🛜 South Korea plans free nationwide AI access

💔 OpenAI is breaking up with Cursor

🤖 Meta tests robots to run its AI data centers

🔬 Anthropic says AI researchers now beat human ones

🔎 Google expands AI Overviews above search results

+ 📊 Daily poll and results

+ 📈 Trending tools and resources

South Korea's science ministry chose three consortia led by SK Telecom, KT and Kakao to build chatbots and public service agents that every citizen can use free, with no usage limits.

The government gives the groups a combined 512 Nvidia B200 GPUs this year and will cover part of the running costs from 2027. Beta testing starts in September, full launch by December.

Rules require at least half of each service's model usage to run on certified Korean models. So Seoul is buying American chips to lower its dependence on American AI.

OpenAI plans to remove its models from Cursor on November 12, 2026, following the coding assistant’s acquisition by SpaceXAI, though Cursor can choose to end access sooner.

OpenAI cited concerns about Elon Musk’s companies following its terms, pointing to Musk’s admission that xAI used OpenAI outputs to train models through a technique called distillation.

Cursor says OpenAI models serve just 5% of its customers and hopes to reach a deal, while Anthropic has already promised more compute for Claude inside the coding tool.

Meta is testing robots inside the data centers powering its AI, with machines handling tasks like swapping networking cables, rebooting servers, moving equipment racks, and checking hardware for problems.

Meta is testing machines from Watney Robotics, Kinova, and ABB, but humans still supervise them because tangled cables, limited batteries, obstacles, and visual inspections remain difficult.

One worker estimated a reliable cable-swapping robot could handle up to 80% of certain jobs, but that was not Meta’s forecast, and today’s robots still work slower than humans.

Anthropic published a paper Friday showing automated AI systems improved model performance on all 10 alignment benchmarks they were given, without degrading overall performance. Fellow Chen Yueh-Han led the work.

The systems copy how human researchers work. They search literature, propose a method, train the model for 30 minutes, then keep what works and discard what fails across repeated rounds.

Anthropic puts the automated researcher at roughly $4 an hour against $150 for its human ones. The paper notes the approach only works if the benchmarks measure the right things.

Google has started automatically expanding AI Overviews for some searches, placing full AI-generated answers and an “Ask anything” box above traditional links and pushing regular search results farther down the page.

Google says its systems expand summaries when they predict users will find them useful, though the behavior varies by query, browser, and session, making the rollout inconsistent for now.

Google says expanded answers encourage deeper follow-up searches and stop opening once users start scrolling, but publishers may care more about how much farther their links now sit below AI.

A) Yes - 68% 🏆

B) No - 32%

1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals

Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning