpenAI API, and OpenAI Codex will carry SynthID and C2PA signals, with a checker to verify provenance and new participation in C2PA conformance.
- Lab0 says its AI FDE cuts enterprise rollouts from 6 months to 10 days
Founders say the agent automates discovery-to-go-live, from drafting docs and test plans to configuration and integrations; Lab0 also says it is already working with Adobe.
- Chinese cities are rolling out AI robot barber kiosks, Cointelegraph says
Cointelegraph said in a thread on X that 3D-scanning robot barber kiosks are appearing in Chinese cities for 60 yuan a cut, but did not name the maker or list specific cities.
- Stanford cs329x publishes human-centered LLM playbook
The cs329x report, led by Caleb Ziems and Dora Zhao, maps design, data, tuning, evals, and deployment tradeoffs, and flags engagement-optimization and sycophancy incentives misaligned with goals like user empowerment and mastery.
- ElevenLabs launches Speech Engine to turn chat agents into voice with one prompt
The new pipeline layers onto existing stacks, adds 70+ languages, enterprise compliance, and pricing from 8 cents per minute via ElevenAPI.
- OpenAI says internal model cracks Erdos unit distance problem
The general-purpose reasoning model produced a polynomial-improvement construction; external mathematicians checked the proof and published a companion paper.
- Sam Altman offers every YC startup $2M in OpenAI tokens for equity
At a YC event, Sam Altman said OpenAI would invest across the class with $2 million in API tokens per startup, trading usage for cap table.
- Nick Frosst initially said Cohere's Command A+ was out and Apache 2.0; later corrected us on X
Frosst's first X post called Command A+ Cohere's best model and described it as "out now" and "open source Apache 2.0." After publication, he corrected us on X; availability and licensing remain unconfirmed.
- This founder took revenue to zero after $9M raise to unwind services-led growth
In an X thread, Gushwork founder Nayrhit Bhattacharya said headline metrics masked a services-heavy business; he paused go-to-market for 4-5 months, automated the content pipeline, and relaunched at 90%+ gross margins.
- Parag Agrawal jumps into AI search as Exa Labs hauls in $250M
Parallel Web Systems, led by former Twitter CEO Parag Agrawal, joins a surge of AI search upstarts while Exa Labs raises big at a multibillion valuation, per TechCrunch.