Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
|
from
login
The Weirdness of LLM Tokenization
(
twitter.com/karpathy
)
2 points
by
tosh
1 day ago
|
past
|
discuss
Jagged Intelligence
(
twitter.com/karpathy
)
2 points
by
mellosouls
1 day ago
|
past
|
discuss
Andrej Karpathy: "LLM model size competition is intensifying backwards
(
twitter.com/karpathy
)
13 points
by
bilsbie
8 days ago
|
past
|
discuss
I am starting an AI+Education company
(
twitter.com/karpathy
)
915 points
by
bilsbie
10 days ago
|
past
|
540 comments
The if-then-else monster
(
twitter.com/karpathy
)
4 points
by
tosh
16 days ago
|
past
[flagged]
Andrej Karpathy on X: "100% Software 2.0 computer.Just a single neural net
(
twitter.com/karpathy
)
25 points
by
bilsbie
26 days ago
|
past
|
23 comments
One built-in UI/UX feature of LLM interfaces I'd love is proof
(
twitter.com/karpathy
)
1 point
by
bilsbie
35 days ago
|
past
These 94 lines of code are everything that is needed to train a neural network
(
twitter.com/karpathy
)
3 points
by
r_singh
35 days ago
|
past
[Andrej Karpathy] Let's reproduce GPT-2, in PyTorch from scratch (nanoGPT)
(
twitter.com/karpathy
)
7 points
by
_giorgio_
46 days ago
|
past
Let's reproduce GPT-2 (124M)
(
twitter.com/karpathy
)
5 points
by
Multiset
47 days ago
|
past
FineWeb-Edu: High quality LLM dataset
(
twitter.com/karpathy
)
3 points
by
tosh
54 days ago
|
past
I had ~30 direct reports and didn't do 1on1s at Tesla and imo it was great
(
twitter.com/karpathy
)
1 point
by
tosh
56 days ago
|
past
|
2 comments
CUDA/C++ origins of Deep Learning
(
twitter.com/karpathy
)
4 points
by
tosh
84 days ago
|
past
llm.c: multi-GPU, bfloat16, flash attention, ~7% faster than PyTorch
(
twitter.com/karpathy
)
121 points
by
tosh
84 days ago
|
past
|
10 comments
LLMs must one day run in Space
(
twitter.com/karpathy
)
6 points
by
tosh
84 days ago
|
past
|
1 comment
Llm.c Update
(
twitter.com/karpathy
)
31 points
by
ibobev
3 months ago
|
past
Karpathy on Llama 3
(
twitter.com/karpathy
)
12 points
by
tosh
3 months ago
|
past
Consider Being a Labeler for an LLM
(
twitter.com/karpathy
)
2 points
by
tosh
3 months ago
|
past
Scheduling Workloads to Run on Humans
(
twitter.com/karpathy
)
2 points
by
tosh
3 months ago
|
past
llm.c is now down to 26.2ms/iteration, matching PyTorch
(
twitter.com/karpathy
)
46 points
by
tosh
3 months ago
|
past
|
8 comments
Llm.c is only 2X slower than PyTorch (fp32, forward pass)
(
twitter.com/karpathy
)
2 points
by
tosh
3 months ago
|
past
Andrej Karpathy explaining llm.c in layman terms
(
twitter.com/karpathy
)
1 point
by
ibobev
3 months ago
|
past
Karpathy: Explaining llm.c in layman terms [tweet]
(
twitter.com/karpathy
)
9 points
by
mellosouls
3 months ago
|
past
LLM training in simple, pure C/CUDA
(
twitter.com/karpathy
)
4 points
by
theaniketmaurya
3 months ago
|
past
|
1 comment
Automating Software Engineering
(
twitter.com/karpathy
)
4 points
by
quick_brown_fox
4 months ago
|
past
Andrej Karpathy on automating software engineering
(
twitter.com/karpathy
)
5 points
by
hubraumhugo
4 months ago
|
past
Love Letter to Obsidian
(
twitter.com/karpathy
)
69 points
by
tosh
5 months ago
|
past
|
40 comments
A Deepdive into the Gemma Tokenizer
(
twitter.com/karpathy
)
2 points
by
tosh
5 months ago
|
past
A lot of weird behaviors and problems of LLMs trace back to tokenization
(
twitter.com/karpathy
)
30 points
by
stared
5 months ago
|
past
|
8 comments
Karpathy on Tokenizers
(
twitter.com/karpathy
)
2 points
by
drcwpl
5 months ago
|
past
|
1 comment
More
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search: