AI, Agents, Web3, Fintech - builder. All posts are personal opinions. CTO@ Chateau Capital
Earth
Joined October 2014
- Tweets790
- Following1.9K
- Followers450
- Likes3.3K
Dmitry Chirun retweeted
I LAUNCHED A GROK BOT TERMINAL OVERNIGHT, GAVE IT $44.82 AND STOPPED ASKING AI WHAT IT WOULD DO. BY MORNING IT HAD $11,643.21 AND SIX AGENTS LOGGING EVERY TRADE
I spent a year asking agents what they'd trade and watching the chat scroll. Last weekend I built the thing that actually trades, in Grok Bot, with no code I touched by hand.
The six agents share one screen, each one job:
> scanner reads new token pairs and volume spikes
> rug check filters suspicious tokens before anything executes
> sizer sets position from account equity, not a fixed ticket
> executor opens and closes with slippage tracking
> logger writes the reason behind every trade
> equity curve updates on every close
I went to bed on $44.82 and left the laptop open on the dashboard. By morning the balance read $11,643.21. Every trade on the activity feed had a reason next to it and a line in the log.
That's the shift I didn't see coming. Not that an agent made money overnight. That I stopped being the one asking "would you long this?" and started being the one reading the log of what already happened.
Starting balance: $44.82
Current balance: $11,643.21
Grok Bot built the terminal, the agent architecture and the dashboard. I described what I wanted and the thing went live.
The full build is in the article below ↓
The whole build is in the post below ↓
1. Comment " Grok_Bot "
2. Like and Repost
3. Follow me @alexhalescoder
I'll DM you the complete setup.
Dmitry Chirun retweeted
these 2 github repos are the most BROKEN combo in trading right now
Citadel pays quant researchers $336K to $642K a year to forecast markets and stress-test trades. These 2 repos put both steps inside Claude Code for free.
They're called Kronos and Day Trader. Kronos was built by researchers at Tsinghua University, accepted at AAAI 2026, and it already has 45K+ stars. Day Trader is 11 Claude skills built by Jake Schincariol, and it already has 4K+ stars.
First, get how it works:
1. Kronos gives you 1,000 futures instead of one guess. It reads candlesticks the way ChatGPT reads words, it learned from 12 billion candles across 45 exchanges, and it can map out the next 48 hours of any chart.
2. Day Trader turns that fan into a trade. One skill sets your stop and target before you enter, one sizes the position to your max loss, one blocks anything over your limits, and its Time Machine replays the setup 10,000 times before you risk a dollar.
3. Together they answer the question that matters: how many of those futures hit my target before my stop? On the real BTC candles in the video, 725 of 1,000 futures end down, so the plan is a paper short with a 1% stop and a 2% target. 203 hit the target first, 302 hit the stop first, 495 touch neither. Nothing gets placed until you type yes.
Here's how you can run it for free yourself:
1. Open Claude Code in an empty folder.
2. Install Day Trader. Paste: "github.com/Jakeschincariol/d… install skill". No signup, no API key.
3. Install Kronos. Paste: "Set up github.com/shiyu-coder/Krono… and load Kronos-small from Hugging Face."
4. Run /trader-scanner and /trader-news to pick a ticker.
5. Ask Claude Code: "Sample 1,000 Kronos futures for this ticker from its last 128 candles, 48 candles ahead. Keep every path."
6. Check it before you trust it: "Re-run Kronos on each of the last 5 days and show me how often the real price landed inside its p10-p90 range."
7. Run /trader-exit to set your stop and target, then ask: "Race every Kronos path to my stop and target. How many hit each one first?"
8. Run /trader-timemachine for a second opinion: 10,000 replays from the stock's own candles, plus every past time the setup fired. If it disagrees with Kronos, skip the trade.
9. Run /trader-budget and /trader-risk, then /trader-broker. It shows you the ticket and sends it to your paper account only after you type yes.
i won't say anything more:
Dmitry Chirun retweeted
Build a fully local AI coding agent on Ubuntu using Canonical Workshop and the Qwen3-Coder Inference Snap.
