• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
  • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    ArXiv Writeup Produced Using Gemini and Claude

    Academic retweets announcement of detailed arXiv writeup made with AI models

    ND
    PM
    KD
    4 Sources, 30d ago, first seen 30d ago

    TLDR

    Peyman Milanfar announced a more detailed writeup posted to arXiv. The message credits help from Gemini and Claude in its creation. Kosta Derpanis, associate professor at York University focused on computer vision, retweeted the post. The announcement supplies the arXiv link but no further details on the writeup's topic or findings. The evidence shows only this public post and its retweet, with no independent corroboration or replies visible in the packet.

    Combined views

    51.6K

    4 Sources, first seen 30d ago

    Combined views

    51.6K

    4 Sources, first seen 30d ago

    496 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    496 likes
    3 comments
    594 saves
    63 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    3 comments
    594 saves
    63 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    4 Sources

    @PackBropagatedReally good mathematical explanation for - diffusion is slow because you're hitting a moving target. - Flow matching is trying to draw straight paths so the target is still hence faster. - But straight paths can cross in n-dim space and you will have 2 images at the target which is not directly optimizable to we do average or any combining op which again bends the paths but it reduces the number of steps. - Solution to intersecting paths is Reflow which says just train on the paths which dont intersect by first sampling paths which dont intersect. - You can take much bigger steps in that case thus reducing steps to find the reverse mapping.
    @NandoDFRT @PackBropagated: Really good mathematical explanation for - diffusion is slow because you're hitting a moving target. - Flow matching…
    @docmilanfarHere is a more detailed writeup I've put on arXiv: https://arxiv.org/abs/2609.00198 - made with help from Gemini and Claude
    @CSProfKGDRT @docmilanfar: Here is a more detailed writeup I've put on arXiv: https://arxiv.org/abs/2609.00198 - made with help from Gemini and Claude

    4 Sources

    @PackBropagatedReally good mathematical explanation for - diffusion is slow because you're hitting a moving target. - Flow matching is trying to draw straight paths so the target is still hence faster. - But straight paths can cross in n-dim space and you will have 2 images at the target which is not directly optimizable to we do average or any combining op which again bends the paths but it reduces the number of steps. - Solution to intersecting paths is Reflow which says just train on the paths which dont intersect by first sampling paths which dont intersect. - You can take much bigger steps in that case thus reducing steps to find the reverse mapping.
    @NandoDFRT @PackBropagated: Really good mathematical explanation for - diffusion is slow because you're hitting a moving target. - Flow matching…
    @docmilanfarHere is a more detailed writeup I've put on arXiv: https://arxiv.org/abs/2609.00198 - made with help from Gemini and Claude
    @CSProfKGDRT @docmilanfar: Here is a more detailed writeup I've put on arXiv: https://arxiv.org/abs/2609.00198 - made with help from Gemini and Claude