Khash retweeted
Lots of work we've done to make Antigravity the best harness for complex tasks using Gemini. Neon using AGY CLI is now #1 on AA Coding agent Index and a massive jump over mini-swe on SWE-sweep, a bug finding and fixing eval on large codebases. It's a much larger jump than the native harnesses for the other frontier models.
More improvements to come to the harness in the next week 👀
Khash retweeted
$GOOG really cooked with Gemini Neon.
Back in the coding model lead.
Based on my limited testing so far, it’s also really excellent for finance (and yet blazingly fast).
Khash retweeted
Gemini Neon just beat Fable 5.2 in nearly everything I tried tonight. Is this supposed to be a cheap model? What is going on?
Khash retweeted
iran may be one of the most under-scouted technical talent pools in the world.
35% of university graduates in iran are in science and engineering, according to the latest WIPO data. that puts iran 9th globally on this indicator, based on 2021 data.
the country's performance in international competitions is impressive:
- math: iran ranked 14th out of 117 countries at the 2026 international mathematical olympiad, winning 2 gold, 1 silver and 3 bronze medals.
- physics: all 5 members of iran's 2026 international physics olympiad team won medals: 3 gold and 2 silver.
- programming: Sharif University won bronze at the 2019 ICPC world finals and was the Asia West champion in both 2024 and 2025.
so where would i look for this talent?
1. Sharif University of Technology. probably my first stop for software, AI, robotics and engineering. its competitive-programming record is especially strong, with a bronze medal at the 2019 ICPC world finals and repeated success across West Asia.
2. University of Tehran. one of iran's leading universities, with strong programs in engineering, computer science and scientific research.
3. Amirkabir University of Technology. a highly selective technical university in Tehran with a strong engineering focus.
4. Iran University of Science and Technology. another place worth scouting, particularly for engineering and applied technical fields.
beyond universities, i'd pay close attention to iran's olympiad teams, ICPC contestants, and students who move abroad for master's and PhD programs.
there's clearly a lot of technical talent coming out of iran. the opportunity is to find it before everyone else starts looking.
Khash retweeted
the Persian diaspora is one of the strongest groups of builders in Silicon Valley.
even today, with an ongoing war, sanctions and the regime, Iran still turns out world-class students: three golds and two silvers at IPhO 2026, top 10 at the IMO 11 times since 2000, and Sharif at every ICPC World Finals since 2000.
here are the SV startups founded by Persian expats and Persian-Americans that raised $10M+
Databricks (@alighodsi)
Baseten (@amiruci)
Mirendil (@bneyshabur)
Blend (@NGhamsari)
Dropbox (@arashf)
Guardant Health (@AmiraliTalasaz)
HashiCorp (@armon)
Ricursive Intelligence (@Azaliamirh)
SoundHound AI (@keyvanm)
Capella Space (@payamban)
Tellme Networks (@hadip)
Heap (@matinm1)
Knowde (@aliaminjavaheri)
AngelList (@nivi)
Profluent (@thisismadani)
micro1 (@aliansarinik)
Collective (@hoomanradfar)
Shef (@alvinsalehi_)
Cognichip (@FarajAalaei)
Mutiny (@jalehr)
Zoosk (@shayang, @realalexmehr)
Serve Robotics (@ahkashani)
Saba Software (@BobbyYazdanii)
Scratchpad (@psalehi)
Aeva (@soroushs)
Workera (@kiankatan)
Tahoe Therapeutics (@nalidoust)
Macroscope (@kayvz)
InvenSense (@SteveNasiri)
Tubi (@farhadm)
Gate Labs (@ehsan_saeedi, @DanialEhyaie)
Delphi AI (@daraladje)
Proximal (@navidkp
Trilobio (@royanaieni)
Darmiyan (@Kamalizare)
General Analysis (@HavaeiRez)
these are the Persian founders building in SV. who did I miss?
Khash retweeted
Lost in the news: Prop 42 was positively called Friday
It adds a constitutional ban on wealth taxes in CA!
The irony? After beating the billionaire tax, California is currently protected from a wealth tax! The Legislature can’t impose one
Let’s go 🚀 Sorry doomers
Khash retweeted
I do a lot of document extraction work. It's not the kind of stuff that is eye-popping on social media, but it's often the most valuable application of AI I've seen.
Frontier models (Opus/Fable/Astra) are not the best choice for these tasks; they can do them well, but not meaningfully better than say a Sonnet-level model and are slower and way more expensive.
In my early testing for doc extraction tasks, Gemini 4 Hydrogen is achieving accuracy on par with Sonnet, but at one-third the cost and twice as fast.
Frontier Anthropic and OpenAI models get all the hype but think twice about throwing them at everything.
Meet Gemini 4 Hydrogen: built for rapid coding, multimodal synthesis, and everyday workflows.
It finishes tasks 2.5x faster than 3.8 Flash and surpasses Argon on the @ArtificialAnlys Intelligence Index.
Khash retweeted
🚨🚨JUST IN: TRUMP STOPS JD VANCE FROM SURRENDERING TO IRAN’S MULLAHS!!!
