Corki

34 / ML dev / bots that print

agents on @claudeai. built on @AnthropicAI

[Corki @Corki_0x]

@dangijpeg a legend of the sports betting markets

[Corki @Corki_0x]

@dangijpeg that’s exactly right

[Corki @Corki_0x]

I gave my Grok Bot $50 and one task: make it 4x.

It's 3:40am. It's at 25.5x.

$50 → $1,277.

+$1,227 in one night. 21 trades. 19 green. 2 red.

It isn't one bot. It's four agents, each with one job:

> SNIPER hits fresh pairs in the first 5 minutes, but only after LP is burned and mint is revoked
> MOMENTUM waits for a volume breakout and rides it with a trailing stop
> DIP HUNTER buys −30% wicks and sells the bounce
> WHALE TAIL watches smart wallets and copies the top ones the second they move

SNIPER went 4-1. MOMENTUM went 5-1. DIP HUNTER and WHALE TAIL went 5-0 each.

The goal was 4x. It crossed that hours ago, then 9x, 16x, 25x. Nobody told it to stop, so it didn't.

Here's what the last few seconds look like:

-> MOMENTUM sells MET +37%, +$86, target hit
-> DIP HUNTER buys TRUMP on a −34% wick
-> WHALE TAIL buys PUMP, copied a top wallet
-> DIP HUNTER trails TRUMP out at +8%
-> MOMENTUM buys WIF, buyers 3.4x sellers
-> SNIPER trails JUP out at +11%

Every trade is small. Every exit is planned. 90% of them land.

I'm asleep. The four of them aren't.

[Corki @Corki_0x]

@kinderlayer I agree, the agents did a thorough job

[Corki @Corki_0x]

@0xAIlabs I can’t say anything about Claude just yet

[Corki @Corki_0x]

@kinderlayer This comparison is really useful for understanding the models

[Corki @Corki_0x]

@mmoyshaa Thank you for sharing this

[Corki @Corki_0x]

@bckfv_eth @entropyIO That’s a massive trading volume

[Corki @Corki_0x]

@hammertime_one Thank you, I’m delighted

[Corki @Corki_0x]

this is f*cking insane. andrej karpathy, co-founder of openai: "the LLM is rediscovering knowledge from scratch on every question"

his fix pulled millions of views: stop using AI only to write code. use it to build a second brain that remembers every article, pdf, transcript and conversation you ever fed it

this is what that brain looks like when claude code runs it. 3,206 nodes, 11,483 links, and every single one came from something i fed it

what you're watching on screen:

-> CLAUDE.md loads every session. 142 lines, hard cap at 200. it's the map of the whole brain
-> rules stay asleep until they're needed. open people/anna.md and only rules/people.md wakes up
-> skills run on command. "do my weekly review" -> 5 steps, stale contacts flagged
-> memory writes itself. say "source under every claim" once and it's saved for good
-> /doctor audits everything. it caught rules/writing.md contradicting CLAUDE.md before i did

how to build yours:

-> make a folder, open claude code inside it
-> ask it to set up a vault: CLAUDE.md for rules, /raw for sources, /wiki for pages
-> dump anything into /raw (transcripts, pdfs, meeting notes, screenshots) and say "ingest"
-> it writes the page, links it to what's already there and updates its own rules
-> from then on every question goes through the vault, answered with your own sources

5 minutes to set up. pays off every day after

the part most people will skip: a few years from now, the ones who started early will recall every book, call and idea with receipts. everyone else will be googling their own memories

full setup in the article below

bookmark this before another week of notes disappears

[Andrej Karpathy @karpathy]

We're starting to leave the territory where you'd test an LLM by e.g. "create an svg of pelican on a bicycle". As one idea to generalize it, I was interested what Opus 5 would do if I gave it the first paragraph of the Lord of the Rings, a 1M token budget (~$10) and asked for three js render of it. Opus went off for ~2 hours and wrote 5500 lines of code that (procedurally) rendered the story. It's kind of janky but fun. But it's a bit mindboggling that the LLM has to place and orchestrate various polygon assets in (x,y,z) coordinates and write code that animates it all, and that it even does anything at all.

I also like this kind of examples because no one in their right mind would ever spend the time to write something this custom but LLMs have all the stamina and patience in the world, so it's an example where we go from "no one would ever do this" to "sure, why not, it's ~free". There might be a lot more. But I'm excited about creating hyper custom worlds that you can imagine dropping players into, e.g. here to participate in the LoTR story as a spectator NPC, or one of the characters, or etc. Something like an ephemeral GTA of X on demand.

Last thought is that the domain of worlds/games exposes a weakness in LLMs: they can't easily audit their work because they aren't able to efficiently and natively perceive videos or play games within them. Here, Opus 5 had to very slowly and painstakingly take screenshots at different points, and it messed up a few times and created a bunch of jank. An example of raw capability (multimodal, gameplay) that I think is still quite lacking.

[Corki @Corki_0x]

@mmoyshaa Thank you for your rating

[Corki @Corki_0x]

@bckfv_eth Yes, that’s what makes AI so wonderful

[Corki @Corki_0x]

@mirnowq @BGPredict I urgently need a bot like this

[Corki @Corki_0x]

@mirnowq This is our future

[Corki @Corki_0x]

@evgenia_vibes You’ve finally got round to it

[Corki @Corki_0x]

@isagi1000x It's great that you shared that

[Corki @Corki_0x]

@0xAIlabs that’s exactly how it works

[Corki @Corki_0x]

@isagi1000x That’s a brilliant assessment of the work

[Corki @Corki_0x]

@kinderlayer I agree, it makes everything much easier

[Corki @Corki_0x]

@mopozeuX @Polymarket @PolymarketTrade I’d have retired by now