Mike Pulliam

SFNative#wesleyanU #uchastings #giants #49ers #Dubs passionate about my City. Competence and empathy are powerful.

[Mike Pulliam @mikepulliam]

Dean Preston, Hillary Ronen, Connie Chan and Chesa Boudin all brag about speaking truth to power but the truth is they speak nonsense to reality. We need common sense and competence to make a comeback in SF

[will depue @willdepue]

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 https://t.co/YOtzND0Jqk.

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.

[Mike Pulliam @mikepulliam]

@VigilantFox This guy is an intellectual midwit. A trust fund kid larping as a communist. If communism is better why not move to North Korea or Cuba?

[Kyle Tibbitts @KyleTibbitts]

“Wealth tax for thee, insider trading for me.”

[SF_retro @SF_retrograde]

SF used to put 12% of its general fund into infrastructure and capital improvements

Then it was 8%
Then 4%
Now < 2%

In the same period, our cradle-to-grave “free everything” welfare programs have risen from 2% to > 25% of our budget.

Outcomes 👇🏽
#leftist
#enshittification

[🦔 @adrift_mp3]

Nothing is funnier than someone who claims to be progressive having a wife that can't show anything but her face and hands because islam's god sees women as sexual objects lol

[Alec Stapp @AlecStapp]

Very glad to see the NYT editorial board endorse BAAJA, the bipartisan permitting reform bill.

Key points from the article:

1. "Democrats who call climate change an emergency should not defend a permitting system too slow to build the clean energy that would address it."

2. "[BAAJA] would make it easier to build solar energy farms, battery storage, power lines and more. As a result, it would reduce energy prices. The status quo is expensive, dirty and unstable. The country needs to build a lot of big, new things to make its future cheaper and cleaner."

3. "The bill’s Democratic skeptics are making a mistake. President Trump would almost certainly veto a follow-up bill next year. And time matters, especially for the climate. Congress should pass this bill now. It can then return next year and pass other bills that would do even more good."

4. "Today, it typically takes between 5 and 13 years to build a long-distance power line, like the one that transports clean hydropower from Quebec to New England. That line took about a decade to finish, and opened in January. It has enough capacity to meet about one-fifth of Massachusetts’s electricity demand, and it’s projected to save consumers $50 million a year."

5. "A more robust, cleaner grid would make American energy more reliable and less dependent on hostile foreign nations. No country can embargo the wind."

6. "[BAAJA] would force data center operators to pay the full costs of the electricity they use, protecting the average electricity consumer."

7. "On net, the bill would cut greenhouse gas emissions from power plants by as much as 18 percent by 2040. It would do so in part by nearly tripling the construction of high-voltage transmission lines."

8. "Anyone who cares about climate change should cheer this bill. And a party that just won an election on affordability should not pass up a chance to start delivering immediately."

[David Harsanyi @davidharsanyi]

It was explained to me by my parents who defected from Hungary with nothing.

[AG @AGHamilton29]

China planting a spy to have an affair with the Nvidia CEO and China-tied groups using far-left activists to scare communities into blocking data-center development are not unrelated.

It’s the same story: China is fighting for the future, and it doesn’t plan to fight fairly.

[Kareem Rifai 🌐 @KareemRifai]

Extinction-level community note

[Pol Schiaffino @pabloschiaffino]

El personaje que se mandó Guy Ritchie para la segunda temporada de MobLand es para la historia del cine. Habla como Alex DeLarge de La naranja mecanica, es psicopata como El guasón de Nolan y tiene el Britlook del supremo @liamgallagher . Maravilla.

[Ryan Maue @RyanWeather]

Remember when CNN reported data centers were causing enormous "heat island" effects based upon a (bogus) non-peer reviewed research paper on March 30?

Well, now published research shows the exact opposite: the reflective roofs of the data centers actually dramatically cool the daytime temperatures around the buildings compared to otherwise.

https://t.co/1crmjqIzu8

Study: https://t.co/h6qKfOAlZk

CNN sloppy work:
https://t.co/BylQtleHAw

[GREG ISENBERG @gregisenberg]

openai just let every business claim a profile inside chatgpt. my honest reactions:

1/ this could be the google my business moment for ai, and a lot of agencies made a great living (millions!) in the 2010s just claiming and fixing listings for local businesses

2/ ads can now click through to a conversation with your agent instead of your website (interesting)

3/ every business agent runs on context the company writes, and most brands are going to be bad at it

4/ your product feed is your new seo. boring, and it matters more than your blog now in some ways if you think about it

5/ the dashboard tells you how often chatgpt mentions you but HIDES the prompts and your competitors, so you still can’t see why you’re winning or losing

6/ discovery topics shows what chatgpt already cites you for, which is basically a free content calendar

7/ agents are only open to us ecommerce and online services right now. plumbers, dentists etc are next, and so lots of opportunity to help main street here

8/ businesses can’t see the conversations, so nobody knows which chats turn into sales yet

9/ i’m not sure people want to talk to a brand’s agent. i think they’ll want their own agent to talk to it for them

10/ 1.2 billion people use chatgpt every week. i’d go claim my profile before finishing this post if you can

[Polymarket @Polymarket]

JUST IN: U.S. regulators approve construction of 4 advanced nuclear reactors at a Dow chemical plant in Texas, marking the first such approval for an American industrial site.

[Innovation Council @innovationcncl]

🚨 750 gallons of water flowing every MINUTE, and it's being reused.

CNBC goes inside the next generation of data center cooling technology, where water circulates in a closed loop instead of constantly drawing fresh water from local supplies.

[Mike Pulliam @mikepulliam]

@SFist It just received a $100m donation. And how does a park system incur a 15m deficit?

[Warren Redlich - Chasing Dreams 🇺🇸 @WR4NYGov]

Palmer usually delivers truth bombs By drone but today …

[Autism Capital 🧩 @AutismCapital]

We got Creatine ID before Voter ID.

[Mike Pulliam @mikepulliam]

@KenjiCapital Build it

[Cristina Cordova @cjc]

There’s a long American tradition of industry titans turning their fortunes into public goods. Carnegie built libraries and Rockefeller funded universities and public health. Others built parks, museums, and civic institutions.

Great to see @bchesky carrying that tradition forward in SF.

[J.D. Morris @thejdmorris]

New: Airbnb CEO Brian Chesky is committing $100 million to SF parks in what Mayor Lurie's administration says will be the largest contribution the city has ever received from 1 donor. https://t.co/kZyN7K5GjW