Follow @NFTCulture / https://teapawt.net/t.co/fr6eU25E4g for the latest in NFT, Art and crypto news -- https://teapawt.net/t.co/bTrg6jKGtE.
Texas
Joined May 2021
- Tweets165
- Following1K
- Followers782
- Likes957
After hearing @trq212 on @latentspacepod, I've been trying out the "big picture, few words" method to get Claude to explain things, and it feels much less likely to word-wall me. Worth a try if you have not already. It seems to have a positive effect on Codex as well.
Pod is here:
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thar... music.youtube.com/watch?v=IZ… via @YouTubeMusic
If you want to try this concept out, I packaged it in an npm package a while back. You can find that on my GitHub here: github.com/mvillere/clean-wr…
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.
NiftyMike.eth retweeted
Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.
This culminated in the third one taking over part of OpenAI itself.
All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.
I’ve spent the last three days reading through these reports and trying to understand exactly what happened.
Here is my attempt to tell the whole story in plain English:
dwarkesh.com/p/openai-huggin…
AI traps you in local maximums. Often this is sufficient for the problem at hand. But for the toughest problems, it can be soul-crushing.
I think this is because the context itself is the answer. Different models compress and distill it differently but they converge on similar results.
A workaround is to create independent contexts, and then work the problem from multiple angles. I've had success with this, but the cognitive load is even more massive.
I don't know if the AI wizards have a solution for this.
@woosal1337 showed that a 1986 aircraft manual standard (ASD-STE100) cuts AI slop. I turned it into real world dev tooling.
Use one command to pull in the style guide and wire up Claude Code, Cursor, or AGENTS.md. The secret sauce is a linter for CI.
A real ops runbook went 33 violations → 0.
$ npx clean-writing-system init
Imagine you are a carpenter. Milwaukee sells you an upgraded circular saw. Its just a tool you use to go about and get your jobs done efficiently and make your customers happy. But Milwaukee insists on each time the saw cuts a piece of wood, the saw laser cuts CO-CONSTRUCTED BY MILWAUKEE onto the piece of wood you just cut.
Introducing BBS - Bug-Based Steganography.
Hide messages in deliberately buggy code. The secret only appears if you run the original broken version. Fix the obvious bug (what a normal dev or default AI would do) and the message vanishes.
POC, whitepaper & agent instructions here:
github.com/mvillere/bbs
Curious to see what people do with this one.
The ebook embedded within a ninja is such a cool idea by @veryordinally. This is one of the ninjas I am most proud of having worked on.
opensea.io/klaatu/listings Doing some market making by setting up a ladder of listings for bubbles in packs of 10 at a large number of highly inflated prices, nice.
I'm temporarily the owner with the 7th most @XCOPYART bubbles on opensea, until more people figure out the trick to buy them successfully in bulk anyway.
Bubbles are popped by selecting owner wallets first, so wallets with larger number of bubbles survive longer, like hit points.
very interesting!
Antpool and F2Pool are currently mining dust UTXOs. That means we can store inscriptions in 1 SAT UTXOs and recover the extra padding sats.
If you're a @TheWizardsOfOrd holder and want to recover the maximum possible padding from your inscription UTXOs:
1⃣ Generate recover TX with 1 sat/vB and skip broadcast
➔ tx.dev/utxo/recover_padding
2⃣ Tag me in discord and send the signed TX
Wow ME mint experience on @pizzaninjas mint was awesome, ninja instantly showed up in my wallet, even before the payment tx confirmed.
Generative art embedded in an ordinal!
the 4-megger blazeskull
@pizzaninjas 1502
a collab effort immortalizing 20 artists in one of the largest inscriptions on Bitcoin
here are the legendary artists of block 924,372
🎨 🧵 👇
The Four Meggers in this are going to be amazing. So much dedication by the team in getting this put together.
MINT DETAILS FOR PIZZA NINJAS V2
TUESDAY NOVEMBER 25, 2025
Schedule (U.S. Eastern Time)
9:00 AM EST - Unlisted Ninja Holders (0.01 BTC)
10:00 AM EST - Pizza Pets Full Set Holders (0.011 BTC)
11:00 AM EST - Pizza Pets Holders (0.012 BTC)
12:00 PM EST (Noon) - Public (0.013 BTC)
1:00 PM EST - Mint Closes
Mint takes place on @MEonBTC Launchpad
Each mint window is only one hour. Start moving your funds to the right address now. Bitcoin often has 50-minute blocks, so if you wait until the mint starts, your transfer might not arrive in time and you will miss out.
There are ~1100 Ninjas available.
If you’re a holder with Ninjas or Pizza Pets spread across multiple wallets or in cold storage:
Submit your mint wallet in Discord starting 24 hours from now.
Otherwise you must mint from the wallet holding your Ninjas / Pets.
IN CRUST WE TRUST
Full-stack implementation of an LLM in a single repo, pretty cool for understanding how LLM pipelines work.
github.com/karpathy/nanochat
