LOL, I got 100% nerd-sniped by my friend Sönke this week and wound up building a small spaceship.
On Monday he's like "Hey, what if you found obscure seed phrases embedded in public texts? You'd only need to remember the name of the book and the paragraph and go from there."
I honestly could care less about crypto(currencies) and I'm 100% sure this is like cryptanalysis 101. But, yeah, it seemed like an interesting problem anyways.
First, I downloaded a few hundred books from Gutenberg, wrote a ruby script and found BIP39 word sequences with a tolerable buffer for filler-words.
Then, I was like, okay, gotta now check them against actual addresses. Downloaded a list of funded ETH addresses. Wrote the checker in ruby. Ran it. No hits but this was now definitely weirdly interesting.
Because: And what if I downloaded the whole pg19 text corpus to scan! And what if I'd add BTC addresses! And what if I checked every permutation of the seed phrase!
Everything got really slow once I got to processing 12G of raw text for finding sequences and then checking a few million candidates with 44.000+ variations per candidate.
So, let's rewrite this into C! And since I've got 16 cores, let's parallelize this puppy! And since it's a MacBook, let's use GCD! Optimize all the things!
Lol, so NOW this thing is so fucking FAST. Takes four minutes to go through the full pg19 corpus and generates 64,205,390 "interesting" seed phrases. The fully parallelized checker (see Terminal screenshot) processes 460 derived addresses per second.
I really don't care if I get a match or not. I feel like I started with building a canoo and wound up with a spaceship is in itself just the best thing in the world.
Also a huge Eno fan here. Put together, I probably have listened to Music for Airports, Another Green World, Taking Tiger Mountain and Discreet Music more than any other artist. Maybe Philip Glass comes in at a close second.
Anyways, in 2016, Tero Parviainen (@teropa) shared this really cool long-form exploration called "JavaScript Systems Music – Learning Web Audio by Recreating The Works of Steve Reich and Brian Eno" that I enjoyed tremendously (and I don't even like Javascript!)
Thanks for sharing. I've been on a path of algo music with JavaScript (I also do not enjoy JavaScript) and have mostly just guess-and-checked my way through it. I'm going to work through this as my advent of code project.
Yesterday I put up a little dictionary of synth sounds that I'm building out to help me on my journey (https://synthrecipes.org). The goal to be able to export any particular sound in a format for different live coding environments. Sounds are defined in a JSON format like https://synthrecipes.org/recipes/acid-bass.json. I'll open source it today so other can submit sounds.
Music is funny. I played the closed hi-hat sound (https://synthrecipes.org/#closed-hi-hat) a couple of times and my brain instantly started playing AC/DC's, Back in Black. I probably haven't listened to that song in 15 years and now I'm shuffling AC/DC on Spotify.
When you feel bored listening to this sort of music you are already half way to the Alpha state (I heard it called that by Quincy Jones). Go a little further, and when your brain fully disengages you can use the space/quiet/calm to go to new places and come up with some amazing ideas.
Hi Max! Thank you for updating my mental model of AI detectors.
I was with total certainty under the impression that detecting AI-written text to be an impossible-to-solve problem. I think that's because it's just so deceptively intuitive to believe that "for every detector, there'll just be a better LLM and it'll never stop."
I had recently published a macOS app called Pudding to help humans prove they wrote a text mainly under the assumption that this problem can't be solved with measurable certainty and traditional methods.
Now I'm of course a bit sad that the problem (and hence my solution) can be solved much more directly. But, hey, I fell in love with the problem, so I'm super impressed with what y'all are accomplishing at and with Pangram!
Off-detail/on-topic: After 10 months of reading Pratchett's Discworld novels, I'm now reading the Cantebury Tales. And by golly, the tales are surprisingly accessible, entertaining and fun to read.
We're a classic XP shop. To build new features in our brown-field app, we defined about 8 sub-agents such as "red-test-writer", "minimal-green-implementer" and "refactorer".
Now all I do in Claude Code is: "Build this feature X using our TDD process and the agents." 30 minutes later the feature is complete, looks better and works better than what I would have built in 30 minutes, is 90% tested and is ready for acceptance testing.
Granted it took us years of working XP, pairing, TDD etc. but I keep feeling confused about posts like this.
We've been shipping production-grade code written 95% by AI for over a year now. Non-trivial, complex features.
There is no secret sauce even, in how we do this. It works. Really, really well for us.
We're practically a 100% XP shop compiled of ex-Pivots and Thoughtworks. Pairing, TDD and client-on-site as our baseline. We've also been using AI as part of our IDEs full-time for 2+ years.
