Bread @BreadPirateRob

baking Joined February 2021

Tweets

230
Followers

37
Following

101
Likes

1K

Bread @BreadPirateRob

3 weeks ago

this man was right the people don't want APIs, they want a box w/500gb VRAM and no setup

0 0 0 15 0

Download Image

Bread @BreadPirateRob

2 months ago

Apple is messing w/their bluetooth protocol

0 0 0 130 0

llm reasoning and thinking is an action that belongs to the action space, e.g., sequential thinking mcp. forget about all of the other anti patterns like react. those won’t scale. decomposition ftw. from my blog post from last november.

2 5 32 4K 21

Download Image

ℏεsam @Hesamation

3 months ago

Ilya’s top 30 paper recommendation is actually cracked, i knew about it but i just got a chance to skim the titles and dude! it has cracked papers by G. Hinton, Karpathy, Chris Olah, covering nlp, vision, rnns, lstm, fundamental deep learning, and so much more.

16 255 3K 191K 4K

Download Image

Bread @BreadPirateRob

3 months ago

Ilya Sutskever compiled a list of 30 papers. “If you really learn all of these, you’ll know 90% of what matters” And, yeah. Wow. If you haven't read ALL of these, take the time. aman.ai/primers/ai/top…

0 0 0 84 0

Bread @BreadPirateRob

3 months ago

Gemini Diffusion is shockingly fast. So fast, it opens a realm of new use-case applications. An obvious leap in architecture. Nice work @bodonoghue85

0 0 1 88 0

Bread @BreadPirateRob

3 months ago

there’s a race to add a continuous layer to LLMs the silence is deafening

0 0 0 68 0

Bread @BreadPirateRob

4 months ago

Why can't Claude's UI fork conversations?

0 0 0 59 0

Andrej Karpathy @karpathy

4 months ago

We're missing (at least one) major paradigm for LLM learning. Not sure what to call it, possibly it has a name - system prompt learning? Pretraining is for knowledge. Finetuning (SL/RL) is for habitual behavior. Both of these involve a change in parameters but a lot of human…