a @yihyunCS

schizo-shitposter the child asked me if god exists. i said 'not yet.' prev @nasa yihyun.ca Seoul Joined February 2024

Tweets

5K
Followers

2K
Following

4K
Likes

14K

kache @yacineMTB

5 days ago

*yet*

137 182 4K 262K 1K

Download Video

0 0 3 165 0

Alexander Doria @Dorialexander

2 weeks ago

Well sorry to insist, but the catch is that the market is still small — that's less than chewing gum industry. Too small for a bubble.

12 4 62 4K 16

Download Image

a @yihyunCS

2 weeks ago

can someone tell me if new grok code model is worth even using

0 0 2 111 0

gemini is the best model purely due to its ability to be critical and evaluate things with lower bias, some models (OAIs in particular) are vulnerable to agreeing with the user even when incorrect about an issue, gemini seems to not fall into this as much (deepseek too)

1 0 3 156 0

a @yihyunCS

2 weeks ago

erm claude pookie?

0 0 3 142 0

Download Image

a @yihyunCS

2 weeks ago

claude is cute 'this issue is persistent!'

0 0 3 117 0

a @yihyunCS

2 weeks ago

they should’ve called it GPT-V but meh wtv

1 0 5 134 0

a @yihyunCS

2 weeks ago

beginning to wonder what percentage of the industry works at Meta now…

0 0 4 115 0

a @yihyunCS

2 weeks ago

it’s a matter of time before we get something definitively better than transformers in every area, it will happen someday

0 0 5 110 0

a @yihyunCS

3 weeks ago

how i look at deepseeks hugging face profile waiting for something, the whale is ghosting us

0 0 2 74 0

Download Image

a @yihyunCS

3 weeks ago

theres a well known person on my tl all the time and everytime i see their posts i just think the dude is such a narcissist not even because HES something, but because he knows people that are something maybe that didn't make sense, but iykyk

2 0 14 491 0

kalomaze @kalomaze

3 weeks ago

working on a somewhat novel approach that is able to adapt qwen0.5b from basically not understanding a format at all, to mostly getting the structure right, just through rejecting synthetic feedback against a target distribution of desirable data working name: "RL-ASF"

8 3 146 14K 62

Download Image

a @yihyunCS

4 weeks ago

why WOULD they classify their own site as unsafe? its easier to just do if url is not openai. com - mark as unsafe its very simple and actually harmless, it could've been done better for sure but it really didn't need to, the added complexity isn't worth the outcome