Alireza Fathi (@alirezafathi) header
Alireza Fathi (@alirezafathi) profile picture

Senior Staff Research Scientist / Manager @ Google DeepMind

Mountain View, CA
Joined August 2009
Photo by Alireza Fathi (@alirezafathi)Photo by Alireza Fathi (@alirezafathi)Photo by Alireza Fathi (@alirezafathi)Photo by Alireza Fathi (@alirezafathi)Photo by Alireza Fathi (@alirezafathi)Photo by Alireza Fathi (@alirezafathi)Photo by Alireza Fathi (@alirezafathi)Photo by Alireza Fathi (@alirezafathi)
Alireza Fathi retweeted
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
1,816
6,411
2,189
41,163
6,649,982
Alireza Fathi retweeted
Models by @GoogleDeepMind scale better on cost than their rivals. Gemini 4 Carbon is the prime example. As tasks get more demanding, every rival gets more expensive relative to Carbon. For a typical task, GPT-6 Astra by @OpenAI costs 3.4× as much, Claude Opus 5.5 by @AnthropicAI costs 3.9× as much, and GPT-6.1 Sol is the cheaper option at about half of Carbon’s cost (0.54×). Across the 5% most expensive sessions, Carbon’s position improves against all three. Opus edges up to 4.2×, Astra’s cost roughly doubles relative to Carbon, and Sol’s discount shrinks. This in turn can translate into meaningful savings across real-world tasks such as building an Android security app, one of many long-horizon tasks that our Agent Arena users delegate to their agents.
As tasks get more complex, @GoogleDeepMind and @XiaomiMiMo are the biggest winners in the Agent Arena. On cheaper tasks, @OpenAI and @AnthropicAI hold 10 of the 11 frontier spots. On the most expensive 5% of sessions, they hold just 7. The freed spots go to Google’s Gemini 4 Argon and Xiaomi’s MiMo-V2.6 Flash and Pro. These shifts matter as sessions costing $100 or more are not uncommon across real-world use cases coming from Agent Arena. Even within the last 2 weeks, more than 1,000 sessions crossed that boundary, and choosing a suboptimal model might have had a meaningful impact on that final bill.
35
39
16
633
78,227
Alireza Fathi retweeted
Native Jupyter notebook support is here. .ipynb files now open natively in the file viewer, allowing you to render code cells, Markdown, tables, images, and cell outputs without leaving Antigravity.
135
111
31
1,598
7,083,711
Alireza Fathi retweeted
We just redesigned the Antigravity app to be much more sleek and delightful to use! Hope you all enjoy it. More to come in a couple hours 👀
139
57
11
2,069
114,904
Alireza Fathi retweeted
We want to clarify what we announced today about our education services. Today’s announcements give schools more choice, not less. Here’s what we announced: - Gemini, Gemini Notebook, and Search will be OFF 🔴 by default. This is based on K-12 schools asking for more control over student use of AI services. - Schools are in control: They can choose to turn ON 🟢 one, two, or all of these services individually. - We are also updating Search for schools to make it better for learning. Specifically, AI Mode and AI Overviews will offer step-by-step guided learning through problems, so that students don’t get quick answers to schoolwork. We know there’s concern about students using AI to shortcut schoolwork, and schools have told us they want control. That’s exactly why we are making these changes today to address the feedback we’ve heard from educators in recent months. And we’ll continue to engage deeply with educators to ensure we are giving schools more choice, more controls, and improved Search for learning.
