Albert

What is going on in France. How do we allow this in Europe :(

100% agree. Defend our next generation.

Wow forever beta crashes on macos Super hard crash. @Warcraft we understand it's beta but that's some 1 person indie QA stuff :(

True Vanilla Experience :D took me back to 2004

Feels like @GoogleDeepMind has made everyone scared, all of a sudden they want to 'unify'.

What is going on with @OpenAI Codex usage from 70% to like 14% or so in seconds, this is a bug, someone else had it? Check your codex! @thsottiaux you know something about it? :o

A time will come when I’ll tell my grandchildren that we couldn’t create worlds instantly. The games we envisioned required us to play games on rails with pre-defined stories and graphics. Heck, it could even be my own children in five years.

I’m astounded by the speed of @antigravity with @GoogleDeepMind Gemini Flash 3.8. The accuracy is also incredible, and the usage percentage remains remarkably stable even when I throw it the most challenging tasks. Google is currently making significant strides, and most people haven’t noticed due to the overwhelming attention on Astra and Fable. ;)

The start of something strong? :)

Spent today stress-testing @GoogleDeepMind Gemini 3.7 Flash on a real production transcription workload. Same prompt, same config, head to head against 3.5, across two independent random samples. It is the better transcriber. On the audio it completed, it landed closer to the human-corrected reference about twice as often as 3.5 did. But it refuses roughly a quarter of calls with PROHIBITED_CONTENT. Completely ordinary material. I looked hard for a content pattern and there isn't one. The refusal comes back as HTTP 200 with zero candidates. A silent empty result. Very easy to ship a bug you never notice. safetySettings BLOCK_NONE does not recover it. Sending the audio with no text prompt at all still blocks, so it is the audio being flagged, not the prompt. What predicts a refusal is length: around 12% under a minute, rising to every single call past four minutes. The exact same audio on 3.5 and 3.6: zero refusals, every call. It also ignores thinkingBudget: 0 (still burns thinking tokens anyway, ~2.2x the latency) and leaks its chain of thought into the output on about 6% of calls. This is a day-one model and all of it reads like an untuned filter rather than policy. Hoping it settles over the next few days, because on pure quality it deserves to win.

Google is making intelligence more affordance, more usable than ever before. Maybe that is the real benchmark that we will all be looking for in the near future.

JUST IN : Seems @GeminiApp Flash 3.7 is listed and usable time to check it out! Really hoping that it's better than 3.5 Flash and accepts thinking level parameters etc. 3.6 Flash was not amazing tbh.

Well well well it seems the next @GeminiApp model is being tested already in the web interface. Lets go @Google ! I am testing as we speak.

When will we get a proper UI-based app like Codex, Antigravity, or Claude for Grok? I’m amazed by the model’s efficiency and would love to make it our daily driver. Frankly my belief is that in the next 2-3 months grok will skyrocket ! @spacexai @grok

is @GoogleDeepMind @GoogleAIStudio down or smth? says i reached my BA cap and i am not even on 0.0001% of my monthly spent alloance on tier3 lol

I find it ridiculous that Google antigravity is the only AI coding platform that cant successfuly open Google Chrome and debug on Macos, whereas Codex + Claude can. "I attempted to use the /chrome browser as requested, but it is currently failing with a system error (local chrome mode is only supported on Linux). Since we are on a macOS environment, the direct browsing tool is blocked." Not sure if I should laugh or cry for my Google Ultra subscription. @antigravity

We are live lads :) Fable times!

Europe at this point needs to really fund @MistralAI and similar European AI models as intelligence is being treated like a Nuclear Program at the moment from the US.