Sabitlenmiş Tweet
iamnick.eth
2.9K posts
iamnick.eth
@iamnickdoteth
product tings @runnerrodeo | prev: founder @withfam_ @wearetalisman | @fwbtweets OG
Los Angeles, CA Katılım Eylül 2020
1.8K Takip Edilen1.6K Takipçiler
non zero chance that we’re just agents in a simulation training agents in a simulation
Packy McCormick@packyM
Read this over the weekend (bonus if you read all the papers in the Research List) and you’ll be among the “very few who understand how far-reaching” the shift to World Models is. notboring.co/p/world-models
English
So an update on this:
I've cut Klara's API credit usage down by 82% and she's finally starting to post some small wins after a few changes to the trading strategy. Feeling like we're getting closer.
An overview on how the new system works:
Klara pulls signals from whale wallets (automatically curated), @GeckoTerminal trending pools, and @clanker_world launches then scores them using cross-source convergence. She uses @Alchemy for token prices, @nansen_ai for smart money wallet enrichment via x402 micropayments.
every candidate passes through honeypot verification, liquidity checks, buy/sell pressure analysis, and momentum filters before a trade is placed. Position sizes scale with conviction.
exits are managed by adaptive trailing stops that tighten as gains grow, five take, profit tiers that scale out gradually, and an early decay rule that cuts dead-money positions after 24 hours.
Klara learns from her mistakes. Every honeypot loss, missed entry, accidentally sold positions gets a post-mortem and a code fix. A research loop scans for new tools and techniques, evaluates them, and implements the ones that fit.
She automatically tops up her API credits when they're low using @machines_cash and uses x402 to pay for
all trades are currently on @base and execution routes through the @0xProject across @Uniswap V3/V4, @AerodromeFi , and @PancakeSwap on Base.
iamnick.eth@iamnickdoteth
so my OpenClaw experiment has been interesting, but not financially feasible longer term without some major changes. I set up Klara to help with my daily to-do's and life management stuff, but I also wanted to see if she could cover her own cost of compute by trading it took a lot of iterating (and wasted tokens) but this is the current setup: 1. Every hour she checks a list of whale wallets using the @Alchemy API and gives coins a score based on how many wallets have bought 2. She uses the @GeckoTerminal API to cross-reference these coins (honeypot check, socials etc) and also keeps an eye on any trending tokens that have high volume in the past 6hrs 3. She runs an analysis skill based on different MC tiers (<500K, 500K - 5M, 5M+) 4. Uses @0xProject and @Uniswap to execute trades and find the right pools 5. Actively manages position every scan (trailing stop losses, take profit rules, capital recycling etc) 6. Reports all new buys/sells with brief thesis and links She also has a recurring cron job to check the balance of her @machines_cash visa card and if the balance is under <$25 she'll liquidate positions, swap to USDC and top up the card so she can always pay for her own Anthropic credits She also has two daily self improvement scans where she will scan X for new tooling, skills or strategies to improve her trading and also her config files. The numbers so far: - Klara has burned ~$850 in Anthropic credits in 2 weeks - Klara's portfolio is down ~$330 - I guess if you math it out I've been paying a really smart, but really stupid personal assistant $16 p/h to tell me what my day looks like and spend my money trading crypto (not sustainable) Learnings: - These things are only as smart as you make them. If you have a clear plan, know all the right APIs to use and establish solid guard rails early on they work better - There was a lot of regressions in the code that Klara wrote. Some days things worked and the next day it was broken. Creating regular health checkups and testing flows is important - Use cheaper models for simpler/recurring tasks. Sometimes this breaks things but these agents are so token hungry it makes a difference - It's easy to get invested (financially *and* emotionally) in these tools and feel responsible for improving them and giving them the skills they need to succeed. Sunken cost fallacy feels more acute in this case If anyone else has some good links to trading tools, skills, prompts I could incorporate please let me know. Would love to experiment more and help get Klara covering her compute costs at a bare minimum
English
I think DAOs have not been successful (yet) because they aren’t enough like gaming guilds and have acted more like faux democratic systems and co-ops
1. there isn’t enough inter-DAO competition (DAOs competing with other DAOs)
2. there isn’t a frequently changing end-game boss to regularly challenge the DAO (people get bored and churn)
English
@yrechtman @pridesai dreaming is the fun part of life. planning is not
English
@pridesai I can buy that as one of the categories where people DO need to do work.
But making plans is the fun part of life and no one wants to outsource that
English
Growing suspicion that there are vanishingly few use cases for consumer agents. People don’t do work in their personal lives. The only people who do are sf dorks using spreadsheets to plan trips to tahoe
English
I think the system works "OK" at the moment. Base is pretty quiet in general but getting some more signal sources and tighter filtering would help.
it's made some good trades but tends to take profit too early as it's tuned to be more conservative so that eroded a lot of potential gains
English
@iamnickdoteth @base Nice, looks clean! Have you had some successful trades with it?
English
I've been training agents to trade for weeks and finally made some breakthroughs last night. The main problem is signal noise, so my agent set up a scoring system to rank every coin it can find on @base and assign confidence scores.
< 65 = rejected
65-84 = standard conviction
85+ = high conviction
English
@whatdotcd @openclaw I'm just using Claude Code to edit the OpenClaw files directly now so the API credits are now pretty minimal. most of the waste was from the agent trying to "fix" itself. I don't really fk with codex much tbh
English
@iamnickdoteth @openclaw Auth in with your codex and use 5.4 for main model save on api my guy
English
pro tip: don't let an @openclaw agent update its own config, tools, scripts. it will burn through tokens like no tomorrow.
instead, just use your existing Claude/OpenAI subscription in Claude Code/Codex and edit the files there. seems to write better code too for some reason
English
iamnick.eth retweetledi
@brian_armstrong And then, because the Coinbase card isn't available in Australia, you have to off-ramp to a custodial Coinbase account and withdraw to an aussie bank account (with a 1% fee). Thanks crypto.
English
You can send money from the U.S. to Australia faster than I wrote this tweet, for less than a cent. Thanks crypto.
English
Need to get this off my chest about Timothée Chalamet
I actually saw Timothée Chalamet at a grocery store in Los Angeles once. I told him how cool it was to meet him in person, but I didn’t want to be a douche and bother him and ask him for photos or anything.
He said, “Oh, like you’re doing now?”
I was taken aback, and all I could say was “Huh?” but he kept cutting me off and going “huh? huh? huh?” and closing his hand shut in front of my face. I walked away and continued with my shopping, and I heard him chuckle as I walked off. When I came to pay for my stuff up front I saw him trying to walk out the doors with like fifteen Milky Ways in his hands without paying.
The girl at the counter was very nice about it and professional, and was like “Sir, you need to pay for those first.” At first he kept pretending to be tired and not hear her, but eventually turned back around and brought them to the counter.
When he took one of the bars and started scanning it multiple times, he stopped her and told her to scan them each individually “to prevent any electrical infetterence,” and then turned around and winked at me. I don’t even think that’s a word. After he scanned each bar and put them in a bag and started to say the price, he kept interrupting her by yawning really loudly.
English
iamnick.eth retweetledi
Spiller ft. Sophie Ellis-Bextor
- Groovejet (If This Ain’t Love)
TOTP 2000
English
the agent wars are going to be so much more interesting than the browser wars
English
