AV (@TheAvtoritet) header
AV (@TheAvtoritet) profile picture
Joined September 2017
Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)Photo by AV (@TheAvtoritet)
Welcome to my library. Feel free to browse around. You can find all my threads in this thread, or via this link: linktr.ee/TheAvtoritet
9
81
12
353
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.
1,821
6,424
2,207
41,250
6,784,531
This skew is older than AI A study cane to mind but there’s multiple related ones apparently • Price's Law: √N people produce half the output. At 400 engineers, that's 20 
• O'Boyle & Aguinis (2012): individual performance follows a power law, not a bell curve 
• McKinsey: the top 25–50 roles drive the bulk of a company's value. AI made it visible on an invoice
Just spoke to the tech leadership at a company with almost 400 engineers. Their head of product asked me what has to change culturally when a company adopts AI. Then they told me where adoption actually is: "Simple answer is mixed. We have some who are heavy adopters burning through tens of thousands of credits a month, and some who haven't adopted it at all." Like most companies, they're seeing the Pareto principle in AI, and it's steeper than 80/20. A small group does most of the work with it, and a big group has barely started. The tools are rolled out across the teams, and the split is still there. So the questions on the table get uncomfortable. Are the heavy users our best people? Is the zero-usage group dead weight? Should the budget follow the usage? Should we cut? I'd be careful answering any of those from a token count, because tokens measure how much somebody uses the tool. A big bill can just as easily be an agent going around in loops on a task it can't finish. Take two engineers. One burns $3,000 of tokens a month and opens 40 pull requests, and reviewers send half of them back. The other burns $300, opens 15, and 14 of them merge with a few comments. By usage, the first one is your power user. I'd take the second one. On our builds we do look at usage, team by team, but it's the first screen, not the answer. What we weigh more: 1. How much of the agent's work the engineer kept. 2. How much of it held up in review. 3. What it cost per ticket. In my experience the split looks pretty different once you measure those, and so do the people decisions you were about to make.
138
AV retweeted
i just built a 24/7 AI video editor inside Claude Code this was supposed to stay internal but f*ck it i'm leaking the whole thing.. you give it a reference video and your product info. it comes back with a finished AI UGC ad ready to post. here's how it works: the Higgsfield video analyzer MCP breaks down any winning ad you feed it. it pulls the structure, pacing, style, and timing from the reference so you're not guessing what made it work. then a custom Claude skill takes that breakdown and generates all the video clips using Gemini Omni. which then it renders each scene, stitches them together, and adds captions all in one go. you're not jumping between tabs or scrubbing timelines or waiting on editors anymore. the whole thing runs inside Claude and outputs a finished 9:16 ad this is faster than any editor could ever do it and costs almost nothing to run.. i'm sharing the full video breakdown, which includes: > the Higgsfield MCP setup for video analysis > the custom Claude skill that handles generation and stitching > the exact workflow for going from reference to finished ad > how to batch this out at scale RT + reply "EDITOR" and i'll send it over (must follow so i can dm)
837
444
15
1,073
101,267
If 90% of humans become economically irrelevant, who buys what capital produces? Stocks in consumer companies, rental properties, the dividend aristocrats, their value depends on a functioning consumer economy. “Owning assets” in a post-consumer world = owning claims on nothing
This is one of the smartest things you will ever read on this site, And it’s why I constantly tell all of you daily to get rich as fuck on here, By the early 2040s we will likely will be in a new structural economic reality, and if you are not rich af by then(depending on your location/class status), then oh boy, I don’t think it’s going to be pleasant. This is the Great Game we’re heading towards, capital is increasingly losing its need for labor and will compound effortlessly by itself. You do not want to be lacking capital when that comes. Normally I wouldn’t even quote tweet alpha like this, because I want as many of you in the dark as possible, but it’s Sunday so let me try to adhere to Christian values and be a good anon neighbor.
1
1
1
405
The people near power DO think in these terms. Whether they’re right or not, capital is being deployed based on these beliefs.
1
1
140
The “neofeudal” framing is partially correct: capital IS increasingly dominant over labor, and this trend will likely accelerate. But the conclusion “buy assets and you’ll be fine” is too simple.
1
1
135
If you just focus on exercising, reading, learning a martial art, picking up an outdoor activity, spending time with loved ones, building a relationship with God, working on something consistently, you will live a vibrant and fulfilling life. Everything will work out. It’s really that simple.
15
160
6
1,519
50,734
At the very least pay only what you must under protest. Do your best to not negotiate with thieves. Engage in legal tax minimization, if you have a stronger stomach, break the law. Every moral and rational person must reject the legitimacy of the state. The moment this happens it will lose power The power it has is based on consent and nothing else. Reject the states authority privately, socially, and politically, whenever and wherever possible. Reduce dependence on the state whenever and wherever possible. You do not overthrow the state, you render it obsolete by living as independently from it as possible. Hoppe talks about this.
7
16
1
148
21,412
The Russian Soul Get a slightly wider peek into the enigma of the Russian soul 👇👇👇
2
1
Heard a German say “we should be grateful to the Soviets” about the Wall coming down. Not taking it for granted, recognizing how easily it could’ve gone differently. How would the world look like with USSR and GDR intact?
133
What the Eastern Bloc actually had: • Soviet oil, gas, rare earths, uranium, timber • East German precision engineering, optics, chemicals • 290M+ people with world-class technical education • Complete resource independence No imports needed for anything critical.
1
84
The fundamental error in most counterfactual analysis: Treating Western baselines as fixed while only varying Eastern outcomes. But this is a systemic counterfactual. Everything shifts. Winners become smaller winners. Losers become competitors.
1
64
The American century might have been much shorter. Or might not have happened at all. One decision in 1989. A different trajectory for everyone. History isn’t inevitable. It’s contingent. And the contingencies compound.
67