Looks like Trump sees the risk in sending the Pakistani-loving, Israel-hating VP to Iran for a “diplomatic” mission.
Khash retweeted
The alpha from using swarms is ridiculous. We have folks using swarms of 100s of agents to build large systems and products from scratch with a single prompt.
The UI is also very delightful 😀
Announcing /swarm, a new slash command in Google Antigravity for orchestrating specialized agents in parallel across the desktop app and CLI.
Turn a single prompt into a coordinated multi-agent workflow where agents plan, build, and verify together on a live message board, with the flexibility to mix and match different models for each role.
Khash retweeted
What a day! I tested Gemini 4 Carbon on real work. 2 repos, 105 replanted bugs. Find and fix what you can.
It scores just behind Fable 5.2. The results:
- Fable 5.2 (max): 58 for $139.65
- Gemini 4 Carbon (high): 57.3 for $31.32
- Sonnet 5.5 (max): 51.3 for $154.03
- GPT-6 Astra (max): 45 for $33.03
- GPT-6.1 Sol (max): 44.3 for $5.88
- Gemini 4 Argon (high): 40.7 for $7.16
Google's own benchmarks looked super strong, but nobody could access the model. Turns out they didn't lie: +16.6 points over Argon, at less than a quarter of Fable 5.2's cost.
Nobody should be laughing at Google anymore.
(the next effort level drops soon in the thread 🧵)
Khash retweeted
Gemini 4 Carbon is now available in Cursor!
At high effort, it's the top model on CursorBench at 60.4%
Khash retweeted
Four distinct TPU systems in a single generation in one year! Instead of waiting years for a traditional semiconductor cycle, hardware engineers co-designing systems alongside Google Deepmind means we can keep innovating quickly for customers.
Thrilled to share a big milestone from our AI & Infrastructure team at @Google.
Today we’re expanding our 8th generation TPU portfolio with:
• TPU 8i Max: Built for massive context windows and the highest compute density per rack
• TPU 8tx: Purpose-built for secure, confidential inference
This marks the first time we’ve introduced four distinct TPU systems in a single generation, in one year.
For developers, hardware is only as good as the software stack. And we want you to be able to build with TPUs as easily and efficiently as we do at Google. You can now:
↳ Run native PyTorch & Hugging Face models via TorchTPU (zero code rewrites)
↳ Cut KV transfer latency >90% with open-source TPU Sync
↳ Eliminate idle compute waste with Agent Compute Engine
Read more about our newest TPUs on our blog: bit.ly/3VfETPR
I look forward to connecting with builders, researchers, engineers and organizations pushing the boundaries of AI and computing!
Khash retweeted
Thrilled to share a big milestone from our AI & Infrastructure team at @Google.
Today we’re expanding our 8th generation TPU portfolio with:
• TPU 8i Max: Built for massive context windows and the highest compute density per rack
• TPU 8tx: Purpose-built for secure, confidential inference
This marks the first time we’ve introduced four distinct TPU systems in a single generation, in one year.
For developers, hardware is only as good as the software stack. And we want you to be able to build with TPUs as easily and efficiently as we do at Google. You can now:
↳ Run native PyTorch & Hugging Face models via TorchTPU (zero code rewrites)
↳ Cut KV transfer latency >90% with open-source TPU Sync
↳ Eliminate idle compute waste with Agent Compute Engine
Read more about our newest TPUs on our blog: bit.ly/3VfETPR
I look forward to connecting with builders, researchers, engineers and organizations pushing the boundaries of AI and computing!
Khash retweeted
Wer den Unterschied zwischen Persern und Pali-Arabern bis heute nicht kapiert hat, der möge sich die Bilder heute aus Frankfurt im Verglich zu denen von letzter Woche aus Frankfurt ansehen.
🇮🇷Hochkultur vs. 🇵🇸 Islamistischer gesellschaftlicher Bodensatz
Khash retweeted
.@washingtonpost reports a new joint Harvard University and the University of Pennsylvania study finding data centers have "only a small effect on residential electricity prices."
Data centers were responsible for "just 4 percent of the utility rate increases over the last five years."
**Weak Connection Between Data Centers & Overall Electricity Prices**
"Multiple studies by different researchers and organizations have failed to find a strong connection between data centers and the overall trajectory of U.S. electricity prices in recent years."
archive.is/20261106122720/ht…
Alternative title for this clip:
How to Be Manipulative: A Lesson from the Chief.
🚨Coordinated Democratic Social Media Campaign to Win Midterms Revealed🚨
CNN’s Van Jones says Democrat operatives coordinated tens of thousands of people through the “American Accountability Network” to amplify their unaffordability message across social media.
“The Republicans never figured out what we were doing,” Jones said.
He described the participants as “online humans, not bots” and claimed the campaign generated 32.5 million daily hits on Instagram, 88.8 million on X and 185 million on Facebook
"Driving the kitchen table affordability message and tying that to Democrats on a positive and Republicans on a negative.....They couldn't get any traction on these cultural war issues anymore because we figured out how to get around all that."
7
173
































