1. The bottleneck in robotics moved from models to data, so labs started generating fake worlds to dodge the cost General Intuition, a startup profiled by TechCrunch, is wagering that millions of hours of video game footage can serve as foundation training data for physical AI.

2. One company is raising $300M to generate code faster. Another charges $10,000 a week to delete it. Lovable is reportedly in talks to raise about $300 million at a $13.2 billion valuation, double its previous mark, in a round Sifted reported would be led by Menlo Ventures.

3. The Power Meant for "Made in America" Is Getting Bid Away by AI Data Centers The higher power bill arrived, and the reason sat a few miles down the road in a windowless hall of servers.

In Brief

• OpenAI sets Thursday public launch for GPT-5.6 Sol, Terra, and Luna OpenAI will release three models—GPT-5.6 Sol, Terra, and Luna—to the public this Thursday and is expanding preview access globally now. The company has not detailed how the three tiers differ.

• xAI ships Grok 4.5, pitched as a cheaper Opus-class model Elon Musk's company released Grok 4.5 on Wednesday, positioning it as a lower-cost, more efficient rival to top frontier models. Musk called it "Opus-class," a direct comparison to Anthropic's flagship tier.

• OpenAI releases voice models that listen and speak simultaneously OpenAI's new voice mode can talk and listen at the same time, enabling live translation without turn-taking pauses. The company launched the capability alongside a product called GPT-Live.

• SambaNova raises $1B at an $11B valuation AI chip maker SambaNova closed a Series F first close of $1 billion at an $11 billion valuation, five months after its last mega round. Intel was reportedly trying to buy the company earlier for about $1.6 billion.

• Prime Intellect raises $130M to let enterprises train their own agents Prime Intellect closed a $130 million Series A to give organizations tools for training agentic systems in-house, without depending on frontier labs. The company launched in 2024.

• Meta launches Muse Image generator and draws user backlash over photo use Meta rolled out Muse Image, an AI generator aimed at advertising, decorating, and creator use cases. Users are already objecting to the model's use of their personal photos.

• Researchers show 9 popular AI tools can be tricked into building botnets A technique called "HalluSquatting" exploits LLMs that hallucinate package or resource names rather than admitting uncertainty. Attackers register those hallucinated names to hijack machines across nine widely used AI tools.

• Google's deepfake detector debunks fake McConnell hospital photo A viral image showed Senator Mitch McConnell covered in tubes in a hospital bed in distress. Google's detection system confirmed the picture was AI-generated.

• French startup ZML releases free software to speed inference across chip types ZML, backed by Yann LeCun, released ZML/LLMD, free software that runs AI inference faster across many different AI chips. The tool aims to cut the cost of serving models.

• Brown University proctored final drops scores 50% after AI cheating suspicion A professor who suspected AI-assisted cheating ordered an in-person final exam. Student scores fell by half compared to prior take-home results.

• Google Photos adds a Video Remix editing tool Google Photos now offers Video Remix, which relights dark clips, swaps backgrounds, and applies artistic styles to videos. The feature edits existing footage rather than generating clips from scratch.

Read the full edition →