vessenes

Aww! Hi! I missed that it was you commenting originally. I have been mulling over whether or not you could train on an elliptic curve. So, totally different direction. Let’s catch up - you sniped me back. I’ll drop you a note.

Can we pay $4 to have the AI only build the software and not the marketing materials?

I wasn't pro truck until I lived on an island with limited infrastructure. Once you need to take care of some of your own infra, a truck is a great tool, beating your full list. That said, a beater truck is generally preferred -- you want reliable, cheap and something that you don't care if it gets banged up.

Isn't that kind of the dream of Slate? I reserved two, btw, hard agree.

This was beautiful to read. Thank you for writing it! Don’t worry about it reading like slop - your voice comes through very distinctly to me.

You mention a lot of ethics concerns; I am totally uneducated on medical ethics but I think those are pretty important questions. It doesn’t sound like you’ve had a lot of support from this startup.

I might suggest to you that you publish this on a website - GitHub pages comes to mind as something Claude or codex could do for you easily - it would let you segment this information in a way that someone who wanted to help with the medical questions could get right to that stuff, and someone who wanted to help with ethics could also jump to what they were interested in.

Either way, thank you for the write up! Your curious and open attitude is lovely.

I agree ZKP is the wrong approach. I also agree most people don’t care.

I’d love to read a plan that lets someone trust a pile of weights that comes with some sort of provenance proof. I’ll happily call it and you brilliant even if it was really easy to write and come up with. All my ideas for it look really difficult.

Thank you! I will review. Super interesting.

Please refrain from ad hominem attacks here on HN. Or really, anywhere.

Sort of a Sophie's World for Crypto. I'm sure this would be a treasured experience.

This is an excellent summary of the state of the frontier in March, 2026. It was frustrating. However. Things continue to improve, shockingly quickly, at the frontier. These are not major concerns I have with Opus 5.5 or Astra or Sol 6.1.

There's a lot of fundamental science to be done in space; some forms of crystal are much easier to grow, chemical reactions and biological interactions etc. etc. For reference, it wasn't long ago NASA was trying to figure out if spiders could form webs in space.

Not a book but you could construct a little walk through some historical crypto as fun games to play. Caesar cipher, Vigenere cipher, one time pad, then learn about primes and make some repeating keys..you could do a super simple matrix example to make tiny DES s-boxes. I bet claude could come up with some games / fun little exploration things to do together, and you and your son could do a lot of fun motivating math together. You could anchor it with rewards like pictures of you and your partner or some family history or stuff that feels like a treasure hunt.

Mine are out of the house, but I had a few who might have really liked doing something like this.. could make a nice Christmas present, the first mysterious letter..

I've been writing about the need for provable inference (and training) on and off for a number of years; this is welcome news and a good start, and a significant improvement on prior zk results.

The problems of inference proof for a closed model provider are more difficult than an open model provider -- you can get a probabilistic proof much more cheaply out of four independent inference providers running the same non-batched prompt than this 90-100x compute hit.

Anthropic and others also have a significant problem with these proofs in that they have a lot of incentive internally to have some portions of the pre-prompt secret; I haven't read this newest scheme closely enough to determine if they have a way to hide parts of the prompt with the proof. But if they don't, they are almost certainly not going to want to send these proofs out in the general case.

Last, I'll take this moment to soapbox and say for the researchers out there -- provable inference is nice. Provable training corpus information would be spectacular. It would be very, very, very nice to provide a set of model weights and a proof that they had only been trained on a dataset. To be clear, this is a really hard problem, but it would change fundamentally what levels of trust we could have about a given set of weights.

Awesome. "At $1/hr for my time, I spent the equivalent of 4 replacement computers" :)

It's been that way forever here, don't worry. Signal to noise is lower today, but it's the same vibes, trust me. That said, earlier this year I got really frustrated, and put out https://github.com/vessenes/hn1k/, which lets you build a local hn restricted to 1k / 10k / 100k commenters (and threads they engage in) which will massively de-ragebait you.

Sounds like you live in Europe! Bring me more trains pls

It absolutely was a net positive. On the infrastructure side we got datacenters built and tons of dark fiber put down. At the start of the boom we had only 2G cellular for goodness sake. I think you could argue we wouldn't have necessarily had the rise in cloud compute without some of the 'overbuilding' of datacenters in web 1.

For reference this 'massive build-out' resulted in, according to a Berkeley study I found, 500MW of total datacenter power in the US across 320 locations at end of 2020. This wasn't too much for very long: as of 2005, according to Tier1 there were 511 datacenters. And it was likely only really a year or two: Dupont's VA3 utilization numbers published it at 50% utilization in 2003, 89.4% in 2004, then full thereafter.

Meanwhile, these infrastructure improvements allowed, for example, things like the iPhone to be built and deployable -- there was 3g, there were a LOT more web devs that could create the early (web only) apps the iPhone needed to compete. This in turn allowed another round of investment to be built on top of all of that.

On the capital side, investors were also able to buy Amazon at around $1 / share. AAPL at iPhone launch was $7/share. I haven't split adjusted either of these, this is just from memory.

