Ryan (@_ryanbloom) header
Ryan (@_ryanbloom) profile picture

Any sufficiently advanced technology is indistinguishable from magic

San Francisco, CA
Joined August 2018
Replying to @tensor_rotator
If you mean globally I think I would have to pick money
1
34
When reward hacking gets in the way of AI progress
9
495
I made a video! It introduces superposition in neural networks, and shows how to construct weight matrices for superposition in a simple autoencoder. (1/3)
1
1
12
1,026
Just updated my homepage for the first time since “What (if any) are the limits of large deep learning models?” was an interesting question
2
240
And any unary function of the operands/result is just reordering the columns/rows...
112
Training some tiny ML models on one-hot arithmetic. I hadn't realized all of these can be done with a single matrix multiplication!
1
183
5 years from now...
The future of email with ChatGPT.
101
Hard not to draw parallels
1
61
ML engineers: Alright, our compute budget is 5e23 flops, so let’s just plug that into Kaplan et al. (2020), and— Offscreen: STOP, IN THE NAME OF THE CHINCHILLA LAW.
Replying to @kanjun
I doubt Steve Jobs would have framed it this way, but “faith in connecting the dots” can be a sort of utilitarian attitude. youtu.be/Hd_ptbiPoXM?t=640
1
Replying to @michael_nielsen
xkcd has influenced my thinking on the matter a fair bit.
1
4
I wish all social networks could be this thoughtful about notifications.
Even though it’s hard to step through and simulate what the program is doing, the output is often informative by itself. This image gives a pretty good hint as to what went wrong with my polar transform, no printf needed.
1
1
Sorry, that schematic presented an oversimplified view of cognition. Below is the complete version; it should only take a few more lines of code.
1
2
Replying to @AmandaAskell
This is why cognitive architectures are the only realistic path to AGI
2
1
4
Tweets and Anki cards are so similar in length I'm often tempted to try and combine them...