Sabitlenmiş Tweet
Neural Bytes
416 posts


Grok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come.
Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release.
This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
English

@MatthewBerman Maybe we should talk! I have a few ideas, but running out of tokens always!
English

@eptwts As a baseline for exploration, current models are so good in generalisation and their ability to connect across domains. And I think the best way to figure out is to actually a lot of exploration and self calibration.
English

@eptwts isn't there a way to create a systemic structure through which we can reduce this hallucinations?
English

the more i talk to LLM's about things i'm already deeply knowledgable about, the more i see their limitations...
a base LLM with no access to good information is actually pretty fkn stupid, and the dangerous part is that it sounds super confident in its stupidity
therefore if you're researching something that you're not familiar in, you're likely gonna get lied to, a lot
this makes me think that LLM's will never kill the info market - in fact, they make good info more valuable, we now have a better way to intake that info & talk to it
English

@ah20im Exactly, it helps to occasionally remind.
x.com/NeuralByte01/s…
Neural Bytes@NeuralByte01
Codex is open source
English

i feel like tibo's the only one in the office who's allowed to do what he wants.
takes breaks whenever, steps out the office without having to tell anyone, orders doordash to his desk, puts his feet up on the desk, big red button on his desk for resets, plays counterstrike in between shitposting on x, gets a special corner, and gets asked weirdly esoteric questions by the rest of the team constantly. 50% of staff are scared to even go anywhere near him. one sweet office lady always bringing him cupcakes.
English
Neural Bytes retweetledi

Introducing performance.dev!
A new space where I explore how the best apps in the world are built.
First piece:
How's Linear is so fast? a technical breakdown.
performance.dev/how-is-linear-…
English
Neural Bytes retweetledi

Excited to see what you are building with Antigravity, We just 3xed the Antigravity limits again, but this time, the weekly quotas. Don't stop building!
Varun Mohan@_mohansolo
Yesterday, we 3x’d limits on Antigravity and are seeing you build so much more. One thing we heard was people are worried about hitting their weekly limits after a couple work sessions. To give you more runway, we’re 3x’ing the weekly Gemini quotas AGAIN on all paid plans. We’ve also gone ahead and reset Gemini quotas on all paid plans. Don’t stop building!
English
Neural Bytes retweetledi

We made a Flash model match Gemini Deep Think!
Two days ago, Google launched Gemini 3.5 Flash.
Today we tested it on LiveCodeBench Pro with the Poetiq Meta-System. It made a new harness that pushed performance up >10%.
Check out our results on LCB Pro: poetiq.ai/posts/recursiv…

English
Neural Bytes retweetledi

Very excited that @GoogleAIStudio is coming to mobile (both Android and iOS) with native apps!
We rebuilt the vibe coding experience to bring it to even more people in a form factor that feels approachable + simple
iOS: apps.apple.com/us/app/google-…
Android: play.google.com/store/apps/det…
English

Total shift in perspective today. Grok’s "real-time truth engine" literally failed to index its own parent company's press release, confidently gaslighting me about a feature drop. Meanwhile, Gemini synthesized the entire cross-platform integration instantly. The ability to attach live tabs and code seamlessly is total leverage. 🛠️⚡️
English

Its about the avilability of Grok Build 0.1 for supergrok users (not supergrok heavy) through opencode platform. x.ai/news/grok-open…
English




