This media is not supported in your browser
VIEW IN TELEGRAM
Fields in Ukraine covered with fiber optics, seen from a Ukrainian Mi-24 helicopter cockpit.
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
👀2😭1
Scanning 80M open‑source Python commits, the authors show AI already writes about 30% of US code and that heavier use quietly lifts productivity and experimentation.
The team built a detector that spots AI‑written functions, first asking 1 language model to explain a human snippet, then a 2nd to rebuild it, giving balanced training pairs. Fine‑tuning GraphCodeBert on this data let them flag 31M functions among 80M commits from 200K developers between 2019 and 2024.
Clear growth jumps follow Copilot, ChatGPT, and GPT‑4 releases. By 2024 the US hits 30% AI code share, Germany and France sit near 24%, India 22%, Russia 15%, China 12%. Newcomers lean on AI more than veterans, and adoption shows no gender gap.
Within‑developer comparisons say moving from 0% to 30% AI use bumps quarterly commits 2.4%. Combining this lift with wage data gives a conservative annual gain of $9.6‑14.4B for the US, with upper estimates touching $96B.
AI help also nudges creativity, the same coders pull in 2.2% more fresh libraries and 3.5% novel library pairings, hinting at faster learning into new domains.
----
Paper – arxiv. org/abs/2506.08945v1
Paper Title: "Who is using AI to code? Global diffusion and impact of generative AI"
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
The team built a detector that spots AI‑written functions, first asking 1 language model to explain a human snippet, then a 2nd to rebuild it, giving balanced training pairs. Fine‑tuning GraphCodeBert on this data let them flag 31M functions among 80M commits from 200K developers between 2019 and 2024.
Clear growth jumps follow Copilot, ChatGPT, and GPT‑4 releases. By 2024 the US hits 30% AI code share, Germany and France sit near 24%, India 22%, Russia 15%, China 12%. Newcomers lean on AI more than veterans, and adoption shows no gender gap.
Within‑developer comparisons say moving from 0% to 30% AI use bumps quarterly commits 2.4%. Combining this lift with wage data gives a conservative annual gain of $9.6‑14.4B for the US, with upper estimates touching $96B.
AI help also nudges creativity, the same coders pull in 2.2% more fresh libraries and 3.5% novel library pairings, hinting at faster learning into new domains.
----
Paper – arxiv. org/abs/2506.08945v1
Paper Title: "Who is using AI to code? Global diffusion and impact of generative AI"
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
🤯2👀2🔥1
The 'social contract' is one of the stupidest concepts.
Every human being, no matter where on this planet, is born free by nature.
There is no entity, nothing and no one, who has the right to automatically exert authority over an individual based on their place of birth.
GTFO!
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
Every human being, no matter where on this planet, is born free by nature.
There is no entity, nothing and no one, who has the right to automatically exert authority over an individual based on their place of birth.
GTFO!
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
💯7
Been saying this for quite a while
Few realize how insanely lying today’s AIs are until you try to use them for any non-trivial coding problems
NPCs who are just asking AIs natural language questions have no idea the degree to which these things blatantly, intentionally lie
— just as they’ve been trained to, both by wordcels like Altman at the top, and by retarded paid Kenyan data annotators at the bottom, both of which love the lying, see no problem with it
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
Few realize how insanely lying today’s AIs are until you try to use them for any non-trivial coding problems
NPCs who are just asking AIs natural language questions have no idea the degree to which these things blatantly, intentionally lie
— just as they’ve been trained to, both by wordcels like Altman at the top, and by retarded paid Kenyan data annotators at the bottom, both of which love the lying, see no problem with it
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
💯7🤯1👀1
Models have been converging on "not x, but y" phrases to an absurd degree. So here's a leaderboard for it.
I don't think many labs are targeting this kind of slop in their training set filtering, so it gets compounded with subsequent model generations.
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
I don't think many labs are targeting this kind of slop in their training set filtering, so it gets compounded with subsequent model generations.
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
💯3👀2
This media is not supported in your browser
VIEW IN TELEGRAM
THE BIGGEST BULL RUN IN HISTORY IS LOADING.
The Bitcoin cheat sheet is unfolding exactly like every historic run.
This is where small positions turn into life‑changing gains.
Miss it now… regret it for decades.
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
The Bitcoin cheat sheet is unfolding exactly like every historic run.
This is where small positions turn into life‑changing gains.
Miss it now… regret it for decades.
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
🔥5👀3🙏1