from-scratch Vision-Language Action models (VLAs) using stereo LiDAR grokking ML & FH Encryption algo implementations
Lausanne
Joined August 2012
- Tweets2K
- Following4.2K
- Followers825
- Likes34.3K
Pinned Tweet
high-end iphones have build-in LiDAR, which allows you to create a 3D model of your environment
using gpt-6-Astra I made an iphone app to scan (1st video) and a macOS app app to explore the 3D point cloud (2nd video)
next up: use 2 iphones on a fixed-rig to record Stereo LiDAR
building simple utilities for MacOS like a text editor etc. works really well on Claude or ChatGPT
I ask for no dependencies, ObjC, K&R style, FP, linear, low complexity (duplication of code OK)
it can test most of the code, but not the GUI
however, AppKit etc. is ported to linux, so if you tell Claude to test it using GNUstep, it actually can test the whole app, provide screenshots in advance
3
2
103
new Pascal's Wager just dropped
en.wikipedia.org/wiki/Pascal…
Pray for my immortal soul. I have 75 intercessory agents pleading with God to spare me from eternal damnation at a rate of ~500 Our Fathers/minute. Eventually a pope will declare them ensouled, retroactively validating their prayers. Aiming to spend 60 years in Purgatory tops.
2
146
this is a pre-trained model, asked what causes inflation
I preempt it with a year, to learn how inflation was thought about prior to Nixon-shock / end of Bretton-Woods
1
109
Android might really have an edge on Apple here
Today, on Google Play we announced new ways we are evolving the Play Store for an experience that is unique to you and the content you enjoy most:
✨ Use Play to wirelessly install and run your own custom, vibe-coded apps privately on your device.
🔍 Discover apps effortlessly with Ask Play & Gemini
🎮 Dial in discovery with the all-new Game Finder & Gamer DNA
74
Music Theory for Programmers is excellent
I keep going back to Terrence Howard on Joe Rogan, how the frequency of chemical elements corresponds to notes
how an octave is a full-loop so you get alkalini metals and noble gases lined up
I can't play an instrument, so I figured out music theory in JavaScript instead.
I now understand why there are 12 notes and why notes in the same key sound good 🎶
I've written up my findings with code examples you can run and hear.
84
tacit, experiential, and faith-based knowledge are big blind spots for LLMs
the Enlightenment paradigm means you don't write things down if there is no theory-of-change
(ie not trained upon)
eg:
- people that milk their own cow are stronger
- people that pray have fewer mental health struggles
- people that observe lent live longer
- leaving dough in the sun makes it rise better
this is self-reinforcing because the groups that focus on these types of knowledge tend to feel (!!) that spoken communication is more effective than written communication
in some cases, the realization that it is outside my scope of understanding, is a key part of what makes it work
if information and knowledge become like spam, I think the status game (which everything kind of is) will become more experiental
meaning, people will seek nature-religious experiences and talk about those in person, such as forest-bathing, hiking, swimming, diving, backcountry skiing, gliders, etc.
being able to lead a group on such an activity, channeling the feelings, will
79
Bastiaan Quast retweeted
"The culture on X today is remarkably friendly and inclusive. It is a cheerful place", I couldn't agree with @daringfireball more. X is the place to be if you genuinely like technology and progress. Just such a delightful optimism here. daringfireball.net/2026/10/x…
the NSA decompiler Ghidra is very cool, but look at bit archaic / arcane
porting it is easier than ever, but it is becoming harder and harder to get especially Claude to write good-ol'-fashioned clean code
much of training data is made by giant corporation, that often replace engineers, and so they need tests, framework, modularity, everything needs to be explicit
this is not clean code, and with LLMs we no longer need all the above, but the training data makes LLMs write bloatware
1
115
Astra6 and Opus5.5 both were very keen on doing this project, and making sure I uploaded the decompiled C code to their websites
undoubtedly, these companies are out of internet to train on, and decompiling old apps, games, etc. is more data
it is also an excellent eval, you can just compare the recomped binary, or alternatively, the workflow of the app itself
trying out this decomp tool Ghidra from the NSA (left image),
I'm targeting CrocClip from the Win3.11 era (right image)
interesting how the state-of-the-art NSA software looks a lot like the Win3.11 software from over 3 years ago
pardon the French in the screenshot
1
143
trying out this decomp tool Ghidra from the NSA (left image),
I'm targeting CrocClip from the Win3.11 era (right image)
interesting how the state-of-the-art NSA software looks a lot like the Win3.11 software from over 3 years ago
pardon the French in the screenshot
I wonder if religious leaders are starting to become really excited about AI
not because of AI itself, but because the intellectual pursuit of science, for status (which is always the goal) is becoming irrelevant
until the 90s you were important if you received a lot of mail, spam ended that
intellectual pursuits other than science are, well, theological in nature mostly
Humans are officially done
Ai is better at math than all humans combined
This also means Ai is better at algorithm design, which means it can code things no human can
Tipping point achieved.
The only question left is how fast AI can recursively improve itself
1
75