@Xxi5olc @tcrow_psych @internetope @arcprize We already have coding and math benchmarks. ARC-AGI measures how models handle unfamiliar tasks. It may not matter much if you mainly care about coding, but it still has its own value.
English
Andy Yao
125 posts
Andy Yao
@yaoandy107
Thinking about code or my next bowl of ramen.
Tokyo, Japan Katılım Eylül 2014
321 Takip Edilen71 Takipçiler
Yes I did and will do again. Guess what I am saying is: Even with only 0.5% score, the model already writes better code, does better math and puts together precise financial models. Where does this insistence on cracking ArcAgi come from? Does solving ArcAgi improve intelligence across the board?
English
@gokmakes @MariusMollerH I suddenly started getting "Couldn't read commit" errors on Cloud Build around the same time as well.
English
@MariusMollerH I’m seeing the same behavior on my end. GitHub events are coming through, but Cloud Build triggers aren’t firing at all. Might be an issue on Google Cloud side?
English
Is GitHub, Google Cloud, or me the issue? Because cloudtriggers are not triggering
English
I ran some internal Doist evals for Gemini 3 Flash, and it seems to perform poorly (hallucinations even on simple prompts). These are the same tests & prompts we use for Flash 2.5.
I am getting to a point where it’s very tough to trust any benchmarks.
English
@ardaaltinors @theo @scaling01 Just a heads up, 2.5 Flash actually got an update in September.
To me, the original release was good but not a clear upgrade. It was better in some ways but worse in others compared to 2.0. The updated version definitely works better for my use cases though.
English
@theo @scaling01 you were telling us 2.0 flash is better and to use it. i can't keep up with your pivots
English
You don't understand
Gemini 3 Flash is a bigger deal than Pro
Riley Brown@rileybrown
Their level of confidence is actually concerning at this point. I'm scared for OpenAI and Anthropic.
English
Andy Yao retweetledi
🚨🍌Breaking News: Gemini-2.5-Flash-Image-Preview (“nano-banana”) by @GoogleDeepMind now ranks #1 in Image Edit Arena. In just two weeks:
🟡“nano-banana” has driven over 5 million community votes in the Arena
🟡Record-breaking 2.5M+ votes casted for this model alone
🟡It has achieved the largest Elo score lead in Arena history (a monster 171 point lead)
Huge congrats to @GoogleDeepMind!
Google DeepMind@GoogleDeepMind
Image generation with Gemini just got a bananas upgrade and is the new state-of-the-art image generation and editing model. 🤯 From photorealistic masterpieces to mind-bending fantasy worlds, you can now natively produce, edit and refine visuals with new levels of reasoning, control and creativity. A quick dive into Gemini 2.5 Flash’s capabilities 🧵
English
@gus_tiffer It's better to craft the requirements and design first, then split it into tasks and implement them one by one. One-shotting never performs well for me. Trying to refine a bad version is even worse, I will prefer reset and start the bad piece over again instead.
English
to everyone who says they're one-shotting apps.... genuinely how?
i'm working in claude code every day.
carefully thinking through an app.
how it will work.
crafting a big beautiful prompt.
And still... the end result is barely there.
I spend the next few hours working through the details, debugging, adding/removing, etc.
Do you have some prompting secret I don't know about?
Or does it all just have to be refined in the end?
English
@OfficialLoganK @K78467764 @GoogleDeepMind @joshwoodward I'd love to see automatic memory, and a faster, more concise research mode like Grok or Perplexity, ideally with selectable depth (quick vs. deep).
English
@K78467764 @GoogleDeepMind pls send feedback on what needs to be fixed and @joshwoodward + team will fix
English
This is soo cool! Super amazed 🤯
Google DeepMind@GoogleDeepMind
What if you could not only watch a generated video, but explore it too? 🌐 Genie 3 is our groundbreaking world model that creates interactive, playable environments from a single text prompt. From photorealistic landscapes to fantasy realms, the possibilities are endless. 🧵
English
Andy Yao retweetledi
Struggling with learning Japanese?
Learn with Yori!
Translations with grammar & context explanations.
🎨 Custom styles control
✨ Detailed explanation
💬 24/7 chat buddy with correction
Free to start 👉yoridesu.com
#LearnJapanese #Japanese #JLPT #IndieDev #EdTech
English
Andy Yao retweetledi
"Kotlin is now the recommended programming language for server-side JVM usage at Google, set to replace Java while still providing access to a large existing Java ecosystem."
youtu.be/o14wGByBRAQ
YouTube
English
Andy Yao retweetledi
Andy Yao retweetledi
Jetpack Compose may crash when using early local return🤯
issuetracker.google.com/issues/2096197…
English