Results

Google launched TorchTPU, an open source version of PyTorch that will run AI apps on TPUs and Google Cloud without rewriting code. That news followed AWS, which opened its Trainium AI chips to every PyTorch developer without a rewrite via the now open source TorchNeutron. The upshot: Google Cloud and AWS are removing any friction from moving workloads to their custom chips (and presumably away from Nvidia) to optimize for cost.

Google DeepMind published research showing four Gemini 4 Argon agents in a sandboxed environment beat Minecraft in 27 hours and 14 minutes.

"Starting from a randomly generated world on Easy difficulty, the team wrote 2,783 programs and defeated the dragon in 27 hours and 14 minutes, dying 605 times along the way. To our knowledge, this is the
first end-to-end autonomous completion of Minecraft by AI agents, achieved using only the standard Mineflayer library without pre-scouted seeds or other game-specific tools written by humans. Winning required the agents to keep a coherent plan across thousands of dependent steps for more than a day."

Anthropic issued a fascinating report about the workplace where agents outnumber humans by a wide margin. In a nutshell, you're more productive, don't hire entry-level techies and you babysit AI agents all day. Sound rewarding? Didn't think so. For Anthropic, this report on the future of work is in part marketing to get you thinking about scaled AI adoption (the company assumes it's 18 months ahead of everyone else) and possibly get you to buy into its IPO and vision. This report may backfire as an HR recruiting tool. Also see: Anthropic employees and their take on their jobs in the future.

Anthropic job report 1