davidbgk
/
larlet-fr-david


			
				
					
						
						
							123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630
							<!DOCTYPE html><!-- This is a valid HTML5 document. -->
<!-- Screen readers, SEO, extensions and so on. -->
<html lang="fr">
  <!-- Has to be within the first 1024 bytes, hence before the `title` element
  See: https://www.w3.org/TR/2012/CR-html5-20121217/document-metadata.html#charset -->
  <meta charset="utf-8">
  <!-- Why no `X-UA-Compatible` meta: https://stackoverflow.com/a/6771584 -->
  <!-- The viewport meta is quite crowded and we are responsible for that.
  See: https://codepen.io/tigt/post/meta-viewport-for-2015 -->
  <meta name="viewport" content="width=device-width,initial-scale=1">
  <!-- Required to make a valid HTML5 document. -->
  <title>
    CompAIgnon
  — David Larlet</title>
  <script>
    function toggleTheme(themeName) {
      document.documentElement.classList.toggle(
        'forced-dark',
        themeName === 'dark'
      )
      document.documentElement.classList.toggle(
        'forced-light',
        themeName === 'light'
      )
    }
    const selectedTheme = localStorage.getItem('theme')
    if (selectedTheme !== 'undefined') {
      toggleTheme(selectedTheme)
    }
  </script>
  <!-- Documented, feel free to shoot an email. -->
  <link rel="stylesheet" href="/static/david/css/style_2024-03-09.css">
  <!-- See https://www.zachleat.com/web/comprehensive-webfonts/ for the trade-off. -->
  <link rel="preload"
        href="/static/david/css/fonts/century_supra_ot_a_regular.woff2"
        as="font"
        type="font/woff2"
        media="(prefers-color-scheme: light), (prefers-color-scheme: no-preference)"
        crossorigin>
  <link rel="preload"
        href="/static/david/css/fonts/century_supra_ot_a_bold.woff2"
        as="font"
        type="font/woff2"
        media="(prefers-color-scheme: light), (prefers-color-scheme: no-preference)"
        crossorigin>
  <link rel="preload"
        href="/static/david/css/fonts/century_supra_ot_a_italic.woff2"
        as="font"
        type="font/woff2"
        media="(prefers-color-scheme: light), (prefers-color-scheme: no-preference)"
        crossorigin>
  <link rel="preload"
        href="/static/david/css/fonts/century_supra_ot_b_regular.woff2"
        as="font"
        type="font/woff2"
        media="(prefers-color-scheme: dark)"
        crossorigin>
  <link rel="preload"
        href="/static/david/css/fonts/century_supra_ot_b_bold.woff2"
        as="font"
        type="font/woff2"
        media="(prefers-color-scheme: dark)"
        crossorigin>
  <link rel="preload"
        href="/static/david/css/fonts/century_supra_ot_b_italic.woff2"
        as="font"
        type="font/woff2"
        media="(prefers-color-scheme: dark)"
        crossorigin>
  <meta name="description" content="Tech companies big and small sell AI as something that thinks for us. It does replace thought with statistics—but it is not intelligent. No one knows what the future will bring. But is a future without thought a better future?">
  <!-- That good ol' feed, subscribe :). -->
  <link rel="alternate"
        type="application/atom+xml"
        title="Feed"
        href="/david/log/">
  <!-- Generated from https://realfavicongenerator.net/ such a mess. -->
  <link rel="apple-touch-icon"
        sizes="180x180"
        href="/static/david/icons2/apple-touch-icon.png">
  <link rel="icon"
        type="image/png"
        sizes="32x32"
        href="/static/david/icons2/favicon-32x32.png">
  <link rel="icon"
        type="image/png"
        sizes="16x16"
        href="/static/david/icons2/favicon-16x16.png">
  <link rel="manifest" href="/static/david/icons2/site.webmanifest">
  <link rel="mask-icon"
        href="/static/david/icons2/safari-pinned-tab.svg"
        color="#07486c">
  <link rel="shortcut icon" href="/static/david/icons2/favicon.ico">
  <meta name="msapplication-TileColor" content="#f7f7f7">
  <meta name="msapplication-config"
        content="/static/david/icons2/browserconfig.xml">
  <meta name="theme-color"
        content="#f7f7f7"
        media="(prefers-color-scheme: light)">
  <meta name="theme-color"
        content="#272727"
        media="(prefers-color-scheme: dark)">
  <!-- Is that even respected? Retrospectively? What a shAItshow…
  https://neil-clarke.com/block-the-bots-that-feed-ai-models-by-scraping-your-website/ -->
  <meta name="robots" content="noai, noimageai">
  <!-- To get attribution when linking on mastodon. -->
  <meta name="fediverse:creator" content="@david@larlet.fr">
  
