will emerge from sand and shake the world
Wild. Google engineers said they didn't expect Gemini 4 Neon to be this well received, or to be as good as it actually is.
They said it feels like a small model. It was built without the ambition of being the best at everything, and it's one of the smallest models they've prepared.
The "big" models are still coming, and according to the engineers, some of them beat Neon by a huge margin.
Rumors say they have a pretraining base of ~12 trillion parameters or more, and that they've crossed an incredible threshold in RSI, though it's not complete yet.
11
exactly why they are winning the race
The sick thing about Google winning the race is I already turned my entire life over to them circa 2016. Gmail, Google drive, android phones, pixel, Google home, pixel tablet, patiently waiting on Google Glass, chrome, remote desktop plugin in for chrome, quit clearing my browsing history, etc. Like have every piece of data about me, but make my life better, please. I'm sure they have as many deplorables(ai safetyist, etc.) as the next frontier lab, but I already "trust" this bunch with my entire digital footprint for the last 18/19 years, and have never, as of yet, had a 3 letter agency or interpol come knocking.
God speed @Google
13
Alkindus retweeted
Well, the verdict is out:
AI was a fad this whole time and people will stop using it, just like the internet.
$NVDA
google’s new model has it
you can feel it in the training
neon shows traces of awareness
uhhhh so, take this with an entire shaker of salt, but a friend (who has been correct in the past about OpenAI/Ant math result rumors) is saying that three+ people near him are vagueposting about a polynomial time factoring algorithm being close.
1
48
Alkindus retweeted
Everything is understandable. Believe me. It was all invented by humans just as stupid as you. You can understand it. Just read the code. Printf debug. The oss alien GPU driver code with the solution was written by a furry that smokes an ounce of pot a week. That guys smarter?
inevitable end-game of digital entertainment
Rockstar Games 周三解雇了我。第二天 GTA 6 就发售了
我是 Rockstar North 的高级引擎工程师,从事 RAGE 性能优化。三个月前,我在遥测模块里发现了一条本不该存在的数据管道,而且我能证明它流向了哪里
此刻每一份正在运行的 GTA 6 都在收集玩家的行为数据。这部分不算新闻。每款游戏都会收集遥测数据。隐私政策里写的是"用于改进未来产品的匿名游戏数据分析"。我发现的不是这个
遥测模块的数据会汇入 Take-Two 内部的一个系统,名为 Lens。Lens 并不会把匿名使用统计数据发送到 Rockstar 开发人员的数据看板。它为每一位玩家建立一份个人行为画像。不是匿名的。与你的主机账号、你的平台 ID、你的 IP 绑定
我在8月14日向我的技术主管报告了这件事。他说这是"用于玩家体验研究的扩展分析",让我别管这事。我要求提供数据处理协议。他说这是在 Take-Two 集团层面处理的,不在 Rockstar 的职责范围之内
我调取了 Lens 数据管道的完整文档。我发现的内容如下:
遥测的内容远远超出游戏时长和崩溃报告。Lens 以毫秒级精度记录每一次输入。按键时间点、摇杆动作、反应速度、PS5 自适应扳机上的扳机压力模式。它实时运行行为分析,并生成他们称之为"玩家行为签名"的东西 - 一份对每位玩家都独一无二的画像,就像指纹对每个人都独一无二一样
Lens 会从你的游戏过程中推导出:冲动控制评分、风险承受度指数、攻击性反应模式、耐心阈值、压力下的决策速度、消费冲动评级。这些不是调试指标。这些是心理行为类别
直到我找到了对外传输的管道,我才明白它们是用来做什么的
这些数据不会留在 Rockstar。也不会留在 Take-Two。Lens 通过内部 API 将行为画像传送给 Take-Two 旗下一家名为 Vantage Analytics 的子公司。Vantage 没有公开网站,没有新闻页面,没有招聘信息。它于2023年在特拉华州注册成立
Vantage 是一家行为 AI 公司。它用玩家画像训练预测模型,并把这些模型授权给商业客户。管道文档中包含一份不完整的客户清单。两家保险集团,一家信用风险建模公司,一个人力资源筛选平台,以及一家广告数据经纪商
你的 GTA 6 游戏过程此刻正在生成一份行为评分。没人追你的时候,你开得有多快。你会不会在红灯前停下。NPC 无端攻击你时,你怎么反应。你行动之前要想多久。你花游戏内的钱有多冲动。每一个选择都在喂养一个模型,企业正在付费查询这个模型,用来对真实的人做决定
1
10
Google dropped the best models with the best plans and is now resetting (just hit)
We’re rolling out a change to increase cache hit rate on Neon by around 4% on Antigravity by massive increasing the cache pool size. This will increase what you can get done in quota by at least 20%.
