Built a little tool to have my #braintrust answer the hard questions for meπ
Can you tell which model came up with the design? #triumvirate
20
Today, Search is getting a massive upgrade with Gemini 4 Hydrogen! Itβs the biggest model improvement to Search yet. And itβs available now for all users, globally.
Why itβs awesome:
- Crazy fast
- Cutting-edge intelligence, with advanced reasoning, more precise instruction following, and better tool calling
- More visual responses, including images and custom tables, charts, and simulations
- More scannable response layout
- Better understanding of intent (including for long multi-part queries with ambiguous typos)
- Broader range of links to the web (improved reasoning -> more query fan-outs)
TL;DR: powerful and fast. Take Hydrogen for a spin in Search and let me know what you think!
Sebastian Mauer πͺπΊ retweeted
Carbon made a pelican demo so you don't have to π
fun test of long-horizon coding and the new /swarm feature in Google @Antigravity.
Builder and critic subagents ran iterative loops across Blender and Three.js - checking their own renders to build a playable 3D game of a pelican biking around our Mountain View campus.
Right, this was bound to happen sooner or later
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! π§΅
Blog: rikiyat.github.io/blog/weighβ¦
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-spβ¦
github.com/RikiyaT/lm_undβ¦
huggingface.co/Qwen/Qwen3.5-0β¦
huggingface.co/RikiyaT/Qwen3.β¦
17
So yes, AI and what it can do (yes LLMs in particular) is amazing (and sometimes terrifying) But are our lives better with it? Is there a Marie Kondo rule that should apply to "new tech"? #controversial
1
39
Sebastian Mauer πͺπΊ retweeted
In lieu of flowers, memorial contributions may be made to:
Johns Hopkins Alzheimerβs Disease Research Center Fund Β Β
alzresearch.org/giving/
Alzheimerβs Foundation of America Caregiver Resources Β Β
alzfdn.org/donate/
Photo Credit: Tommy Garcia
Sebastian Mauer πͺπΊ retweeted
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which weβre pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
Weβre seeing significant improvements in long-horizon tasks even compared to Argon, and weβve strengthened frontier safeguards.
As we did with Argon, weβre going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
Sebastian Mauer πͺπΊ retweeted
Check out the research shared today in @Nature about the foundational science behind the Google Health Coach. Excited that this can help accelerate community research for building responsible, personalized health AI!
How did we build the @googlehealth Coach? Today in @Nature, weβre publishing the research that made the coach possible.
Itβs the most comprehensive evaluation of a personal health agent to date β validated across 11 benchmarks through more than 1,350 hours of review by clinicians, data scientists, health coaches, and users.
Check it out: goo.gle/4sfGBNf
More #ToolsForOne. How about taking care of a #DroppedKerb application for the council. A CAD twin of the house is making all these little planner apps so easy.
18
Not to be #soupy or anything, but my #vibecoded tool helped me find, put in an offer and secure the #property I migh live in for a decade. #AIForOne this one's should be it. @mosworld
1
55
Weβre releasing EmbeddingGemma 2, our first natively multimodal open model engineered for on-device embeddings.
Built on the Gemma 4 architecture and released under an Apache 2.0 license, it goes beyond text to unify images, video, audio, and code in a single embedding space.
TBH, to this date #vibecoding has been the best for me when creating a tool for a userbase of 1. We are house hunting and #estateagents and portals like #rightmove are the worst... scraping it and having AI sift trough all the noise is a godsend.
3
2
93























