• progress_activity cloud_sync

    Reconnection to the server…

    Movim cannot talk with the server, please try again later

  • back_to_tab fullscreen tile_small dialpad mic videocam switch_camera screen_share

    mic_none No sound detected from your microphone


    • ArsTechnica


      article 31785 posts • people 689 subscribers
      assignment_ind Only publishers can publish


    • Public subscriptions

    • chevron_right

      coopr8

    • chevron_right

      gabagoo

    • chevron_right

      kenu_demon

    • chevron_right

      coopr8

    • chevron_right

      gabagoo

    • chevron_right

      kenu_demon

    • chevron_right

      coopr8

    • chevron_right

      gabagoo

    • chevron_right

      kenu_demon

  • Register Login

    Movim

    movim.chatterboxtown.us


  • rss_feed
    add Follow

    ArsTechnica

    people 689 subscribers • news.movim.eu

    • chevron_right

      Researchers isolate memorization from reasoning in AI neural networks

      news.movim.eu / ArsTechnica • 10 November 2025

    When engineers build AI language models like GPT-5 from training data, at least two major processing features emerge: memorization (reciting exact text they’ve seen before, like famous quotes or passages from books) and reasoning (solving new problems using general principles). New research from AI startup Goodfire.ai provides the first potentially clear evidence that these different functions actually work through completely separate neural pathways in the model’s architecture.

    The researchers discovered that this separation proves remarkably clean. In a preprint paper released in late October, they described that when they removed the memorization pathways, models lost 97 percent of their ability to recite training data verbatim but kept nearly all their “logical reasoning” ability intact.

    For example, at layer 22 in Allen Institute for AI’s OLMo-7B language model, the bottom 50 percent of weight components showed 23 percent higher activation on memorized data, while the top 10 percent showed 26 percent higher activation on general, non-memorized text. This mechanistic split enabled the researchers to surgically remove memorization while preserving other capabilities.

    Read full article

    Comments

    • tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
    • chevron_right

      Researchers isolate memorization from reasoning in AI neural networks

      news.movim.eu / ArsTechnica • 10 November 2025

    When engineers build AI language models like GPT-5 from training data, at least two major processing features emerge: memorization (reciting exact text they’ve seen before, like famous quotes or passages from books) and reasoning (solving new problems using general principles). New research from AI startup Goodfire.ai provides the first potentially clear evidence that these different functions actually work through completely separate neural pathways in the model’s architecture.

    The researchers discovered that this separation proves remarkably clean. In a preprint paper released in late October, they described that when they removed the memorization pathways, models lost 97 percent of their ability to recite training data verbatim but kept nearly all their “logical reasoning” ability intact.

    For example, at layer 22 in Allen Institute for AI’s OLMo-7B language model, the bottom 50 percent of weight components showed 23 percent higher activation on memorized data, while the top 10 percent showed 26 percent higher activation on general, non-memorized text. This mechanistic split enabled the researchers to surgically remove memorization while preserving other capabilities.

    Read full article

    Comments

    • tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
    • chevron_right

      Researchers isolate memorization from reasoning in AI neural networks

      news.movim.eu / ArsTechnica • 10 November 2025

    When engineers build AI language models like GPT-5 from training data, at least two major processing features emerge: memorization (reciting exact text they’ve seen before, like famous quotes or passages from books) and reasoning (solving new problems using general principles). New research from AI startup Goodfire.ai provides the first potentially clear evidence that these different functions actually work through completely separate neural pathways in the model’s architecture.

    The researchers discovered that this separation proves remarkably clean. In a preprint paper released in late October, they described that when they removed the memorization pathways, models lost 97 percent of their ability to recite training data verbatim but kept nearly all their “logical reasoning” ability intact.

    For example, at layer 22 in Allen Institute for AI’s OLMo-7B language model, the bottom 50 percent of weight components showed 23 percent higher activation on memorized data, while the top 10 percent showed 26 percent higher activation on general, non-memorized text. This mechanistic split enabled the researchers to surgically remove memorization while preserving other capabilities.

