Lightnews — Scholar-powered news

LightNews

max 🌌

@sneptech.bsky.social

530 followers 450 following 2.9K posts

the future is a nation we will become citizens of together founder & frontend dev @ Dreamtime

Posts Media Videos Starter Packs

Pinned

max 🌌 @sneptech.bsky.social · Nov 21

I've got an idea for commercially viable mass asteroid mining that, roughly condensed in four words is: SOLAR BULK CENTRIFUGE PEEL MINING

laying out the system design in a thread 👇

(artist's conception by @CalebABingham on 🐦🖥️)

max 🌌 @sneptech.bsky.social · 12h

nice try

max 🌌 @sneptech.bsky.social · 12h

max 🌌 @sneptech.bsky.social · 13h

max 🌌 @sneptech.bsky.social · 14h

and his High Showeriness presided over the greatest golden age in all of human history up until that point

* technologically and economically anyway. socially some stuff was left to fix

max 🌌 @sneptech.bsky.social · 15h

funnily enough I had that exact same argument with someone and their argument boiled down to "nuh uh deflation is fine because bitcoin is deflationary and bitcoin is good"

ideally you'd want something that expands with use algorithmically, early experiments like Ampleforth kinda tried doing that

max 🌌 @sneptech.bsky.social · 15h

song is www.youtube.com/watch?v=hdM2...

YouTube video by NeurofunkGrid

www.youtube.com

max 🌌 @sneptech.bsky.social · 17h

max 🌌 @sneptech.bsky.social · 18h

all my flowers gone

max 🌌 @sneptech.bsky.social · 18h

max 🌌 @sneptech.bsky.social · 18h

using the hot shower index alone predicted the regression of Germany 10 years ago as prices for power exploded and the culture shifted to treat those who would want such decadence as immoral

max 🌌 @sneptech.bsky.social · 18h

"but it uses lots of water" -> massive desal expansion, recycling
"that's a lot of energy" -> heat exchangers, nuclear-maxxing
"what about work" -> automation revolution, UBI, resource-based economy
etc etc.

the index is that you can measure the success of a civ based on length of shower too

max 🌌 @sneptech.bsky.social · 18h

years ago i went on a rant about how if your only driving force was "everybody should have a hot shower that lasts indefinitely if they want it to" you can reliably assemble a gameplan for utopian civilization from that alone

max 🌌 @sneptech.bsky.social · 18h

the moonboi hypothesis runs in direct opposition to this because anything that looks problematic to lawmakers hurts line going up

max 🌌 @sneptech.bsky.social · 18h

the best usecase for cryptocurrency was and continues to be being able to pay for things without a bureaucrat standing in your way going "no"

Reposted by max 🌌

S.M. Pritchard @smpritchard.bsky.social · Sep 30

A cold rocky world languishes in the feeble glow of a brown dwarf, itself part of a distant binary pair whose partner can just be seen as a small magenta speck just over its limb. Perhaps a view like this exists around the real-life brown dwarf binary Luhman-16, the closest brown dwarfs to Sol.

An image depicting a rocky landscape bathed in deep red light. In the center, a brown dwarf dominates the scene, with alternating bands of incandescent cloud layers glowing various shades of magenta, red, and orange. From our vantage point the cloud bands are oriented roughly vertically. Over the north-eastern limb of the brown dwarf, a distant second companion is visible only as a tiny magenta dot.

max 🌌 @sneptech.bsky.social · 19h

goddamn

max 🌌 @sneptech.bsky.social · 20h

internet poisoning

Reposted by max 🌌

Louie Zong @everydaylouie.bsky.social · 21h

manul

a stylized 3d render of a large manul sitting on a georgian orthodox church.

max 🌌 @sneptech.bsky.social · 20h

www.youtube.com/watch?v=LJ5n...

Eric Andre Let me in

YouTube video by Flashervee7

www.youtube.com

max 🌌 @sneptech.bsky.social · 20h

cough it up sir
www.youtube.com/watch?v=MwSl...

Give Him To Me Auhhh 😫 (HQ)

YouTube video by Dakota Wolf

www.youtube.com

max 🌌 @sneptech.bsky.social · 20h

finally i can pursue my true calling in life. going triple platinum on facebook reels

Reposted by max 🌌

stellz @piss.beauty · 1d

solved my laptop being cold in the morning in a dumb way

@piss.beauty/heater

CPUs can make a little heat, as a treat

max 🌌 @sneptech.bsky.social · 1d

and then came fucking Nixon
bsky.app/profile/snep...

max 🌌 @sneptech.bsky.social · Sep 4

i have +infinity hatred towards Richard Nixon because if he didn't cancel the MSRE solely because he wanted to look good in his home state (where the LMFBR was) then 9/11 and all that subsequent horseshit would probably not have happened

yes. Nixon being a piece of shit Butterfly Effect'd 9/11.

SwiftOnSecurity @swiftonsecurity.com · Sep 4

"Hey what if humanity had infinite energy in the 70's instead of inventing global ecocide and dictatorial petrostates and century-long populous displacement"

max 🌌 @sneptech.bsky.social · 1d

AFAIK the MSRE didn't do continuous online chemical reprocessing where they removed fission products but I could be wrong

Reposted by max 🌌

Tim Kellogg @timkellogg.me · 1d

Karpathy: nanochat

A small training+inference pipeline for creating your own LLM from scratch

$100 will get you a somewhat functional model

$1000 is more coherent & solves math

detailed walkthrough: github.com/karpathy/nan...

repo: github.com/karpathy/nan...

Andrej Karpathy & @karpathy
X.com
Excited to release new repo: nanochat! (it's among the most unhinged I've written).
Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web Ul.
It weighs ~8,000 lines of imo quite clean code to:
- Train the tokenizer using a new Rust implementation
- Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics
- Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use.
- SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval)
- RL the model optionally on GSM8K with
IPDDOI

- RL the model optionally on GSM8K with
"GRPO"
- Efficient inference the model in an Engine with
KV cache, simple prefill/ decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUl.
- Write a single markdown report card, summarizing and gamifying the whole thing.
Even for as low as ~$100 in cost (~4 hours on an
8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions.
About ~12 hours surpasses GPT-2 CORE metric.
As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and
70s on ARC-Easy, 20s on GSM8K, etc.
My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow

developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved.
Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.
nanochat