hoanh
@hoanhle_learning and creating new things, under @jaakkolehtinen’s lab, core contributor of a few things running on million devices
Similar User
@Fabel38564808
@HectorM93178597
@unhingedorca
@Robert38636577
@selfimprooover
@MichaelRPatters
@BoztekeMustafa
@UgurAltin16
@TheChukaa
@larkywind
@vividverbiage
@HuaweiMate6
@K_Hoff6626
@moonNbeyond
@ben_filloon
Excited to share a new article, From Transformers to ChatGPT In it, I go back in time, starting with the invention of the transformer model by @ashVaswani, and leading up to the release of ChatGPT by @OpenAI last year hoanhle.github.io/blog/2023/chat…
just realized doing a phd is a multi-year exercise in speculative decoding.. and i am the draft model every week i generate a sequence which may be many tokens long. however all tokens are not always correct. that’s why i have an expert model (phd advisor) to correct my mistakes
23/ But, by following the gradient of interestingness, you're increasing your chances of stumbling upon success because:
I think a lot about how Feynman got his Nobel Prize from “piddling around with wobbly plates”. I’m always looking for wobbly plates to piddle with. And of course the point isn’t to get the Prize, but to enjoy the piddling
jira? sprints? what are you talking about dave? get over here and check out this neat optimization carmack figured out
> simplify the problem as much as possible. > add a debugger and step through the code to ensure it works exactly as expected. > gradually introduce more complexity. every time i skip this process, i end up wasting more time flailing around.
When I retrospectively think about all the globally successful projects I worked on, the common denominator wasn't the buy-in from everyone. A few strongly opinionated people came together, identified a major problem, built a solution extensible enough. Growth was organic from…
I resonated with Leonardo da Vinci's ideas so I read stuff by him ~every night for like two or three years. I'm glad I didn't diversify, I never ran out of stuff and after a while you have so much context, hard to get deep context if you're always reading different people
United States Trends
- 1. Bengals 71 B posts
- 2. Chargers 58,9 B posts
- 3. McPherson 10,4 B posts
- 4. Herbert 31,8 B posts
- 5. Joe Burrow 17,1 B posts
- 6. #CINvsLAC 9.738 posts
- 7. #BoltUp 4.809 posts
- 8. Zac Taylor 3.837 posts
- 9. #BaddiesMidwest 17,5 B posts
- 10. #SNFonNBC N/A
- 11. JK Dobbins 3.947 posts
- 12. Money Mac N/A
- 13. Chiefs 154 B posts
- 14. #WhoDey 1.705 posts
- 15. Harbaugh 14,5 B posts
- 16. Ladd 5.133 posts
- 17. Tee Higgins 3.516 posts
- 18. WWIII 152 B posts
- 19. Josh Allen 66,7 B posts
- 20. Russia 741 B posts
Who to follow
-
Fabel
@Fabel38564808 -
COINHUNT.X
@HectorM93178597 -
unhinged orca
@unhingedorca -
Robert
@Robert38636577 -
selfimprooover
@selfimprooover -
Mike
@MichaelRPatters -
Mustafa 1907
@BoztekeMustafa -
Ugur Altin
@UgurAltin16 -
TheChuka
@TheChukaa -
Fairy Sakana
@larkywind -
DoubleEntendres
@vividverbiage -
Baseball Bonk Inu
@HuaweiMate6 -
K-Hoff
@K_Hoff6626 -
Max 👾
@moonNbeyond -
Ben Filloon
@ben_filloon
Something went wrong.
Something went wrong.