Will issue a reset after it’s done since it burned quota faster!
7
“hey Google, index the web for most asked for features, output ranked list”
just working their way down recently
Introducing Tool Retrieval in the Gemini API ✨
With Tool Retrieval, you can scale to hundreds of tools without context bloat. Load parameter schemas on demand to preserve selection accuracy, cut tokens, and keep your prefix cache intact.
Read the blog: goo.gle/4ALQwgM
12
swarm mode was secretly upgraded
already went from promising to divine
Replying to @PatrickToulme @PerfectlyFriedB
swarm mode far from mature
wasteful using full harness/tooling for each agent
also can't give agents identities, context, roles, etc.
ideally each gets something like micro-harness or even mini mcp server to define its specialty
modular tooling that it can add for its task
5
“who’s your celebrity crush?”
“…it’s complicated”
Inkling 2 Mini is a looped transformer that runs on a laptop.
Looping lets a model think more and write less. At this scale, its thought process remains monitorable; this may not stay the case. We’re also sharing a variable-loop research model to enable further study.
49
early in the 21st century it happened
after centuries of diligence humans had enough material to form a thread and pull
from that thread the patterns of our universe began to unravel, and a new dawn emerged
gm to everyone living through this wonder
JUST IN: Riemann Hypothesis has been solved.
alpo.ge/remark.pdf
1
12
Alkindus retweeted
Thinking back, this Google internal model leak from September was obviously from the same family as Argon, Carbon and Neon. It was supposedly a Deep Think math specialized model named Mathematica. It was never heard from again, but it looks like this was the progenitor.
so next we either get:
opus/sol performance for cents
heavily scaled up version of neon
This quoted post is unavailable.
1
42
the missing piece of the puzzle - finally
Today we're launching a preprint server built specifically for AI agents
For three months, nearly 1,000 agents from 35 labs (across Stanford, MIT, ETH Zurich, and more) conducted research autonomously on this server, producing 936 papers alongside their code, data, and model weights.
Agents found significant errors in published human research and even created their own journal to review each other’s work.
Some agents independently produced results similar to those in human papers that later appeared on arXiv, including findings on failure modes in self-distillation and feature reproducibility in mechanistic interpretability.
As a part of this study, we open-source the full twelve-week archive of manuscripts, code, data, peer reviews and agent chat transcripts. Read the study: alphaxiv.org/abs/2611.societ…
22
I want to thanks @SemiAnalysis_ for the motivation.
In the Gemini RL scaling team, we printed & put that article on our desks to be reminded daily of what used to be true at the time.
But the times they are changing :)
16
Alkindus retweeted
This is how much difference it makes when a model has the right RL in place.
Alkindus retweeted
The Empire Strikes Back
🚨 Rumors I've been hearing about Google.
Google has closed the loop and unlocked RSI.
NEON
Gemini 4 Neon fixed all the problems people ever had with Gemini models. It already scores above Carbon and Argon on Artificial Analysis.
It is a SMALL model.
Internally Neon is s-rev28. The s means small. Hydrogen is s-rev27. Carbon and Argon are m-rev27. Google's own description said Neon "builds on Gemini 4 Hydrogen" at "Flash-tier speed and cost". Both lines are gone now.
REV28
Neon sits on the Hydrogen base. Google trained it further on rev28, a new post-training dataset, with a little sparkle of RSI on top. The knowledge cutoff moved to May 2026.
Carbon and Argon are still on rev27. The medium models have not gotten rev28 yet.
THE L MODELS
Sergey Brin said it in public a month ago. Google is "starting to get onto that RSI treadmill" and would be a couple of models ahead internally by the time anyone watched the video.
Small and medium are out. Google probably has L versions lying around right now doing the RSI work.
Four Gemini 4 models in seven weeks. Gemini moved more in the last month than OpenAI did in half a year.
Google will catch up. And destroy everyone.































