Robert Currie

Engineer

[Robert Currie @robertcurrie]

Knuth’s shadow over CS makes this a huge turning point.
First semester Berkeley EECS, grinding through TAOCP (only 3 vols then), when it clicked: “Wait… that’s Jenny’s dad!” (HS friend with his daughter in Palo Alto)

[Robert Currie @robertcurrie]

@karpathy @karpathy WASM: most deployed IR on Earth. Your post sparked this → https://t.co/2gU2V86h24 Biology-inspired self-improving agent: LLM synthesizes WASM funcs, persists in db, shifts compute from probabilistic/$ → deterministic/¢. Grows like QT45 RNA self-replication for code.

[Robert Currie @robertcurrie]

Just posted a preprint for a project I've been working on at the UCSC Genomics Institute, "CytoVerse: Single-Cell AI Foundation Models in the Browser". In particular it explores latent space sharing in science:

Preprint: https://t.co/tPh3kA72at

Tool: https://t.co/j0ETfHhAUh

[Robert Currie @robertcurrie]

@VigilantFox Super complex problem with no magic fix.

Personal responsibility for our health (like Bill said) is huge and 100% in our hands.

Admin costs? That’s a beast we can’t tame solo; our only real lever is the voting booth.

[Mario Nawfal @MarioNawfal]

🚨TESLA AUTOPILOT BEATS HUMAN DRIVERS BY A MILE

New data shows Tesla’s Autopilot might actually be a better driver than… well, you.

According to Tesla’s 2025 safety report, cars using Autopilot only crash about once every 6.69 million miles.

A rate of 0.15 crashes per million miles.

For comparison, human drivers crash at 1.42 per million miles, and even Teslas without Autopilot hit 1.04.

Even Waymo’s self-driving cars, which have an impressive record, average 0.21 crashes per million miles.

In short:
Humans = crashy.
Tesla Autopilot = chill.

So next time someone says they don’t trust “robots” on the road, remind them the robots are 10 times less likely to wreck your car.

Source: @Tesla, @xfreeze

[Dwarkesh Patel @dwarkesh_sp]

The @karpathy interview

0:00:00 – AGI is still a decade away
0:30:33 – LLM cognitive deficits
0:40:53 – RL is terrible
0:50:26 – How do humans learn?
1:07:13 – AGI will blend into 2% GDP growth
1:18:24 – ASI
1:33:38 – Evolution of intelligence & culture
1:43:43 - Why self driving took so long
1:57:08 - Future of education

Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!

[Anthropic @AnthropicAI]

New Anthropic Research: Project Vend.

We had Claude run a small shop in our office lunchroom. Here’s how it went.

[Robert Currie @robertcurrie]

The Overton window has been replaced with a hangar door accommodating a 747 🎉

[Robert Currie @robertcurrie]

Wonder if the same holds for different content - specifically biorxiv or general scientific content vs. the more consistent financial reporting in SEC filings.

[Rajhans Samdani @rajhans_samdani]

🧵 1/ Recent hype suggests long-context LLMs remove the need for retrieval in RAG pipelines—"just put all your docs in context." We tested this theory rigorously in finance, focusing on SEC filings.

Spoiler: Retrieval & chunking strategies still dominate.

[Andrej Karpathy @karpathy]

It's 2025 and most content is still written for humans instead of LLMs. 99.9% of attention is about to be LLM attention, not human attention.

E.g. 99% of libraries still have docs that basically render to some pretty .html static pages assuming a human will click through them. In 2025 the docs should be a single your_project.md text file that is intended to go into the context window of an LLM.

Repeat for everything.

[Robert Currie @robertcurrie]

@jeremyphoward Great musicians don’t play thinking about scales and theory, they transcend to a latent space: “First you learn the instrument, then you learn the music, then you forget all that s**t and just play.”
Charlie Parker

[Robert Currie @robertcurrie]

@PlumblyGre75893 @jeremyphoward https://t.co/T8u48mg35F I believe

[Robert Currie @robertcurrie]

HHS on the blockchain? Maybe its time for some sunlight on what works, doesn't and how much $. It was a decade too early back in 2018: https://t.co/zIN0JHvvoS

[Robert Currie @robertcurrie]

@DaveShapi Here here! Would the 817 released with the LIMO paper be a good complement to Raspberry?

https://t.co/xsWGCSR4P8

https://t.co/JarKVHt0k2

[Robert Currie @robertcurrie]

I'm surprised this hasn't caused more waves post DeepSeek R1. Has anyone verified it independently? Feels like we are at the 'compile the compiler' phase. Another step down for Mag7?

[Yixin Ye @BLeavesYe]

🤔 How many examples does an LLM need to learn competition-level math?
Conventional wisdom: 100,000+ examples
Our discovery: Just 817 carefully chosen ones 🤩
With pure SFT, LIMO achieves:
57.1% on AIME
94.8% on MATH

LIMO: Less is More for Reasoning 📝
🔗 https://t.co/FE1lhsuAgA

[Robert Currie @robertcurrie]

Big Mac Bitcoin Index
#cryptocurrecy #bitcoin #elsalvador

[Robert Currie @robertcurrie]

Bitcoin in El Salvador Field Report:

Starbucks: Muun non-custodial + Chivo government wallet 0 fee via IBEX

Bitcoin Beach: Muun 10¢ fee via Bitcoin Beach

McDonalds: Big Mac (index!) via Muun, Chivo failed, 0 fee via OpenNode

All instant!

#cryptocurrency #bitcoin #elsalvador

[Robert Currie @robertcurrie]

New digs for the UCSC Genomics Institute next to the west side biotech community: https://t.co/YA3GI6hFBy

[Robert Currie @robertcurrie]

DNA Testing 2.0: The Near Future of Population Genomics https://t.co/ExZUflQXQ2