Christian-H.eth (๐Ÿ’™,๐Ÿงก) ๐Ÿ›ธ (๐Ÿ“ฝ๏ธ, ๐ŸŒฟ)

352 posts

Christian-H.eth (๐Ÿ’™,๐Ÿงก) ๐Ÿ›ธ (๐Ÿ“ฝ๏ธ, ๐ŸŒฟ)

Christian-H.eth (๐Ÿ’™,๐Ÿงก) ๐Ÿ›ธ (๐Ÿ“ฝ๏ธ, ๐ŸŒฟ)

@ChristianHCode

Currently learning web3 & Solidity | JS | Python | โค๏ธ Open Source | empathy & respect are ๐Ÿ”‘

๊ฐ€์ž…์ผ Ekim 2022
919 ํŒ”๋กœ์ž‰56 ํŒ”๋กœ์›Œ
Christian-H.eth (๐Ÿ’™,๐Ÿงก) ๐Ÿ›ธ (๐Ÿ“ฝ๏ธ, ๐ŸŒฟ) ๋ฆฌํŠธ์œ—ํ•จ
Chubbyโ™จ๏ธ
Chubbyโ™จ๏ธ@kimmonismusยท
MYTHOS BENCHMARKS, OFFICIAL. HOLY MOLY Anthropic cooked!!
Chubbyโ™จ๏ธ tweet media
English
110
191
2.6K
446.2K
Christian-H.eth (๐Ÿ’™,๐Ÿงก) ๐Ÿ›ธ (๐Ÿ“ฝ๏ธ, ๐ŸŒฟ) ๋ฆฌํŠธ์œ—ํ•จ
Ziming Liu
Ziming Liu@ZimingLiu11ยท
Drifting generative models are so fun! I explain the physical interpretation and discuss why anti-symmetry is needed. Lots of fun visualizations in today's blog. The kernel is indeed tricky to tune for my toy cases. More study is needed, but definitely an exciting direction!
Ziming Liu tweet media
English
11
72
714
34.9K
Adam Hibble
Adam Hibble@Algomancerยท
So, pushforward in this context is the distribution you get by sampling and passing it through the generator. It's just "what distribution do the outputs live in?". In flows and diffusion, this is what you get after one step of iteration at inference. I think an intuitive way to think of this paper is that, because training itself is iterative, ie, training gives you a series of generators as you update the gradients, the push forward evolves over training, ie, the parameters implicitly 'drift' each sample. So, the formulation basically says, rather than iterate the push forward at inference, like diffusion or flow matching does when you sample via there respective ode/sde, we effectively push that into train time. The drifting field tells each generated sample where to move; when V -> 0 everywhere, the pushforward matches the data distribution and training has converged. For this to work in high dimension they need a really good measure of similarity, which is why they measure that in feature space. I love explaining things, so feel free to spam me questions here, or in dms.
English
2
0
22
1K
Adam Hibble
Adam Hibble@Algomancerยท
Drifting is all you need. Spent the morning going through the paper. Here is my minimum implimentation I made while working through the paper. github.com/Algomancer/Minโ€ฆ It's an excellent formulation, and fits so nicely into so many things. This will be an important paper.
English
15
34
399
29.4K
Christian-H.eth (๐Ÿ’™,๐Ÿงก) ๐Ÿ›ธ (๐Ÿ“ฝ๏ธ, ๐ŸŒฟ) ๋ฆฌํŠธ์œ—ํ•จ
ๆœบๅ™จไน‹ๅฟƒ JIQIZHIXIN
New paradigm from Kaiming He's team: Drifting Models! With this approach, you can generate a perfect image in a single step. The team trains a "drifting field" that smoothly moves samples toward equilibrium with the real data distribution. The result? A one-step generator that sets a new SOTA on ImageNet 256x256, beating complex multi-step models.
ๆœบๅ™จไน‹ๅฟƒ JIQIZHIXIN tweet media
English
15
163
1.3K
319.4K
Christian-H.eth (๐Ÿ’™,๐Ÿงก) ๐Ÿ›ธ (๐Ÿ“ฝ๏ธ, ๐ŸŒฟ) ๋ฆฌํŠธ์œ—ํ•จ
Zephyr
Zephyr@zephyr_z9ยท
Very interesting
Zephyr tweet media
English
4
26
173
35.1K
Teortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž)
He literally did just that, and he will literally launch space datacenters, and it will literally bring us towards Kardashev Type II civilization; very straightforwardly so. Believing People is actually a strong baseline. The utility of cynicism is very overrated by ๅฐไบบ.
Teortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž) tweet media
Unit Accord@unit_accord

@teortaxesTex Do you know what's insane cope? To believe he merged Twitter, Grok and SpaceX so he can make space datacenters that bring us to Kardashev Type II civilization, which is literally the mental illness in their merger press release.

