puffybsd
@puffybsd
Fan of Clojure, F#, Prolog and Rust. Linux Fan. Fan of Deep Learning and AI. Fan of AVR programming. LaTeX and MD Fan. Neovim Fan. Keyboard Fan. Arch btw.
คุณอาจชื่นชอบ
TIL: the tmux command to have what you type sent to all panes: setw synchronize-panes on/off
TIL: AM5 cpus have igpus, including the 9950x. Not everyone knows this, including AI. Spent a few hours debugging wayland+kde instability. Disabling igpu support in the bios fixed it.
Don't sleep on mt3 /dev/tty PBT keycaps with everglide oreo switches. Except they're from a drop 5 years ago. Best combo.
Slightly annoying that people referring to the training phase as the pretraining phase. Pretrained models were previously trained during the training phase.
No problems with it really, but the barrier between building a superior gpu airflow adapter in freecad is... wayland
May be overkill, but would like to model airflow for a combo fan mount and intake redirect to target GPUs. Good case but airflow could be better. 3d printing a fan mount that targets GPU + redirects some other intakes. looking for airflow modeling software. @TheAhmadOsman
Also not fully sure of the claude code z.ai glm4.x model. Was using it, then it requests funds and hasn't reset, so maybe just trial credits? Vs counters resetting daily?
Qwen code has suffered a weird edit problem recently where it just stopped succeeding at making edits "Failed to edit, 0 occurrences". Love qwen and qwen code. Not sure why this is the case, but it seems that it can't stop doing this.
It will take your nvidia training environment up to 10 mins to start: "This process can take up to 10 minutes depending on your course requirements." ouch.
Entirely too challenging for Gemini: - confused labels - wrong shaped curves - terrible label placement Had to: - have it use canvas - write a list of labels from originals - repeated placement (broke out the wacom to annotate) - runs in canvas, can't see(needed export to colab)
There is idris2 and idris2-git on arch: both perform long compiles. What is this, gentoo or bsd?
It's possible, just possible my interests are too diffuse and I've been spinning wheels not focusing enough.
Dgx spark is like a professional portable amp for a musician, and music listeners are complaining it gets bad radio reception.
Watching Karpathy interview, thinking of Hebbian learning and attention, in-context learning as short term memory and residual connections. Conversations with chat is a good way to explore ideas. Looking back at fast weights, hopfield networks, associative memory.
DGX Spark is the ultimate Dunning-Krueger Machine. Even people reviewing it seem to miss the point.
It is probably obvious, but TRM is not a general model. It's demonstrating something that might not be gleaned from the headlines at first glance.
It's cool to see that sparse auto encoders are so useful, and feel like the dual of vaes. Instead of bottleneck to generalize, expand to organize.
United States เทรนด์
- 1. Cheney 157K posts
- 2. Colts 15.9K posts
- 3. Election Day 142K posts
- 4. Jakobi Meyers 6,201 posts
- 5. Mamdani 605K posts
- 6. Shaheed 4,960 posts
- 7. Chris Ballard N/A
- 8. #weloveyouari 2,868 posts
- 9. Welcome to Indy N/A
- 10. #csm219 4,695 posts
- 11. Iraq 60.1K posts
- 12. Cuomo 296K posts
- 13. TWO FIRSTS N/A
- 14. Shota 18.2K posts
- 15. Waddle 6,659 posts
- 16. New Jersey 223K posts
- 17. #いい推しの日 756K posts
- 18. Logan Wilson 9,034 posts
- 19. Quinnen Williams 1,478 posts
- 20. GO VOTE 109K posts
คุณอาจชื่นชอบ
-
John K. Waters
@johnkwaters -
Savo
@savorocks -
Marco Giovanni Marini
@saidone -
Maris
@Marutks -
bsky @s-exp.com
@mpenet -
Gusty
@gmpl -
Kimmo Koskinen
@KimmoKoskinen -
Alex Dixon
@a_dixon -
FIORELLA DOINO
@DoinoFiorella -
إِسْرَ𓂆ء
@EsraaTharwat13 -
Bizpartners.cl
@bizpartners_cl -
Daniel Glauser
@danielglauser -
Aaron Brooks
@0x1B -
Claudiu Apetrei
@claudiuapetrei -
dev
@TheDeveshShetty
Something went wrong.
Something went wrong.