Wednesday, January 29

Geek

Daily News Stuff 29 January 2025

Pink And Blue Edition

Top Story

  • A lot of stuff is being written about Chinese AI DeepSeek right now, and most of it is probably wrong.  Somehow The Verge seems to have been skeptical where skepticism was appropriate for once.  (The Verge)  (archive site)
    It took about a month for the finance world to start freaking out about DeepSeek, but when it did, it took more than half a trillion dollars - or one entire Stargate - off Nvidia’s market cap. It wasn’t just Nvidia, either: Tesla, Google, Amazon, and Microsoft tanked.
    This is of course true.  The sky-high valuations were irrational, and the drop was also irrational.
    Even if critics are correct and DeepSeek isn’t being truthful about what GPUs it has on hand (napkin math suggests the optimization techniques used means they are being truthful), it won't take long for the open-source community to find out, according to Hugging Face's head of research, Leandro von Werra. His team started working over the weekend to replicate and open-source the R1 recipe, and once researchers can create their own version of the model, "we’re going to find out pretty quickly if numbers add up."
    DeepSeek claims 100x improvements in training efficiency, but its published papers are full of micro-optimisations, which do not create 100x performance gains.
    There are some people who are skeptical that DeepSeek's achievements were done in the way described. "We question the notion that its feats were done without the use of advanced GPUs to fine tune it and/or build the underlying LLMs the final model is based on," says Citi analyst Atif Malik in a research note. "It seems categorically false that 'China duplicated OpenAI for $5M' and we don’t think it really bears further discussion," says Bernstein analyst Stacy Rasgon in her own note.
    My take as well.  DeepSeek did some useful work, and they published it.  But there are very good reasons to believe that they didn't do everything they said - such as the fact that on release, DeepSeek was convinced it was ChatGPT.


Tech News


Musical Interlude



Disclaimer: Ring ring!  Banana milk!

Posted by: Pixy Misa at 05:55 PM | Comments (1) | Add Comment | Trackbacks (Suck)
Post contains 519 words, total size 5 kb.

1 "But it would still be a little surprising to see AMD buck the naming conventions it's stuck with up until now."

Eh, what?  Are they talking about some mustachioed, mirror-universe AMD?
Also: "But circling back however although maybe it would still sometimes just be little bit slightly surprising to see AMD buck the naming conventions it's stuck with up until this moment right now today of course."

Posted by: normal at Wednesday, January 29 2025 09:23 PM (bg2DR)

Hide Comments | Add Comment




Apple pies are delicious. But never mind apple pies. What colour is a green orange?




Save
Bold
Italic
Underline
Strikethrough
Superscript
Subscript
Foreground Color
Background Color
Hyperlink
Special Characters
Undo
Redo
View/Edit Source
 

52kb generated in CPU 0.0132, elapsed 0.1076 seconds.
58 queries taking 0.0977 seconds, 350 records returned.
Powered by Minx 1.1.6c-pink.