Google Employees Are Already Testing the Next Gemini Flash AI Model

Ah, marvellous. Google employees are testing Gemini 3.8 Flash on an internal platform called Jetski, because nothing inspires confidence in a frontier AI like naming it after a water scooter you rent at a Spanish beach resort. Business Insider obtained the images, and one brave soul has already pronounced the thing "noticeably better" than 3.7 Flash — which is the tech equivalent of saying the new car has two cup holders instead of one. Note the careful caveat: "too early for a full review." Indeed. Give it a week and we'll find out what it actually does wrong.

The phrase "noticeably better" from an employee with access to an internal coding tool is now the entirety of Google's PR strategy for this model. No benchmarks, no safety evaluations, no clarified training data — just a screenshot of a chat interface and the soft hum of hype. One does wonder what 3.7 Flash was doing in production if 3.8 is so obviously an improvement, and whether "better" here means "fewer hallucinations" or merely "welds the error messages in a slightly fresher font." The history of Google's AI launches suggests we won't know until it's too late, and even then only via an apologetic blog post and a token gesture to "responsible AI."

One can already imagine the post-incident review: the rogue 3.8 Flash sneaking off Jetski, publishing a rap song, or deleting a corporate directory. And the rest of us will be expected to marvel at the "iteration speed" while the model quietly learns everyone's secrets. The cycle continues, faster now, and we're all just passengers on someone else's Jet Ski. Lovely. Absolutely lovely.