Bitcoin and rust
Joined October 2013
- Tweets1.1K
- Following725
- Followers771
- Likes3.3K
Pinned Tweet
After working on Sv2 for more or less 5 years, we finally proved it with a block on bitcoin mainnet!
Can't tell you how happy I am about it.
🔽 continue below
joereis.substack.com/p/build…
Without deep domain knowledge and a sense of taste, you lack the context to judge whether AI-generated outputs are generally useful or complete bullshit
fi3 🏴☠️ 🦀 ₿ retweeted
Dubious. Do you think P=NP? Some problems are fundamentally higher complexity, and accelerating mathematically capability doesn't change that any more than gravity. Plus for what bitcoin cares about, even if all algebraic were hardness problems lost, we have hash-based SHRINCS.
fi3 🏴☠️ 🦀 ₿ retweeted
This approach works perfectly well when you're working with a frontier AI model. And it should be the approach you take, especially when you're early in exploring/solving a problem. It reduces friction and helps you hone in on a solution early and with little effort.
It is also completely viable when you're not constrained by budget or AI token spend.
But you should also learn the disciplines of engineering AI harnesses, such as how to preserve the accuracy of your agent while downgrading the backbone model to a faster/cheaper LLM. Or if you've reached a solution and want to be able to scale it to hundreds, thousands, or millions of repetitions per day.
In those cases, having specific structure and rules in your instructions/harness can help reduce token consumption and AI costs.
I am surprised that people are surprised this is how I prompt Claude.
Talk to Claude the way you would a coworker. There's no secret to prompting. There's no need to be overly scaffolded or prescriptive for most tasks -- give Claude a goal, and it will figure it out.
Back in the Sonnet 3.5 days, your prompt mattered a lot. Nowadays, it's much more important to communicate to the model:
1. What you want it to do
2. How much effort you want it to spend
3. How it should verify that it did the right thing
I built ab-lab to test whether changes to a coding-agent workflow actually help. It lets you compare models, harnesses, instructions, and author/reviewer configurations against a fixed repository snapshot, with explicit budgets and independent evaluation.
👇
You can use the included benchmarks or define tasks for your own repository.
blog.filippomerli.com/articl…
fi3 🏴☠️ 🦀 ₿ retweeted
the way these subscription plans work is there's a pool of compute and you maximize how many people it can serve
that's the only way to justify not using it for high margin api demand the sales team keeps bringing in
individuals with a lot of accounts makes it all less viable
fi3 🏴☠️ 🦀 ₿ retweeted
Stop caring so much about model releases.
Focus on the harness, and improving the environment your agent operates in.
You'll find yourself far ahead of the curve.
fi3 🏴☠️ 🦀 ₿ retweeted
Does your Claude model really need Claude Code…? 🤔
We evaluate 7 models on Claude Code, Codex, and Pi. Three surprising findings emerge:
1️⃣Harness choice has little effect on task success rate, but can significantly affect the cost
2️⃣A simple harness can be competitive
3️⃣The native harness isn’t always the best.
Millions of people are using coding agents, but the impact of harness choice remains unclear.
(1/n) More details in the thread. 🧵
fi3 🏴☠️ 🦀 ₿ retweeted
Oof. See, this is why I’m not worried about “pure” LLMs in the long run.
There’s new evidence that the human brain doesn’t use natural language centers for logical reasoning at all.
Study just came out where stroke patients with complete aphasia (near-chance on written comprehension) performed….great on pure logic tasks! Even in healthy adults, fMRI showed language area activation didn’t scale with logical difficulty.
tl;dr words are too ambiguous for formal thought and our brain doesn’t use language symbols for logic
fi3 🏴☠️ 🦀 ₿ retweeted
‼️ Linus Torvalds told Linux kernel developers this week that Linux is not one of those anti-AI projects and that AI is now a clearly useful tool.
He says AI's usefulness was still arguable a year ago but no longer is, and that the doubters simply haven't used the tools. Nobody is forced to use AI, but he will ignore anyone trying to stop others from using it. Anyone who wants an anti-AI project, he says, can fork the code or walk away.
In 2024, Torvalds dismissed AI as mostly marketing. By this January he was shipping AI-generated Python in his own hobby repo, and the kernel now accepts AI-written code under disclosure rules.
💯
The importance of having, or not having, IP (Intellectual Property) for economic and social progress.
#Against_IP
fi3 🏴☠️ 🦀 ₿ retweeted
"This is not surveillance. This is love."
#ChatControl 🇪🇺
fi3 🏴☠️ 🦀 ₿ retweeted
We will never switch an endpoint to signal for a BIP that hasn't reached consensus.
Our miners can signal for anything they want.
Both statements are true at the same time. That's the point.
Miner Signaling via DMND StratumV2:
blog.dmnd.work/miner-signali… #Bip110
fi3 🏴☠️ 🦀 ₿ retweeted
DMND has mined the first known Bitcoin block using Stratum V2 Job Declaration.
Block 955,318, on mainnet, mined by @GoMining, who built and declared their own block template and included their GoBTC Pay transactions.
mempool.space/block/00000000…
Another year passed and we now have enough hashrate to find blocks on mainnet. For sure this is one of the biggest milestones.
🔽 continue below
A big thank you to everyone that worked on the Sv2 specification, to anyone that worked
on SRI and on DMND, and to anyone who belives in it to the point to putting their money in.
mempool.space/block/00000000…