简体中文
English Indonesia Türkçe Deutsch Español Português Pусский Français Italiano Nederlands Polski العربية ไทย Tiếng Việt 繁體中文 简体中文 日本語 한국어
soboleffspaces's profile picture. 𝗦𝗰𝗵𝗼𝗹𝗮𝗿/𝗔𝘂𝘁𝗵𝗼𝗿/𝗧𝗲𝗮𝗰𝗵𝗲𝗿 • causality in plain language •
Boris Sobolev
@soboleffspaces
年10月14日

While bureaucrats and the “concerned citizenry” panic about chatbots, LLMs are busy doing what they can’t — making us more productive. Here’s a figure Overleaf generated from LaTeX code that Gemini 2.5 Pro writes from my verbal (3 sentences) description of this NCM architecture!…

soboleffspaces's tweet image. While bureaucrats and the “concerned citizenry” panic about chatbots, LLMs are busy doing what they can’t — making us more productive. Here’s a figure Overleaf generated from LaTeX code that Gemini 2.5 Pro writes from my verbal (3 sentences) description of this NCM architecture!…
soboleffspaces's tweet image. While bureaucrats and the “concerned citizenry” panic about chatbots, LLMs are busy doing what they can’t — making us more productive. Here’s a figure Overleaf generated from LaTeX code that Gemini 2.5 Pro writes from my verbal (3 sentences) description of this NCM architecture!…
soboleffspaces's tweet image. While bureaucrats and the “concerned citizenry” panic about chatbots, LLMs are busy doing what they can’t — making us more productive. Here’s a figure Overleaf generated from LaTeX code that Gemini 2.5 Pro writes from my verbal (3 sentences) description of this NCM architecture!…
1
0
6
2
875
murat_kocaoglu_'s profile picture. Asst. Prof. at Johns Hopkins CS. Research on causal inference, causal discovery, generative AI, info theory, online learning.
Murat Kocaoglu
 @murat_kocaoglu_
年10月14日

Nice! Yes, they are definitely useful here. Not sure what prompt you used, but I noticed latent U confounds all three variables, and the trained model will not accurately represent true interventional samples since they are non-identifiable in this graph. Perhaps an LLM mistake?

1
0
1
1
332
soboleffspaces's profile picture. 𝗦𝗰𝗵𝗼𝗹𝗮𝗿/𝗔𝘂𝘁𝗵𝗼𝗿/𝗧𝗲𝗮𝗰𝗵𝗲𝗿 • causality in plain language •
Boris Sobolev
@soboleffspaces
年10月14日

glad you noticed. that’s by intent, though; that’s how I asked. we can condition on a global confounder U, the study-unit variable, and identify an ITE, right? but, conventinally, all U_i=g_i(U) are sampled from independent distributions. If you have time next week,…

0
0
1
0
99
United States 趋势

Something went wrong.


Something went wrong.