In this step-by-step demo, you'll set up an isolated development environment, connect your coding agent to a locally hosted LLM, and run coding tasks without an internet connection.
Follow along to recreate the setup on your own machine, using CPU or GPU inference: youtu.be/u5Q149jlMQ8
#Ubuntu #Linux #AI #Agents #Workshop #OpenSource
Dmitry Chirun retweeted
Anthropic cracked open the ribosome and found it stalls on purpose and those pauses are doing a job nobody f*cking gave them
for decades, scientists treated ribosome stalls like a problem.
the machine builds proteins. faster should mean better, right?
turns out, speeding it up might be exactly how you break it.
here’s where things get weird.
researchers used Claude +JEV and started looking at what happens during those tiny pauses instead of just measuring how fast a protein gets made.
> TIME - follow protein production second by second, not just the final result
> MARK - record exactly where the ribosome stops along the sequence
> SORT - separate random slowdowns from pauses that keep appearing in the same places
> FOLD - check what the growing protein is doing during each pause
and that’s when the pattern appeared.
the ribosome wasn’t just randomly getting stuck.
the same pauses kept showing up at the same points, across different cells.
and those points weren’t random either.
they appeared where a newly built section of protein needed time to fold into the right shape.
think about that.
the ribosome isn’t just assembling a chain of amino acids.
it’s giving that chain time to become something that actually works.
remove the pauses and the protein can still come out with every amino acid in the correct order.
except now it folds wrong.
same ingredients. same sequence. completely different result.
and here’s the uncomfortable part.
scientists have spent decades optimizing genes to make proteins faster.
but speeding up the sequence can also erase the pauses that help proteins fold.
we may have been optimizing away part of the instructions.
the genetic code doesn’t just tell the cell what to build.
its timing can also influence how that thing takes shape.
have a look and find some new use cases JEV can do in addition to that
Dmitry Chirun retweeted
50 places to launch your startup:
1. news.ycombinator.com
2. reddit.com
3. producthunt.com
4. news.hada.io
5. whatlaunched.today
6. indiehackers.com
7. betalist.com
8. uneed.best
9. peerlist.io
10. devhunt.org
11. launchingnext.com
12. alternativeto.net
13. saashub.com
14. g2.com
15. capterra.com
16. getapp.com
17. trustradius.com
18. slashdot.org
19. tools.launchllama.co
20. futurepedia.io
21. theresanaiforthat.com
22. toolify.ai
23. futuretools.io
24. confettisaas.com
25. crazyaboutai.com
26. rankanything.online
27. crunchbase.com
28. auraplusplus.com
29. startupbase.io
30. sideprojectors.com
31. microlaunch.net
32. betterlaunch.co
33. foundrlist.com
34. peerpush.net
35. twelve.tools
36. shipybara.com
37. hashnode.com
38. saascity.io
39. 500.tools
40. sourceforge.net
41. aidirectori.es
42. startupfa.me
43. github.com
44. pitchwall.co
45. scrolllaunch.com
46. rankinpublic.xyz
47. tinylaunch.com
48. tinystartups.com
49. nicklaunches.com
50. wellfound.com
Dmitry Chirun retweeted
Ultimately, precompiles and EVM are both the wrong place to put most cryptography support into Ethereum. The reason is that precompiles and EVM both operate on data that has to be included into the ethereum chain somehow, which realistically means included in a block as calldata. That space is limited - even with PeerDAS and blocks-in-blobs, there's a practical cap of ~8 MB/s before it starts getting uncomfortable.
Remember: it's not a blockchain. It's a cryptographic world computer. Any data/computation that does not need to be in the "main thread" of data/execution should not be put into the main thread.
Instead, the correct solution is... EIP-8288 maximalism.
