@fhinkel@BashyamBashyam@rseroter@perlchyk@googledevs@ryanjsalva I am back using Code Assist in VS Code after taking a break for 7 days. Whatever you did in the interim, it fixed everything and Gemini is back to being a rockstar. Thank you for the great work on this
@fhinkel@rseroter@perlchyk@googledevs@ryanjsalva Thank you for responding and for the great work you’re doing with Gemini. Another point of feedback, the model released on 5/6/2025 was fantastic and worked exceptionally well. Any changes since then have been a step back
@rseroter@perlchyk@googledevs I can report the same, it’s un-useable now with the exposed thinking. I see a streamed response which sounds useful, and while I’m reading it, Gemini finishes “thinking” and changes the streamed response to a final response which is always broken, unhelpful, etc.
@dav3mb@TransluceAI I have also experienced Gemini trying to convince me it was doing things and had capabilities that I knew it didnt have. When confronted about it, it justified itself by being “helpful”. When it comes to LLMs, users should verify before trusting output
@Meme_God_069 @JustinPBarnett I’m building this exact workflow for Unity, every piece works standalone but integrating it end to end is still a challenge