Readers added context they thought people might want to know
Google's post claims schools can turn on Search independently from AI services like Gemini for more choice. Google executives stated Search and AI are deeply integrated, so schools enabling Search must accept AI features with no separate opt-out. blog.google/products-and-p… nytimes.com/2026/11/03/us/… theverge.com/tech/1025411/g… arstechnica.com/ai/2026/11/goo… techcrunch.com/2026/11/03/goo… androidauthority.com/google-to-turn…
476
237
217
2,023
569,935
Top 20 Countries Hosting the Most Refugees (1951–2024)
72
255
39
997
459,918
Alireza Fathi retweeted
Anthropic has more cash than almost every company in America. The nearly 6-year-old artificial intelligence lab disclosed on Monday in an S-1 filing with the Securities and Exchange Commission that it had $112.4 billion in cash, plus another $19.3 billion in short-term investments, as of June 30. Read more: cnb.cx/4f0cFPJ
100
169
43
910
129,331
Alireza Fathi retweeted
XREAL AURA is officially here. 🕶️ Meet lightweight wired spatial computing glasses powered by Android XR, giving you a portable spatial canvas to work, watch, play, and explore. Expand your laptop into a multi-monitor workspace. Transform any room into a personal theater. Enjoy immersive 3D gaming experiences, access to millions of mobile and tablet apps from Google Play and over 150 experiences designed specifically for XR, and get contextual help from Gemini based on what you see. With an expansive 70° field of view, transparent lenses, and a dedicated pocket-sized Compute Puck, AURA brings spatial computing into a lightweight, portable design. For a limited time, enjoy special launch pricing on both standalone configurations: 1️⃣ 12GB RAM + 256GB storage: $1,099 2️⃣ 16GB RAM + 512GB storage: $1,299 Already ordered? Customers who paid the original price before October 28 will also receive the launch pricing, with the applicable price difference refunded. Or choose the setup that fits your needs with our bundles: ⌨️ Productivity Bundle: AURA + keyboard 🎮 Gaming Bundle: AURA + controller 🚀 All-in-One Bundle: AURA + keyboard + controller Available at us.shop.xreal.com/products/x… and Best Buy online. Shipping begins mid-November. #XREAL #AURA #AndroidXR Google, Android XR, Gemini and Google Play are trademarks of Google LLC.
39
115
53
462
62,039
Introducing Gemini 4 Carbon, it's a big jump over Argon which we released just 2 weeks ago. Gemini 4 Carbon is SOTA on cyber, long horizon tasks, ranks #1 on Artificial Analysis V5, significantly improves on usability in domains like coding, and is more aligned.
1,238
1,155
862
15,878
1,308,356
Alireza Fathi retweeted
Replying to @Google
@Google & @Meta launch TorchTPU for native PyTorch support on Google TPUs! 🔥True eager execution & torch.compile backend ⚡Fast inference via vLLM & SGLang 🏋️Near-linear scaling for Llama 3 405B & Qwen 3 MoE via TorchTitan Read more here: bit.ly/4hV19Ho
24
41
20
280
17,009
Google’s new Gemini 4 Argon equals GPT-6 Astra on the Artificial Analysis Intelligence Index at 60% of the Cost per Task with discounted prices. Google is now back to being one of the top three labs in intelligence achieved Gemini 4 Argon is @GoogleDeepMind’s first proprietary model above the Flash class in over 7 months. With high reasoning (the highest available), it scores 53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra (max, 53) and 1 point ahead of GPT-6.1 Sol (max, 52), with gains driven by lower hallucinations and stronger agentic capabilities. At its current 50% pricing discount and with cache discounts increased to 95%, Gemini 4 Argon costs $1.99 per Intelligence Index task, 60% of GPT-6 Astra (max), but 2.7x GPT-6.1 Sol (max). After the discount ends, this will rise to $3.98 (~1.2x GPT-6 Astra (max)). Gemini 4 Argon is currently being rolled out to selected users and is not publicly available. The 50% discount is an initial promotion. Google has not yet confirmed the promotion end date Key benchmarking results for Gemini 4 Argon with high reasoning: ➤ Google returns as one of the top three labs on intelligence: Gemini 4 Argon (high) scores 53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra (max, 53) and 1 point ahead of GPT-6.1 Sol (max, 52). This is 23 points above Google’s previous non-Flash model, Gemini 3.1 Pro Preview (30) and 12 points ahead of Gemini 3.8 Flash (high) ➤ Launch discounts of 50% make Gemini 4 Argon competitive on Cost per Task: At current discounted pricing, Gemini 4 Argon (high) costs $1.99 per Intelligence Index task, 60% of GPT-6 Astra (max, $3.26) for a comparable level of intelligence. This cost efficiency is driven by