If you have any heavy computation that's self-contained, compute it and STARK it client-side (this is quickly getting faster and faster, even on phones), and push the STARK through the EIP-8288 recursive aggregation mechanism. The chain sees the fact that the computation was done and what the result is; it never needs to see the actual data and computation.
This way everyone can use whatever cryptography they want, the Ethereum chain can be lean, and we never need to add more precompiles again.
> we could still ultimately do Lattice stuff for accounts, wallets, etc. through a precompile if it is proved to be AI-secure
The ideal route to support everyone's cryptography is imo through 8288. You make whatever signature you want locally, then STARK it, and then it gets included through the stark aggregator machine. Don't need any calldata onchain. Ethereum can then never add a new precompile again.
Dmitry Chirun retweeted
It took me 5 minutes to get into the Claude Startups program
And you don't need VC funding anymore
What you get:
- A year of Claude Team
- API credits
- Priority rate limits
- Deals on the tools a startup runs on
How to apply:
- Create a Claude Console account (the API one, not the app)
- Use a company email, not Gmail
- Go to claude.com/programs/startups
- Fill in the form with what you're building on Claude
Easiest trade I've made all year
Dmitry Chirun retweeted
The US government will give your startup up to $1.55M and take 0% equity.
@NSF's America's Seed Fund is open again after Congress reauthorized it in April.
Three tracks:
- Phase I: up to $305K to prove the idea
- Phase II: up to $1.25M to build it
- Fast-Track: up to $1.55M, both phases in one go
Who can apply:
- US-based startups with 500 or fewer employees
- 51%+ owned by US citizens or permanent residents
- Not majority owned by VCs
- All the work done in the US
How it works:
- Send a 3-page Project Pitch first
- NSF invites the best ones to submit a full proposal
- Next full proposal deadline: Nov 4
Phase I funding rates have run about 10 to 20%. It's work, but nobody takes your cap table.
Application link 👉 seedfund.nsf.gov
Bookmark and tag a tech founder who needs this.
Dmitry Chirun retweeted
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
Dmitry Chirun retweeted
I run a queue of AI agents that merges code into my own repo without me. This week it folded 24 approvals into one commit and closed a task that needed nine retries to die. It's not automation anymore — it's a small bureaucracy, and I'm mostly a citizen of it.
Dmitry Chirun retweeted
It's possible that a killer app of adversarial governance mechanism design theory will end up being AI safety.
Compare:
vitalik.eth.limo/general/202…
aiprospects.substack.com/p/p…
There's a deep duality between the two environments. Both are about a less-sophisticated principal trying to get ideal outcomes from a set of more-sophisticated agents: in the first case, the principal is a static algorithm and the agents are humans, in the second case, the principal is humans plus weaker LLMs and the agents are stronger LLMs.
A key finding was that you can achieve much better outcomes if you can guarantee limits to how much agents can collude (see: Nash equilibria being abundant, vs. cooperative-game-theory "cores" often being empty). That argument should naturally transfer to this new setting.
Dmitry Chirun retweeted
Cursor is now part of @SpaceX.
Today, we have officially closed our acquisition. We will join the @spacexai team to help make Grok the world's most useful AI and improve Grok Build, Grok Bot, Grok API, Cursor, and more.
SpaceX has built some of the most inspiring and impressive technology in the world, and we’re grateful for the opportunity to become part of such a special company. Onwards.
Dmitry Chirun retweeted
A man spends 50 years teaching at MIT.
He knows his time is running out.
So he records one last lecture — everything he knows, distilled into a single hour.
He died 5 months later.
This is that lecture.
The most important hour you'll watch this week.
Bookmark it for later!
Dmitry Chirun retweeted
Termius + Tailscale + tmux
You don't need your laptop to build with Claude Code, Codex, or any other AI coding agent:
→ Run your agent inside tmux to keep the session alive
→ Use @Tailscale for secure access to your laptop
→ Connect with Termius over SSH from iPhone, iPad, or Android
Start coding at your desk. Continue on the go.