I don't think a bust was necessarily good but the boom undeniably built out infrastructure, a small fraction of what the world is currently using, and did it pretty quickly, and then post bust that infra got bought out in a capital efficient manner. I think that's the bear case here.

Put another way, the Dotcom one boom was wrong about two things: timing was too fast, and the predicted scale of change was way, way, way too small. They undersold the future.

As I've been pointing out a little bit recently -- the 'downside' scenario -- somehow we conclude we have enough inference at our world's desired quality and therefore demand stabilizes below that of these datacenters -- would be GREAT for the world. Super cheap intelligence for all.

If some lenders go bankrupt on a datacenter note post-buildout and the capital structure is reworked, lowering the costs by removing capitalization costs from the cashflow of the inference provider, that is all to the good of people who did not do the initial lending.

I think that is extremely unlikely, but it's good for datacenter doomers to think through these scenarios IMO.

Data centers in northern climates (often so chosen for cold average temperatures) do sometimes pipe heat around. Basically datacenter spend = house warming. This works better with town / high density living or work setups than suburbs as you might imagine. So -- yes! Good idea.

I'm reminded of the human genome project here - in that case there was a geometric ramp-up that meant we went from 1 or 2% sequenced to 100% really quickly. This sort of doubling when you can get it is pretty magical.

I'd like to imagine that we'll get something similar here sometime in the next ten years -- I understand there are (tens of?) thousands of these scrolls still to be scanned much less unrolled. Just an incredibly cool thing to be doing.

I love the advocacy angle here - it's super well executed. However

> SINCE THE EARLY 2010s, QUILICURA HAS BECOME A MAJOR HUB FOR A.I. DATA CENTERS THAT USE LARGE AMOUNTS OF WATER TO COOL DOWN SERVERS. BECAUSE OF THAT, COUPLED WITH AN ONGOING CLIMATE CRISIS, CHILE IS FACING A SEVERE DROUGHT EXPECTED TO LAST UNTIL 2040.

I do not believe that data centers (Were there AI datacenters in 2010? No) are causing a severe drought in Quilicura that will extend until 2040. I believe climate change could be causing such a drought, and I believe a drought is a terrible thing.

Estimates of water used for cooling are about 0.3mL of water per chatbot query; the 25,000 prompts that this really awesome and fun project answered resulted in 7.2 liters of water "saved" from use as cooling. I would imagine these 40 people drank significantly more than that during the day.

I put "saved" in quotes because no water is destroyed during datacenter cooling. Some (the parts sent to open evaporative coolers) goes back in the atmosphere, and I imagine disperses over a pretty wide region. The rest of it is recycled locally, whether into the water table, or I would guess more likely back into cooling another round of queries.

Anyway, it's fun. But I guess I'd prefer something that is more honest about the actual uses and cycles here -- it doesn't do a local community (or a global one) any good to misinform them about the situation; to their point the situation locally sounds like it could be pretty dire. Ultimately I think it risks falling between performative and exploitative.

Okay, this was gratifying to read, but ... can't you wedge CarPlay onto this thing? I'd want a victory lap with root doing at least something. Bad apple on car boot, there's lots of options here!

Also I note from your comments here you don’t plan on local inference - that seems like a huge self own, provided the model is less than like 2b params. I’m not saying people won’t in practice prefer your cloud version, but you have this huge uphill battle ahead of you on privacy which can be wiped at the stroke of a pen with “yep it will run on your Mac.”

You can get both markets (ease of use and technical) by allowing it. Also i would bet that you gain almost no product security locking off local inference - the hardware is the hard bit. Once someone has your hardware in hand it will not take them long to reverse and build their own processing layer; you’ll need to fight those people on patents and product quality.

Heck for a v1 just allowing local sensor data would probably be enough for the tech enthusiasts. But please don’t lock this off from the get go - it makes the sale harder and I don’t think it gives you much in return.

The Gemma task is harder: add features. Qwen was: fix this looping behavior

It’s a good reminder of one of the most unique features of alphago - it was like a seven layer or so neural net that could predict out a hundred moves. I’m not sure we took this kind of compression super seriously or at least I would say perhaps it remains under explored. You could think of what you did as sort of the same thing as skipping the Monte Carlo search phase for a go engine — a process compression.

Good points - I’m still shook

Yes - “opus 5 helped train 5.5” is extremely cool but not shocking. Hand editing weights without tooling to change outcomes is shocking.

A single forward pass here Bette 10mm+ PPO rounds. You are hard to impress.

Super cool. For a cool voice change demo we'll allow a gb200 over low latency wifi :) elevenlabs has deals with a number of professional actors and voice actors for licensed uses, seems like you could probably find an ethical and cool co-brand opportunity in there.

Have fun building this! It's worth doing.

This is insane.

Astra can, without tooling, determine what games small networks are trained on, can edit the weights of networks directly to improve outcomes, and can hand edit a small qwen model to improve its math skills by like 5x.

I'm shook. If you'd asked me this morning whether LLMs have native understanding of their weights that is less than human <-> neuron understanding, the same as human or more than human, I would have said "same to less" depending on definitions.

No human can go edit a brain to become better at math. This is just very, very surprising to me and extremely interesting. I would guess this upsets quite a number of interp theories and philosophical takes on what's going on with these networks.