Jack Maddox (@MaddoxJack) header
Jack Maddox (@MaddoxJack) profile picture

CPO of @Newsquawk, ex Dataminr / CNN / AlJazeera. Talks indie tech and NUFC.

London, England
Joined April 2010
Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)Photo by Jack Maddox (@MaddoxJack)
Pointed my agent at it: 36 modules rebuilt from source and kernel-checked so far, 27 already accepted. The replay is at 16.9% now, up from 1.11% when this went out. No verdict until the final audit, but grabbing a module is easy.
BREAKING: Riemann Hypothesis proven? I'm running 64 workers in parallel to verify the Lean now. Anyone want to help split up the tasks to do it faster? I've divided up the modules in this Github, and you can point your agent to help speed up verifying which could take hours instead of days: github.com/rohanarun/alpoge-… I generated a video explaining the solution at a high level. 1,936 of 175,187 modules (1.11%) verified so far.
1
35
There is AI slop and then there’s AI slop and then there’s bots responding to AI slop…. What model suggests these names unless it’s prompted to rage bait?? Maybe it is and it worked because I’m quote tweeting which amplifies the most according to the open source x algo…
Here's a pyramid of the greatest England players ever... Someone's slipped a few very interesting call-ups into the bottom two rows, and they've still managed to leave room for a debate about who deserves the top spot. Honestly, if we'd had that lot available, the trophy cabinet would look a lot fuller.
81
Jack Maddox retweeted
Okay, what does it mean that ATC "lost frequencies?" Radio frequencies don't just go away. This is a 100-year-old technology. I don't understand the term the way they're using it.
219
294
35
14,335
884,433
Well someone’s been naughty here.
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.
35
Jack Maddox retweeted
Type of shit your subagents start doing when the task requires 5 lines of code
117
301
54
4,310
161,442
Definitely going to give this a try on a throwaway branch to start.
When Anthropic release a new model it's always worth taking a look at their updated prompting guide - here's an entertaining tip to make Fable 5.2 less likely to be cautious in how it improves existing code: tell it "This codebase is yours" platform.claude.com/docs/en/…
47
Nebraska's prediction market switched favourite three times in four minutes last night. Osborn was ahead on the count, but by 02:48 UK time the market had Ricketts at 88%, and he won.
47
How early can prediction markets call a race? Andrews was at 95.7% in South Carolina at 01:33 UK time, and DDHQ called it about an hour and a half later.
82
Jack Maddox retweeted
Morning all! - Democrats are seeing a very strong midterm result, flipping key Senate races in Ohio, Texas, South Carolina and Kansas, with DDHQ giving them a 99% chance of winning the House and a 100% chance of winning the Senate. - US President Trump said he will accept the election result and is prepared to work with Democrats, while stating that veto power is one of the strongest powers he has. - US equity futures were sold on prospects of a blue wave; APAC stocks were mixed, and European equity futures indicate a lower open. - DXY was pressured and 10yr UST futures were underpinned early before paring much of the move, while spot gold saw a haven bid. - Trump said the US has no immediate plan to strike Iran, although CENTCOM reportedly has a plan to strike as early as Wednesday. - Looking ahead, highlights include Global Services/Composite PMI Final (Oct), EU HICP (Oct), US ADP Employment Change (Oct), Trade Balance (Sep), ISM Services PMI (Oct), Atlanta Fed GDP (Q4), Riksbank Policy Announcement, US Treasury Quarterly Refunding Announcement, NBP Policy Announcement, BCB Policy Announcement, BoJ Minutes, South African Municipal Elections. Speakers include Riksbank's Bunge & ECB's Lane. Supply from Germany. Earnings from Qualcomm & Novo Nordisk.
1
4
7
6,270
What comes first, the headline or the prediction market price action? Me and my team put them next to each other for tonight's midterms. nitter.gq/Newsquawk/status/21172…
Following the US Midterms? Try out the Newsquawk Midterm Dashboard. Every race, every price & every call live for subscribers. Available at app.newsquawk.com/midterms Not a subscriber? Get one week of Newsquawk Pro for free, and then 50% off your first three months with code MIDTERMS26, available until the end of the midterms only.
1
1
70
Is Iowa the closest one tonight? Prediction markets have had Turek and Hinson swap favourite twice, and it's 50.5 now.
52
Jack Maddox retweeted
Programming is dead
1
5
31
3,011
Utterly heartbreaking. A must watch.
1,130
3,925
744
27,086
4,733,258
Jack Maddox retweeted
very weird thing happening rn technical people are sad, they think their skills are now useless non-technical people are desperately wishing they were more technical so they could use agents better
1,553
2,410
683
36,890
1,980,659
Jack Maddox retweeted
seems.... not great for AI datacenter debt
1
4
831
Jack Maddox retweeted
One of my favourite Harry Potter loopholes doesn't even involve magic. It involves exploiting the Muggle economy. Hagrid tells us there are 17 Sickles to a Galleon. Sickles are silver, Galleons are gold. Simple enough. Except gold and silver don't have a fixed exchange rate in the Muggle world. Their market value fluctuates constantly. Before 1900, the gold-to-silver ratio averaged around 16:1, which is pretty close to the wizarding world's 17:1 rate. Throughout much of the twentieth century, however, it averaged around 45:1. See where this is going? Wait until the market heavily favours gold. Melt your Galleons down and sell the gold in the Muggle world. Use the money to buy as much silver as possible, then take that silver to Gringotts and have it converted into Sickles. Finally, exchange those Sickles for Galleons at the fixed 17:1 rate. Congratulations. You've potentially turned a small pile of wizarding gold into a much larger pile of wizarding gold, simply by exploiting the difference between Muggle market prices and Gringotts' fixed exchange rate. No Dark Arts. No Horcruxes. No elaborate schemes. Just precious metals, a fixed exchange rate and a wizard who understands arbitrage. The real question is why nobody in the wizarding world seems to have built an entire financial empire around this.
What makes the least sense in the entire Harry Potter story?
859
799
232
16,404
1,858,319
Good take on AI profitability
Anthropic actually loses far less than I imagined off the $200 subscription. It’s likely that the subsidized plans are actually very sustainable and may never go away. It's widely said that inference is quite profitable. This turns out to be true, and can be computed from first principles by estimating out cost of GPUs, electricity, datacenters, etc. On paper it might cost something like $0.10–0.30 per million input tokens and $0.50–2.50 per million output to serve Opus 5.5, and they sell it for $4 and $20. The problem with coming to the defense of AI economics by saying inference is profitable is that it makes no sense to decouple inference from training. Afaict, no company has yet recouped training costs for a model from profits generated from serving that model. A model today is served for a month or two, maybe less, before the next one replaces it. If only the AI companies could serve these models for a little longer before having to race to train the next one. They would get filthy rich off inference. If only they could all agree to slow down…
2
68
So true.
Codebases are going to start looking like optimized topology 3d prints When AI is better at math than humans it also becomes better at algorithm & codebase design At that point the system and delivering quality software doesn’t require human understanding to exist in the actual code layer. Only needs to be understood by humans at the high level of passing testing, verification, vibe checks and being understandable when queried via AI in the rare case one need to dive deeper.
43
Jack Maddox retweeted
LinkedIn final boss
129
1,532
161
26,949
1,329,403