Yet, the most unexpected thing happened this year on my team of 4 senior/staff-level developers:
Instead of "splintering/pairing off with AI" individually even further, we wound up quadrupling (mobbing) full-time on our biggest project to date. That meant four developers, synchronously, plus Claude Code typing for us, working on one task at a time.
That was one of the most fun, laser-focused and weirdly effective way of combining our XP practice with people and AI.
Technically my wife would be a perfect customer because we literally just prototyped your solution at home. But I'm confused.
For context:
My wife does leadership coaching and recently used vanilla GPT-4o via ChatGPT to summarize a transcript of an hour-long conversation.
Then, last weekend we thought... "Hey, let's test local LLMs for more privacy control. The open source models must be pretty good in 2025."
So I installed Ollama + Open WebUI plus the models on a 128GB MacBook Pro.
I am genuinely dumbfounded about the actual results we got today of comparing ChatGPT/GPT-4o vs. Llama4, Llama3.3, Llama3.2, DeepSeekR1 and Gemma.
In short: Compared to our reference GPT-4o output, none (as in NONE, zero, zilch, nil) of the above-mentioned open source models were able to create even a basic summary based on the exact same prompt + text.
The open source summaries were offensively bad. It felt like reading the most bland, generic and idiotic SEO slop I've read since I last used Google. None of the obvious topics were part of the summary. Just blah. I tested this with 5 models to boot!
I'm not an OpenAI fan per se, but if this is truly OS/SOTA then, we shouldn't even mention Llama4 or the others in the same breath as the newer OpenAI models.
Ollama does heavily quantize models and has a very short context window by default, but this has not been my experience with unquantized, full context versions of Llama3.3 70B and particularly, Deepseek R1, and that is reflected in the benchmarks. For instance I used Deepseek R1 671B as my daily driver for several months, and it was at par with o1 and unquestionably better than GPT-4o (o3 is certainly better than all but typically we've seen opensource models catch up within 6-9 months).
Please shoot me an email at [email protected], would love to work through your use cases.
To your point: Do I understand correctly that, for example, by running the default model of Llama4 via ollama, the context window is very short even when the model's context is, like 10M. In order to "unlock" the full context version, I need to get the unquantized version.
For reference, here's what `ollama show llama4` returns:
- parameters 108.6B # llama4:scount
- context length 10485760 # 10M
- embedding length 5120
- quantization Q4_K_M
Feedback: First off, I really like your app's style. I love bold colors. The screenshots and text are clear and understandable - maybe except on how the data gets in there. Even if that's by hand, I still think this is a great first version and a solid product.
While I'm not in your workout target group - nor on iOS - it still resonates with me because I use Oura (the ring) specifically for their detailed heart-rate tracking and stress tracking. My most-used feature in their app is my stress-tracking throughout the day.
Feature request: Only to explain how data gets inserted.
For reference, here are the two heavy-lifting workers:
- https://github.com/akaalias/bipscan/blob/main/src/c/find_seq...
- https://github.com/akaalias/bipscan/blob/main/src/c/check_se...
and here's a screenshot of the thing running:
- https://x.com/SpringStreetNYC/status/1996951130526425449/pho...
and here's the full story:
LOL, I got 100% nerd-sniped by my friend Sönke this week and wound up building a small spaceship.
On Monday he's like "Hey, what if you found obscure seed phrases embedded in public texts? You'd only need to remember the name of the book and the paragraph and go from there."
I honestly could care less about crypto(currencies) and I'm 100% sure this is like cryptanalysis 101. But, yeah, it seemed like an interesting problem anyways.
First, I downloaded a few hundred books from Gutenberg, wrote a ruby script and found BIP39 word sequences with a tolerable buffer for filler-words.
Then, I was like, okay, gotta now check them against actual addresses. Downloaded a list of funded ETH addresses. Wrote the checker in ruby. Ran it. No hits but this was now definitely weirdly interesting.
Because: And what if I downloaded the whole pg19 text corpus to scan! And what if I'd add BTC addresses! And what if I checked every permutation of the seed phrase!
Everything got really slow once I got to processing 12G of raw text for finding sequences and then checking a few million candidates with 44.000+ variations per candidate.
So, let's rewrite this into C! And since I've got 16 cores, let's parallelize this puppy! And since it's a MacBook, let's use GCD! Optimize all the things!
Lol, so NOW this thing is so fucking FAST. Takes four minutes to go through the full pg19 corpus and generates 64,205,390 "interesting" seed phrases. The fully parallelized checker (see Terminal screenshot) processes 460 derived addresses per second.
I really don't care if I get a match or not. I feel like I started with building a canoo and wound up with a spaceship is in itself just the best thing in the world.