Words, code, truth, kindness, other arcana ∙ https://xitter.cf/t.co/sRhJ3En7qD
SF
Joined December 2013
- Tweets8.2K
- Following5.6K
- Followers5.4K
- Likes35.9K
Pinned Tweet
The danger is not in letting students use tools like ChatGPT. It is in letting them learn that they don’t have to think, or believing that the tools are smarter than they are.
I think research like this is very cool for three reasons.
Firstly, it illustrates that we are moving toward a direction of observational science when it comes to complex mathematical patterns.
I believe a new frontier in mathematical sciences and computational biology will arise soon and it will start by having deeper explorations of simple systems with complex behavior. A few people, notably @drmichaellevin, have been pursuing this vision for years but it will become more obvious soon.
Secondly, in the 19th century and earlier it was much more common to have “citizen scientists” finding results at the frontier of human knowledge. I think in purely theoretical subfields, where agents work best, we're moving back to a world a little like that.
Another example of this is work of dozens of people exploring square packing (including me and in fact this is why I gave myself math psychosis this summer!). Anyone with a coding agent can explore new mathematical ideas.
Thirdly, this illustrates how the initial narrative that the surge in notable AI proofs eliminates mathematics as a human pursuit is completely backwards.
We are bottlenecked in some areas of science not by the ability to do experiments, but by the ability to ask new questions. As more people have the freedom to ask questions and do experiments, it leads to a renaissance in understanding.
Conway’s Game of Life has gliders that travel at 45° and spaceships that move horizontally or vertically. Patterns that travel at other angles are called “oblique spaceships.”
By extending rule-switching cellular automata, I’ve discovered a large number of new oblique spaceship structures - with diverse shapes and directions, from nearly horizontal to almost 45°.
The nine examples in this video are just a small selection. Explore the discoveries, including motions that may not have been reported before, through interactive demos:
Joshua Levy retweeted
i’m spending the weekend reading about the science and theory of consciousness. plz reply with recommendations! dms open also
Joshua Levy retweeted
WTF, $3000 Samsung TV has an always-on mic, automatic content recognition, geo tracking, network mapping, and ships your data to Samsung + 400 partners.
Bro Rooted a brand-new 2025 Samsung TV after noticing absurdly targeted ads.
Firmware was locked down, so zero-days it was.
Got a root shell.
then found: always-on voice service, ACR, geo, full telemetry pipeline, and four separate DRM systems.
This isn’t a bug, It’s the product bro.
To be honest, I'm surprised when people are surprised by this reaction.
I live at AI ground zero in San Francisco. I am in the top 1% in terms of agent use. *And* I have more optimism for its positive impact long term then most people I meet.
But let's be honest. For the past three years the main narrative Silicon Valley has promoted is:
“AI is powerful beyond your comprehension, rapidly making us rich as it is taking your job, and experts say there is a chance it might kill humanity.”
If you wanted to explain the real value of AI to normal people, it is hard to imagine of a more confusing and counterproductive way to do it.
I hope It is clear that I mean this in a constructive way.
What I am trying to say is that I think if you're building with AI for anyone outside of Silicon Valley, it is wise develop empathy for people’s concerns. And explain the value (and the mitigations for its negative effects) better.
People really do change their minds. People all over the world will use more and more AI products. But dismissing people as ignorant is rarely an effective way to begin an explanation.
As you delegate more and more to your coding agents, remember there is still one weird trick that only you can do properly:
Deciding which files on your constantly full drive are safe to delete.
💾 Disk Saving Tip #1:
I just saved 200gb 😮
1. Remove old Docker containers (80gb)
2. Purge old pnpm stores (50gb)
3. Remove old emulators - iOS + Android (70gb)
But seriously this is such a big problem you should install my fdu skill for it.
I'm pleased to announce the release of fdu, the world's fastest disk usage and file tree analytics CLI+API.
Over the past few months, I found myself managing numerous worktrees with hundreds of thousands of files. Agents struggle to tally disk usage, file age, or know where to begin to clean up large folders.
fdu is an ground-up rewrite of both du and more modern tools like dust.
It's pre-built and distributed via uv so you can run it right now with `uvx fdu@latest`.
Thanks to an agentic performance iteration loop, it is the fastest tool of its kind, typically walking and tallying 5-10GB/s or 100-200K files/s on my old MacBook. That is ~9x faster than du and 10-100% faster than other modern native/Rust/Go options.
It's instant to install as a skill or use as a CLI. Or can be used as a Python or Rust API.
Other features: colored terminal output, tree views, last modification time for every directory (recursive), gitignore awareness, document analysis, and programming language detection (with code/comment/blank line tallies), filename and age filtering, native directory watching, a file cache for faster reruns, and parsable outputs like YAML and JSONL.
All my agents now use the fdu to disk usage. Give it a try and let me know if it's useful! Or what you'd like to see in it next.
Joshua Levy retweeted
"Picking archetypal song lyrics as a glimpse into the heart or soul of a place"
Very into this. I'd love to see an interactive audio exhibit where you travel the world through songs, listening to the poets and musicians who have best articulated the soul of each city. A musical atlas of the world’s collective psyche!!
Noticing the wild swing in my Twitter feed’s p(doom) again today makes me want to tap the sign.
Take a breath.
Stay humble when making predictions.
But never underestimate human ingenuity.
Go do good work.
Hug your family.
It’ll be okay.
More
Well worth checking out this article from @Mark_J_Perry about how expert consensus can be spectacularly wrong.
This is not to dismiss all predictions of doom. But rather to encourage healthy skepticism and humility when it comes to predictions.
aei.org/carpe-diem/18-specta…
My friends Evan and Babur are setting up this event in Nov. I think it is interesting because it might be a glimpse of the future.
So there are obvious analogies of buying compute today with financial markets and with traditional cloud compute marketplaces.
But I think the dynamics of the market are quite different because AI compute is so new (we don’t know all the form factors), so scarce, and so opaque.
As crazy as AI buildout may seem already, the demand for compute is almost insatiable: yes, there is speculation and over-use of AI by some early adopters, but fundamentally, most consumers and enterprises are still only getting started with using AI.
Tokens are going to be scarce for a while. Few people have visibility on the market. Prices wildly vary. A lot of the digital economy will rest on compute deals and how they are brokered.
Introducing the Haight Street Compute Market.
A one-day trading floor where vetted buyers and sellers of compute meet and sign deals in person. Nov 17, at the old Federal Reserve building in San Francisco.
Request your invite: luma.com/86cjwpui