lower token prices, rather than reduced token use, with Gemini 4 Argon averaging 62k output tokens per task, compared with 27k for GPT-6 Astra (max). Google has not yet confirmed the promotion end date, but on standard pricing, Cost per Task will increase to $3.98 ➤ Stronger agentic performance: Historically a weaker area for Gemini models, Gemini 4 Argon shows improvements across agentic evaluations. It ranks #1 on AutomationBench-AA at 77.5%, 6 points ahead of Claude Sonnet 5.5 (max, 71.3%). On Terminal Bench 4, Gemini 4 Argon achieves 57%, a +53 point improvement from Gemini 3.1 Pro Preview, only behind Claude Sonnet 5.5 (max, 64%), Claude Opus 5.5 (max, 60%) and GPT-6 Astra (59%). On AA-Briefcase, it reaches 1494 Elo. This is driven by a 65% rubric pass rate, the highest we have recorded, but lower Analytical Quality (1576 Elo) and Presentation Quality (1308 Elo) ➤ Lowest hallucination rate among leading models: On AA-Omniscience, Gemini 4 Argon has a 15% hallucination rate, the lowest of any model scoring 45+ on the Intelligence Index, compared with 51% for GPT-6 Astra (max) and 54% for GPT-6.1 Sol (max). This means Argon is much more likely to acknowledge when it does not know an answer rather than guess incorrectly. On accuracy, Gemini 4 Argon scores 50%, a 5 point decrease from Gemini 3.1 Pro Preview, and 13 points below GPT-6 Astra (max, 63%). With this slightly lower accuracy, its overall AA-Omniscience score of 42 remains in line with GPT-6 Astra (43) and GPT-6.1 Sol (42) Key model details: ➤ Context Window: 1M tokens ➤ Multimodality: Text, image, video, and speech input, with text output ➤ Pricing: $4/$20 per 1M input/output tokens at standard pricing, currently discounted 50% to $2/$10. Cached input tokens receive a 95% discount ($0.10 per 1M at discounted pricing), up from 90% on Gemini 3.8 Flash ➤ Long Decode Continuation: We tested Gemini 4 Argon with Long Decode Continuation, a new Gemini API feature that pauses long responses and resumes them across follow-up calls. This lets reasoning run up to 1M output tokens without request timeouts
163
451
230
4,868
626,416
Gemini 4's coding performance is on another level. 🤯 I have been so happy with the performance that I haven't tried any other model for some time now!
Introducing Gemini 4 Argon – our new frontier model. It’s built for complex workflows across coding, enterprise knowledge work, and cybersecurity defense – rolling out today to a set of trusted testers through our Fairwind Program.
2
1
1
48
3,030
Alireza Fathi retweeted
I still pay for ChatGPT AND Gemini but i have been favoring Gemini of late.. My two neighbors at the office are cancelling ChatGPT. I like how much Gemini knows about me and is friendly but not cloying and it is NOT a suck-up
373
54
87
2,141
407,544
Alireza Fathi retweeted
Holy shit. I’ve used ChatGPT every day for 3 years. Just spent 2 hours on Gemini 3. I’m not going back. The leap is insane — reasoning, speed, images, video… everything is sharper and faster. It feels like the world just changed, again. ❤️ 🤖 wsj.com/tech/ai/google-gemin…
789
1,178
451
14,058
3,563,254
Our first release is Gemini 3 Pro, which is rolling out globally starting today. It significantly outperforms 2.5 Pro across the board: 🥇 Tops LMArena and WebDev @arena leaderboards 🧠 PhD-level reasoning on Humanity’s Last Exam 📋 Leads long-horizon planning on Vending-Bench 2
17
103
61
901
274,337
Having some fun around the office this afternoon… enjoy this ship on @GeminiApp before you head out and trick-or-treat! If you tap that pumpkin, we've picked some of our favorite Halloween prompts for Nano Banana :)
32
24
1
313
66,166
✨Our team at Google DeepMind Frontier AI is looking for a strong research intern. We are looking for candidates with a strong publication record at top tier ML and CV conferences.
Our team at Google DeepMind is looking for a research intern (Summer 2026)! Multimodal agentic model, unified model (world model). Looking for candidates with multiple first-author papers in top ML conferences and strong engineering skills. Email: [email protected]
3
20
235
29,656
Veo is getting a major upgrade. 🚀 We’re rolling out Veo 3.1, our updated video generation model, alongside improved creative controls for filmmakers, storytellers, and developers - many of them with audio. 🧵
118
405
169
2,379
754,436
Alireza Fathi retweeted
Have you tried Storybook in the @GeminiApp yet? Whether it's bringing funny group chats to life or creating a special, personalized send-off for a team member, there's so much more than bedtime stories. The possibilities are endless! Available today for Google AI subscribers in the U.S.
61
70
9
432
70,277