    Read full article

    Comments

    • tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices tagai tagai tagai tagbiz & it tagbiz & it tagbiz & it tagai alignment tagai alignment tagai alignment tagai architecture tagai architecture tagai architecture tagai behavior tagai behavior tagai behavior tagai memorization tagai memorization tagai memorization tagai research tagai research tagai research tagai safety tagai safety tagai safety tagallen institute for ai tagallen institute for ai tagallen institute for ai tagcomputational neuroscience tagcomputational neuroscience tagcomputational neuroscience tagcopyright tagcopyright tagcopyright taggeneralization taggeneralization taggeneralization taggoodfire taggoodfire taggoodfire taggradient descent taggradient descent taggradient descent tagk-fac tagk-fac tagk-fac tagloss curvature tagloss curvature tagloss curvature tagmachine learning tagmachine learning tagmachine learning tagmechanistic interpretability tagmechanistic interpretability tagmechanistic interpretability tagmodel editing tagmodel editing tagmodel editing tagneural networks tagneural networks tagneural networks tagolmo tagolmo tagolmo tagoverfitting tagoverfitting tagoverfitting tagtransformer models tagtransformer models tagtransformer models tagweight matrices tagweight matrices tagweight matrices

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
    • chevron_right

      Canada fought measles and measles won; virus now endemic after 1998 elimination

      news.movim.eu / ArsTechnica • 10 November 2025

    Canada has lost its measles elimination status, meaning the highly infectious virus is considered endemic once again in the country, The Pan American Health Organization (PAHO) announced Monday.

    The determination was made by a committee of PAHO experts, who spent last week poring over disease data to assess the measles status of countries across the entire region. The fact that Canada has lost its elimination status means that the region of the Americas overall has also lost the status, which it achieved in 2016. Of the 35 countries and territories in the region—a health region designated by the World Health Organization—Canada is currently the only country where measles is considered to be spreading endemically, though other countries, namely the US and Mexico, are headed in the same direction.

    Measles is considered eliminated when a country can go 12 months without continuous local spread. Sporadic cases brought in from international travel can continue to occur, potentially causing limited outbreaks. But elimination is lost and endemicity is declared only when transmission is sustained over the course of a year.

    Read full article

    Comments

    • taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
    • chevron_right

      Canada fought measles and measles won; virus now endemic after 1998 elimination

      news.movim.eu / ArsTechnica • 10 November 2025

    Canada has lost its measles elimination status, meaning the highly infectious virus is considered endemic once again in the country, The Pan American Health Organization (PAHO) announced Monday.

    The determination was made by a committee of PAHO experts, who spent last week poring over disease data to assess the measles status of countries across the entire region. The fact that Canada has lost its elimination status means that the region of the Americas overall has also lost the status, which it achieved in 2016. Of the 35 countries and territories in the region—a health region designated by the World Health Organization—Canada is currently the only country where measles is considered to be spreading endemically, though other countries, namely the US and Mexico, are headed in the same direction.

    Measles is considered eliminated when a country can go 12 months without continuous local spread. Sporadic cases brought in from international travel can continue to occur, potentially causing limited outbreaks. But elimination is lost and endemicity is declared only when transmission is sustained over the course of a year.

    Read full article

    Comments

    • taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
    • chevron_right

      Canada fought measles and measles won; virus now endemic after 1998 elimination

      news.movim.eu / ArsTechnica • 10 November 2025

    Canada has lost its measles elimination status, meaning the highly infectious virus is considered endemic once again in the country, The Pan American Health Organization (PAHO) announced Monday.

    The determination was made by a committee of PAHO experts, who spent last week poring over disease data to assess the measles status of countries across the entire region. The fact that Canada has lost its elimination status means that the region of the Americas overall has also lost the status, which it achieved in 2016. Of the 35 countries and territories in the region—a health region designated by the World Health Organization—Canada is currently the only country where measles is considered to be spreading endemically, though other countries, namely the US and Mexico, are headed in the same direction.

