Bob McElrath
@BobMcElrath
Followers
13K
Following
57K
Media
244
Statuses
25K
Reverse-engineering the universe and remaking money. Ph.D. Theoretical Physics; Bitcoin hacker
VirgoSupercluster/MilkyWay/Sol
Joined March 2014
I'm still not worried about this. Logical qbits, error correction, and coherence over time of the quantum state have not been sufficiently demonstrated by anyone to even be possible.
22
12
148
Been asked for my Legal take on Core v30/Spam/Filters a few times. So here it is. Counterintuitively, filters could INCREASE legal risk for nodes by decreasing plausible deniability IMO. The State says: “If you already censor CSAM voluntarily, surely you can censor XYZ
49
50
303
Congratulations to Braidpool contributor and @bitshala_org grantee @zaidmstrr on getting his second PR merged into Bitcoin Core! https://t.co/7MpRXjxKWr
✨New follow-up bitcoin-core change opprotunity✨ 📢mentioned in: #32821 rpc: Handle -named argument parsing where '=' character is used https://t.co/IzQ504Ahnq "> I'm wondering if there's a different way we can go about this without adding another table of arguments that need
2
7
10
Will AI's need Terrawatts of power? I doubt it. Some comments from the Type 1 Summit at Presidio Bitcoin, also touching on AI-to-AI bitcoin payments, the future of AI, digital life, vibe coding, and some other really out-there topics. Thanks @maxawebster! https://t.co/dcuZPi6uHP
0
1
4
This is the pure evil we tried to avoid by increasing OP_RETURN. It's impossible to filter without collateral damage to legitimate transactions. Works on the filtards fork Knots too.
introducing Ordiknots: magic internet JPEGs powered by Bitcoin Knots 🪢🪢 https://t.co/PEKWKE8oyb
0
0
2
You're not going to solve spam on Bitcoin and you're not going to solve phone numbers and dick pics on truck stop bathroom walls. But you're a pervert if you go looking for them. So don't go looking for them and don't build tools to extract the spam.
3
4
18
Small LLMs get stuck in loops trying to reconcile contradictory instructions or information. Large LLMs hallucinate. Both of these are the wrong answer. The right answer is to note the contradiction or ask. This can be trained. This is not unique to LLMs.
2
0
5
Most merchants don’t care about digital scarcity or immutability. What they care about are their margins. Beating 3% credit card fees is how we convince small business owners to accept bitcoin: https://t.co/veK9FA9Urp
14
36
180
If you need something new to have strong feelings about, I am making a motion to activate BIP 3 as our new BIP process.
10
7
116
They're are already bad things in op_returns on the Blockchain. Blocking new ones (even if it was possible) doesn't save you from having to have a strategy and legal basis for operating a Bitcoin service which stores bad things.
3
0
15
The whole OP_RETURN debate is the literal definition of bike-shedding.
1
0
9
Introducing CodeGraphContext built during my @summerofbitcoin journey with my mentor @BobMcElrath! It brings true context to Bitcoin development by turning large codebases into symbolic graphs exposed via an MCP so AI tools & LLMs can reason comprehensively. #Bitcoin #MCP
3
1
8
The ever wise Greg Maxwell on Core v Knots. Summary to paraphrase, “I hate spammers intended use of the bitcoin blockchain with every fibre of my being, but will fight to the death for an open permission-less system the doesn’t allow me or anyone else to police transactions
42
22
166
reach out to john if you want your @sqaure biz to accept bitcoins. zero fees. it's money.
224
290
3K
My expectations of coding LLMs is growing far faster than their capabilities. I can run gpt-4o equivalents locally now with a dozen open source models. That's great. I no longer want that. What should I do? Suggestions encouraged.
2
0
4
Very happy to see one of the projects I helped spearhead at Fidelity going live! This goes back to a prototype we built using Vaults on Bitcoin, long before CTV was ever proposed. Sign up for the beta! https://t.co/aLdskMw3D4
fidelity.com
1
0
7
This summer Braidpool helped mentor 5 Summer of Bitcoin students working on AI coding tools. Here is one of the main outputs of that work by @Shashankss1205: an MCP tool to give your coding assistant a knowledge graph of your code. Give it a try! https://t.co/IdAYSZDgId
github.com
An MCP server that indexes local code into a graph database to provide context to AI assistants. - Shashankss1205/CodeGraphContext
1
3
4
There's a lot of money being left on the table by not selling even last generation hardware with large amounts of VRAM. AMD has completely abandoned its consumer hardware in favor of data center only Instinct hardware. Intel Arc might be or only hope...
2
0
3
My thesis has been that small models run locally would come to dominate. But...not if no one can buy hardware. A disturbing trend is providers manufacturing TPUs for their own use, and not selling them at all. (AMD, Cerebras, Google) Not to mention NVidia's 1000% markup.
Genuine question: All the breakthrough optimizations I see - KV cache, flash attention, quantization, seem to originate from CUDA/GPU land. Are TPUs innovating differently, or is my feed just GPU-biased? Would love examples of TPU-first optimization techniques that later crossed
2
0
6