levzzz

160 posts

levzzz

levzzz

@levzzz5154

Katılım Temmuz 2024
78 Takip Edilen4 Takipçiler
levzzz
levzzz@levzzz5154·
@cheatyyyy @teortaxesTex drawthings is such a joke lol they still group samplers and schedulers together like it's 2023 slower than comfy too
English
0
0
0
3
cheaty
cheaty@cheatyyyy·
@teortaxesTex the fuck did they use drawthings for also flux.1 dev at q8 would be at a minimum 5x bigger with the text encoders and vae included
English
1
0
0
123
levzzz
levzzz@levzzz5154·
@Sauers_ first model to have 100k context even if it was partly fake if only we could get claude 1.2 the mad poet back
English
0
0
2
150
Sauers
Sauers@Sauers_·
Fun fact: the first Claude in spring of 2022 was only deployed to certain partners, similar to Mythos right now. The first public Claude came out a year later
Sauers tweet media
English
5
6
198
9.2K
advaith
advaith@advaithj1·
Discord Linux users, must be your lucky day: you won't be seeing this anymore, the app now auto-updates! it's also now available in rpm and pkg.tar.zst formats, in addition to deb and tar.gz
advaith tweet media
English
94
183
3.4K
109.7K
Blake Izayoi
Blake Izayoi@IzayoiBlake·
@minionmomUWU I've 1cced 1, 6 (twice), 7, 8, 9, 19 (three times) and FDF so far! They're all good fun.
English
2
0
1
29
Blake Izayoi
Blake Izayoi@IzayoiBlake·
Holy crap!! I 1cced on my first time playing Marisa shot type B. The last part was so intense because I had no bombs, no lives... but somehow I got it!! Hip hip hooray!!
Blake Izayoi tweet mediaBlake Izayoi tweet media
English
12
3
78
1.2K
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
Many are saying V4 (speaking purely of Pro) is strange. It is often unreliable (hallucinates, falls apart) and often rock solid. At its best, it's way more "Frontier" than other 1T class Chinese open models. And yes, *good* at debugging. It's like something not fully merged.
Michael Anti@mranti

因为小米的API赠送,我尝试着用mimo-v2.5-pro接Claude Code了几天,体验总结是:能干基本的活和编程,但要debug的时候,就不如Deepseek V4 pro了(我现在把这个模型当成了benchmark)。所以,我还是切换回Deepseek了。

English
14
0
125
12.5K
levzzz
levzzz@levzzz5154·
@Tommywebster77 Since we already got the fighting games, extending abilities to 3d doesn't seem like much of a problem
English
0
0
1
118
Theweb77
Theweb77@Tommywebster77·
Touhou Hero shooter and what role the character would be. Feel free to comment and share your thoughts.
Theweb77 tweet media
English
20
11
139
4.9K
BNelsey (comms clogged 4/3)
This week's non-Alice meme is characters that are popular for very specific reasons!
BNelsey (comms clogged 4/3) tweet media
English
19
60
659
8.9K
Remy
Remy@Remihime7·
making up a guy who only refers to touhou games by the translation of their japanese title
English
22
48
733
13.5K
levzzz
levzzz@levzzz5154·
@SergioOSINT @MTSlive possible to just include everything besides thinking and RL the actual thinking process may be more beneficial too but i dunno i'm no LLM researcher
English
0
0
0
114
Sergio
Sergio@SergioOSINT·
@levzzz5154 @MTSlive Mostly when you mean distilling its taking full trajectories including thoughts
English
1
0
1
117
MTS
MTS@MTSlive·
LIVE TRIAL UPDATE: OpenAI's counsel asked Musk whether xAI has ever "distilled" technology from OpenAI. Musk: "Generally AI companies distill other AI companies." "Is that a yes?" Savitt asked. Musk: "Partly."
English
38
52
1.7K
314.7K
Seven Stars Messenger
Seven Stars Messenger@7starsmessenger·
A search through GNSK-5.5's SFT data found many datapoints containing "youkai" and "oni". Further investigation revealed a whole family of other odd creatures: tanuki, tengu, kappa and kitsune were identified as other tic words, while most uses of kami turned out to be legitimate
English
1
5
22
612
levzzz
levzzz@levzzz5154·
@UpgradeFanboy still friends with zun (created lots of 500+ year old characters with non-matching looks)
English
0
1
21
1.9K
levzzz
levzzz@levzzz5154·
@lyncpk @catgirlprostate *variable length sequences are a big problem conv can't consider the entire sequence at once rnn/lstm has a fixed size memory and bottlenecked in performance it's all transformers now
English
0
0
1
51
levzzz
levzzz@levzzz5154·
@lyncpk @catgirlprostate wtf are you on? transformers ARE an architecture for ML? i'm assuming you don't know any AI/ML theory so in short it's all ml, you're trying to create difference where there just isn't any and nearly every single model for anything nowadays is a transformer other arches are dead*
English
1
0
3
148
Anotherthem🇨🇺
Anotherthem🇨🇺@EnbyCreature999·
I hope Deltarune's final boss references another Touhou villain Not made by Toby but UNDERTALE Yellow's final boss Ceroba is also basically just Junko, completely accidental as well, just a coincidence
Anotherthem🇨🇺 tweet mediaAnotherthem🇨🇺 tweet media
Cüneyt@Cuneyt_S_Sevi

Does he know

English
10
25
239
5.2K