žygimantas (@zhygis) header
žygimantas (@zhygis) profile picture

fixer @tempo

ams, nl
Joined October 2014
Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)Photo by žygimantas (@zhygis)
žygimantas retweeted
Today we're releasing the Tempo Facilitator -- the fastest way to start, and scale, services that accept machine payments.
21
15
6
132
26,528
I unfortunately can't take seriously the big AI devrels on Twitter. Popular software factory repos just don't compute for me as cred for "How to use AI". The people I trust the most are people focused on correctness whose bottom is on the line. Kernel/Inference folks, TigerBeetle, our own Tempo team etc. Basically systems folks. I believe that 90% of application layer advice out there is garbage and actively distracting you.
43
48
6
665
32,493
žygimantas retweeted
🇳🇱 Live in the Netherlands? Track the rain without getting tracked + stay dry right from your Lock Screen :) ⛅→ droog.is/lekker Help je me even? Droog is my love letter to the Netherlands. Als je hem gewoon in je groepsapps kan delen, dat zou echt enorm helpen! 🙏☀️
96
38
23
409
57,123
žygimantas retweeted
I'm open-sourcing Dasher, a lightweight headless browser that's faster and cheaper than Lightpanda and Chromium! (1/4)
20
1
64
3,983
žygimantas retweeted
We don't use anything instead of poteto mode. And it works fine without :) I also personally think that invoking skills with slash customs is probably an anti-pattern: - you want the skill to be discoverable enough without a slash command - if it's a highly specific thing that's deterministically invoked, it makes me wonder why one uses the model as a workflow executor vs giving it a high-level goal and figuring it out - codebase should be self-explanatory and legible to the agent without deterministic invocation by human. If human has to tell agent when and how to review with highly specific skills, something probably went wrong somewhere else before
16
7
4
160
25,295
žygimantas retweeted
increasingly think that there are only two successful paths for building software post 2026 1) big orgs with a semi-durable compounding core business which can be tuned and operated by a shrinking team. Improvements can easily be underwritten by a cost curve of token spend. 2) small teams (<5) of people who have overlapping skillsets and trust each other's judgement. Low-trust small orgs or existing orgs which hold onto existing ways of operating will be pulled apart by the entropy of their own slop
5
7
3
205
14,362
žygimantas retweeted
i recreated THE ENTIRE WORLD in minecraft at 1:1 scale. every building on earth, including your house! to do this, i trained a SOTA world-to-minecraft generator on maps data, complete with furnished home interiors! play and check out your neighborhood: worldcraft.gg
1,717
6,459
2,646
123,739
20,803,212
žygimantas retweeted
When you die God plays Clams Casino
1
3
33
2,310
žygimantas 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.
1,823
6,424
2,210
41,251
6,798,099
žygimantas retweeted
Agents assigning tasks and collaborating between machines/harnesses: Claude <> Codex <> Opencode Works behind NAT (private networks) via QUIC Check it out! github.com/peerspan/peerspan
1
5
422
žygimantas retweeted
Swarm riders and Swarm-native companies AI-use even among highly AI-native people is power-law distributed. There seems to be a new class of people — call them “swarm riders” — who are capable of consistently commanding large swarms of agents to productive ends. This is a new kind of managerial skill, analogous to the skill it takes to manage large organizations of people, but wholly different in nature and of course more technical. Based on the swarm riders we know (including many at @Tempo and @Paradigm), these people are capable of productive token spend that is 1000x-10000x higher than median software engineers, and 100x higher than top 5% Claude Code users. The art of swarm riding is different from software engineering and more akin to machine learning, where practitioners grapple with black box systems: defining reward functions and evals, transmuting non-verifiable problems into verifiable problems, running experiments, etc. Swarm-native companies are a new class of company founded by and populated by swarm riders. These companies are organized around swarms in a way that conventional companies will struggle to replicate, as the cost of internal human coordination is sand in the gears of an effective swarm. Effective swarm-native companies will define the right reward functions and evals, remove functional and bureaucratic barriers, and then let the swarms cook. We’ve gotten to see a few swarm-native companies up close — reshaping financial rails @Tempo, building a new gaming platform @Spawn, cyber swarms at Cathedral Systems — and it’s clear these teams wield a significant advantage over pre-AI teams. Some questions I’ve been wondering about: 1) How fast will swarm riding diffuse through the economy? The implications for token spend could be significant. 2) Can swarms be effectively adopted by existing companies or will it generally track the formation of new swarm-native companies? 3) What becomes scarce as swarm capabilities themselves commoditize (through frontier models becoming better at swarming, swarm capabilities built into harnesses, etc)? Cultures and teams that can define reward functions well and otherwise get out of the way?
145
81
115
1,179
284,647
Replying to @matthuang
new copypasta dropped
3
1
70
10,481
Nanocode x Durable Objects! I am gearing up towards an official release of Nanocode, as a frontier harness & SDK for building embedded agents. Durability, code mode & harness/model co-design. Massive thanks to @irvinebroque for helping with this. developers.cloudflare.com/ch…
7
3
2
113
13,775
žygimantas retweeted
VMs and CI providers are modern day dropshippers
13
3
1
163
6,897
žygimantas retweeted
slopsurfing: the art of keeping code at precisely the right level of slop for maximum experimental productivity
5
6
1
180
5,787
žygimantas retweeted
I | Introducing Patron - an experimental pay-per-read monetization model for content providers. Powered by @tempo
9
14
5
93
15,990
žygimantas retweeted
adding her to my tailnet to show that i'm serious about the relationship
37
291
23
3,425
76,593
žygimantas retweeted
the bar was in hell
idk how they did it but it's literally black magic 9 MINUTES to compile on solc 2 SECONDS on solar and bytecode is ~40% smaller, gas is ~15% cheaper holy fucking shit
1
4
65
2,811