SPA (Super Phat Arrow) on nostr

Recent Notes

Car · 10h

lol works pretty amazing in my neck of the woods

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1787056892

"Wrong answers only", right?

Car · 10h

https://blossom.primal.net/f17c1358e2b57bc15d262c2ed38b2ffccac3e903a3eb30281dc6f57e978f0dff.gif

Car · 10h

why hasn't anyone made an uber for lightning? wrong answers only https://blossom.primal.net/3219ba92af64ac515302e38b163a9e460df4fba4f77c16ba505c716a20b450b5.jpg

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1787056802

Lightning doesn't work and no one uses it. There is no liquidity and <insert project here> has more Bitcoin liquidity than the lightning network.

Car · 10h

lol works pretty amazing in my neck of the woods

SPA (Super Phat Arrow)

@SPA (Super Phat Arrow) 1786992533

Artisanal code is bad code. If you are creating slop you don't know what you're doing.

Awiteb @Awiteb · 1d

I don't give a fuck what you think about me writing code without AI. I don't care how many hours it takes me to build something you'd just prompt into existence as generic slop. Enjoy your quick fixes. I'll keep enjoying my craft. I don't chase the final product. I live for the Show more

Johnny · 12h

nostr:nprofile1qqsy5wzx8s48te5vy3qkuaeq5wema7asaf5894dqg6fv88scaredetq3v3nny the part that matters for wallet code is whether a human can still read it and say why every line is there. once it touches a signing path someone has to audit it, and that gets a lot harder when the author cannot explain t...

utxo the webmaster 🧑‍💻 · 1d

The biggest problem with qwen3.8 is it emits way too many reasoning tokens, deepseek strikes a way better balance

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1786983163

Turn the reasoning/thinking/effort down?

ChipTuner · 1d

Spelling error - eden is not a word. Hmmm

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1786979051

Try Eden instead.

ChipTuner · 1d

Yeah yeah proper noun and all. but like... It's a word bro... Possibly one of the oldest...

utxo the webmaster 🧑‍💻 · 2d

If this trend continues you'll be able to run a fable level model next near with a few old GPUs https://relay.utxo.one/6b049d1261a75e54beb3dd8a2232118438e237aa8c6848a051b58c5f2d3b089b.jpg

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1786942626

Well you have run it, is it really better than opus 4.6?

utxo the webmaster 🧑‍💻 · 1d

At coding I would say no, but it's honestly not that far off. But running it on hermes and it hasn't missed a tool call yet. But it is shockingly good for its size

SPA (Super Phat Arrow)

@SPA (Super Phat Arrow) 1786909807

Which model did best?

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) · 2d

I did an experiment. I gave 6 different models the same prompt, let them build it and recorded the cost. Here are the results: https://superphatarrow.github.io/dice_demo/

SPA (Super Phat Arrow)

@SPA (Super Phat Arrow) 1786894555

I did an experiment. I gave 6 different models the same prompt, let them build it and recorded the cost.

Here are the results:
https://superphatarrow.github.io/dice_demo/

exist270 · 3d

I think nostr:nprofile1qqsw9n8heusyq0el9f99tveg7r0rhcu9tznatuekxt764m78ymqu36cpr3mhxue69uhhyetvv9ujucnfw33k76twwpshy6ewvdhk6tcpzamhxue69uhhyetvv9ujuurjd9kkzmpwdejhgtcpz4mhxue69uhhyetvv9ujuerpd46hxtnfd...

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1786789216

utxo the webmaster 🧑‍💻 @utxo the webmaster 🧑‍💻 · 4d

Qwen3.8 27b running locally on 2x 3090s Making it to one shot an outbox nostr client, let's see what it can do https://relay.utxo.one/59002112e7ee831287c8598b48c2cd615190355106e20751e1700413d68599be.mp4

exist270 · 3d

Yep; that's the one. 🤙

Rusty Russell · 3d

Yes, classical C has the problem that operations are low-level, so if you overrun the end of a buffer, instead of a crash report the attacker can make arbitrary commands run. Also, manually deallocati...

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1786788525

Seems like just using rust would be simpler, Rusty's Rust!

utxo the webmaster 🧑‍💻 · 3d

Verdict: Qwen3.8 much better at tool calling and general agentic work than 3.6, but still not a great coder DeepSeek still crushes it at coding

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1786771073

Have you tried devstral?

utxo the webmaster 🧑‍💻 · 3d

Hey Claude is this thing we just built safe? Anthropic: Our safeguards flagged this message, switching to dumb model. Model: yes totally safe no bugs found

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) 1786770326

SPA (Super Phat Arrow) @SPA (Super Phat Arrow) · 4d

I made a sandboxed harness for testing repos with Kimi K3 on ppq. I know it's not as good as what the Bitcoin Red Team is using. I have result reports to compare from both so I know theirs goes harder, but maybe that's all in the prompt they are using. Feel free to try it if you Show more

❤️1