English
10
11
238
10.1K
Financelot
Financelot@FinanceLancelotยท
Did Sam Altman commit the biggest theft in world history? Elon Musk finally got revenge today by bringing SpaceX into xAI. All investment liquidity will be redirected away from OpenAI into xAI, leaving Sam with nothing A 5D chess move to destroy Sam & Elon made the final blow๐Ÿ”ฅ
Financelot@FinanceLancelot

Anyone else think Jensen Huang looks extremely uncomfortable when asked about pulling the plug on $100B for Sam Altman's OpenAI? "There never was a commitment. They invited us to invest up to $100B. You keep putting words in my mouth." Who's going to buy $NVDA's inventory now?

English
139
479
5K
2.3M
Teortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž)
I have a system to my hatreds. It seems completely rational to me, but people often get confused and interpret it as something weird like face-saving or ethnic pride or whatever. For example, I do not forgive "faking to have read a source". I'd flog people for it. how dare you?!
Teortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž) tweet mediaTeortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž) tweet media
Teortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž)@teortaxesTex

@EliasMingos Can you not fucking read? Did you just skim and smugly assume that he agrees with you? I hate people like you.

English
2
0
24
4K
Teortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž)
Insane, mentally ill cope. Inferior men blinded by the light of revealed superiority and writhing in ressentiment. I remove such followers. Thankfully, in due time they'll be removed period. If that's your idea of what Elon is about, you're ngmi.
Teortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž) tweet media
Unit Accord@unit_accord

@teortaxesTex His power is only evidence of how corrupted our systems are. You can do whatever you want with infinite P/E when your stock performance is completely unhinged from your performance, by simply constantly spamming sci-fi nonsense of which 0.01% ever happens by random chance.

English
26
27
513
28.3K
Sebastian Heyneman
Sebastian Heyneman@s_heynemanยท
Garry, browser is solved (playwright mcp). mobile is harder. we built a local server that wraps xcrun simctl. it pipes the simulator screenshots into the agent loop so it can see the css breaks. otherwise it misses the visual bugs the dom tree hides. happy to send you the repo.
English
17
0
60
10.1K
Garry Tan
Garry Tan@garrytanยท
Is there any way I can connect Claude Code to use my browser and iOS simulator so it can find mobile css bugs and make things responsive on tablet, web, and mobile automatically?
English
196
19
762
166.1K
kat kampf
kat kampf@kat_kampfยท
We started internal testing some big updates to the @GoogleAIStudio experience today! Coming to you early next year but reply below if youโ€™d like early access in the coming weeks ๐Ÿ‘€
English
3.1K
127
3.7K
308.3K
Logan Kilpatrick
Logan Kilpatrick@OfficialLoganKยท
Reply here or DM me :) will add folks in as much as we can
English
2.2K
13
941
78.6K
Logan Kilpatrick
Logan Kilpatrick@OfficialLoganKยท
Big upgrade to vibe coding in @GoogleAIStudio lands in Jan, but if you want to test earlyโ€ฆ ๐Ÿ‘‡๐Ÿป
English
3.8K
190
5.5K
553.7K
Zephyr
Zephyr@zephyr_z9ยท
This doesn't change anything No point comparing a next-gen model with the current crop Claude 5, Grok 5, GPT's 5 update will be a fair comparison
strictrope@strictrope

@zephyr_z9 Best model is no longer on Nvidia GPUs, but on TPUs ๐Ÿค”

English
10
2
110
14.8K
Bojan
Bojan@bjnpckยท
Itโ€™s pretty clear that @base has won the L2 race. @arbitrum will stick around as itโ€™s perceived as the most Ethereum aligned one. The rest is dead. Not what I want, but itโ€™s the truth.
English
129
17
442
59.1K