    Measles is considered eliminated when a country can go 12 months without continuous local spread. Sporadic cases brought in from international travel can continue to occur, potentially causing limited outbreaks. But elimination is lost and endemicity is declared only when transmission is sustained over the course of a year.

    Read full article

    Comments

    • taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination taghealth taghealth taghealth tagcanada tagcanada tagcanada tagmeasles tagmeasles tagmeasles tagmexico tagmexico tagmexico tagpaho tagpaho tagpaho tagus tagus tagus tagvaccination tagvaccination tagvaccination

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
    • chevron_right

      Apple TV execs dismiss introducing an ad tier, buying Warner Bros. Discovery

      news.movim.eu / ArsTechnica • 10 November 2025

    The heads of Apple TV have “no plans” to bring ads to the streaming service, balking, at least for now, at a strategy that has driven success for Apple’s streaming rivals.

    In its November 2025 issue, British movie magazine Screen International asked Eddy Cue, SVP of Apple Services, if there are plans to launch an ad-based subscription tier for Apple TV. Cue responded:

    Nothing at this time. … I don’t want to say no forever, but there are no plans. If we can stay aggressive with our pricing, it’s better for consumers not to get interrupted with ads.

    The comments follow reports over the years suggesting that Apple has been seeking knowledge on how to build a streaming ads business. Most recently, The Telegraph reported that Apple TV executives met with the United Kingdom’s ratings body, Barb, to discuss what tracking ads on Apple TV would look like. In 2023, Apple hired advertising exec Lauren Fry as head of video and Apple News ad sales.

    Read full article

    Comments

    • tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
    • chevron_right

      Apple TV execs dismiss introducing an ad tier, buying Warner Bros. Discovery

      news.movim.eu / ArsTechnica • 10 November 2025

    The heads of Apple TV have “no plans” to bring ads to the streaming service, balking, at least for now, at a strategy that has driven success for Apple’s streaming rivals.

    In its November 2025 issue, British movie magazine Screen International asked Eddy Cue, SVP of Apple Services, if there are plans to launch an ad-based subscription tier for Apple TV. Cue responded:

    Nothing at this time. … I don’t want to say no forever, but there are no plans. If we can stay aggressive with our pricing, it’s better for consumers not to get interrupted with ads.

    The comments follow reports over the years suggesting that Apple has been seeking knowledge on how to build a streaming ads business. Most recently, The Telegraph reported that Apple TV executives met with the United Kingdom’s ratings body, Barb, to discuss what tracking ads on Apple TV would look like. In 2023, Apple hired advertising exec Lauren Fry as head of video and Apple News ad sales.

    Read full article

    Comments

    • tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
    • chevron_right

      Apple TV execs dismiss introducing an ad tier, buying Warner Bros. Discovery

      news.movim.eu / ArsTechnica • 10 November 2025

    The heads of Apple TV have “no plans” to bring ads to the streaming service, balking, at least for now, at a strategy that has driven success for Apple’s streaming rivals.

    In its November 2025 issue, British movie magazine Screen International asked Eddy Cue, SVP of Apple Services, if there are plans to launch an ad-based subscription tier for Apple TV. Cue responded:

    Nothing at this time. … I don’t want to say no forever, but there are no plans. If we can stay aggressive with our pricing, it’s better for consumers not to get interrupted with ads.

    The comments follow reports over the years suggesting that Apple has been seeking knowledge on how to build a streaming ads business. Most recently, The Telegraph reported that Apple TV executives met with the United Kingdom’s ratings body, Barb, to discuss what tracking ads on Apple TV would look like. In 2023, Apple hired advertising exec Lauren Fry as head of video and Apple News ad sales.

    Read full article

    Comments

    • tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming tagapple tagapple tagapple tagtech tagtech tagtech tagads tagads tagads tagapple tv tagapple tv tagapple tv tagstreaming tagstreaming tagstreaming

    • Pictures 3 image

    • visibility
    • visibility
    • visibility
  • history

    Get older posts

  • cloud_queue

    Powered by Movim