<style type="text/css">
  .tippy-content {
    min-width: 280px;
    padding: .5rem;
    font-size: calc(var(--fluid-0) * 0.8);
    font-family: var(--labor-font);
    letter-spacing: initial;
    text-align: left;
  }
  .tippy-content h3 {
    margin-top: 0;
  }
  .tippy-content h3 img {
    max-width: 2rem;
    max-height: 2rem;
    display: inline-block;
  }
  .tippy-content .tippy-links {
    display: flex;
    justify-content: space-around;
  }
  .tippy-content a {
    padding: .4rem;
    color: #F06048;
  }
</style>

  <body data-instant-intensity="viewport-all">
    <article>
      
  <header>
    <hgroup>
      <h1>CompAIgnon</h1>
      <p>Le <time datetime="2024-06-28">28 juin 2024</time></p>
    </hgroup>
  </header>
  <nav>
    <p>
      
        <a rel="prev"
           href="/david/2024/06/22/"
           title="Publication précédente : Jour 2">← Précédent</a> •
      
      <a href="/david/" title="Aller à l’accueil" rel="up">Accueil</a>
      •
      <a href="/david/recherche/"
         title="Aller à la page de recherche"
         rel="search" data-no-instant>Recherche</a>
      
        • <a rel="next"
            href="/david/2024/08/10/"
            title="Publication suivante : Kudopamine">Suivant →</a>
      
    </p>
  </nav>
  

<blockquote lang="en">
<p>Tech companies big and small sell AI as something that thinks for us. It <em>does</em> replace thought with statistics—but it is not intelligent. No one knows what the future will bring. <mark>But is a future without thought a better&nbsp;future?</mark></p>
<p>Now, with a tool that might help us think… How about using AI not to think less but <em>more</em>?</p>
<p><cite><em><a data-link-domain="ia.net" href="https://ia.net/topics/turning-the-tables-on-ai" hreflang="en"
            title="Consultation de l’article (anglais)">Turning the Tables on AI</a>
            <a href="/david/cache/2024/8f9d64c455a9246d5e23810ed10e3fe2/" hreflang="en"
                data-tippy data-description="AI is sold to make us think less... How about using AI not to think less but *more*?"
                data-source="https://ia.net/topics/turning-the-tables-on-ai"
                data-date="2024-06-24"
                data-favicon="https://ia.net/wp-content/themes/iA-library/assets/hotlink-ok/favicon-192x192.png"
                data-domain="ia.net"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>
<p>Un article qui ne considère pas les <abbr lang="en" title="Large Language Model">LLM</abbr> comme des machines à écrire mais comme des aides à penser. Ce n’est <a href="/david/2024/01/13/" title="Wuwei">toujours pas pour moi</a> mais j’apprécie cette tentative d’inverser la&nbsp;relation.</p>
<p>Quelles doutes / confidences serait-on prêt·es à confesser à une IA plus qu’à un humain&#8239;? Qu’est-ce que cela raconterait de nos peurs et&nbsp;insécurités&#8239;?</p>

<blockquote lang="en">
<p>The people who are most confident AI can replace writers are the ones who think writing is&nbsp;typing.</p>
<p><cite>Andrew Ti sur&nbsp;X</cite></p>
</blockquote>
<a href="#hr-189" title="Lien vers cette section de la page"><hr id="hr-189" /></a>

<blockquote lang="en">
<p>Trust is a precious commodity. <mark>It takes a long time to build trust.</mark> It takes a short time to destroy&nbsp;it.</p>
<p>I am honestly astonished that so many companies don’t seem to realise what they’re&nbsp;destroying.</p>
<p><cite><em><a data-link-domain="adactio.com" href="https://adactio.com/journal/21160" hreflang="en"
            title="Consultation de l’article (anglais)">Adactio: Journal-Trust</a>
            <a href="/david/cache/2024/d4c01d246dc4f43aeb95c427ac5422a5/" hreflang="en"
                data-tippy data-description="How to destroy your greatest asset with AI."
                data-source="https://adactio.com/journal/21160"
                data-date="2024-06-24"
                data-favicon="https://adactio.com/icon.png"
                data-domain="adactio.com"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>
<p>Lorsque la confiance est perdue avant même d’avoir vraiment été gagnée. <a data-link-domain="blog.kagi.com" href="https://blog.kagi.com/what-is-next-for-kagi#9" hreflang="en"
            title="Consultation de l’article (anglais)">Coucou Kagi</a>
            <a href="/david/cache/2024/463a5fd436bb887f90f9e01b01b7716b/" hreflang="en"
                data-tippy data-description="Two years ago, on June 1st, 2022, Kagi introduced ( https://blog.kagi.com/kagi-orion-public-beta ) a search engine that challenged the ad-supported version of the web."
                data-source="https://blog.kagi.com/what-is-next-for-kagi#9"
                data-date="2024-06-24"
                data-favicon="data:image/svg+xml,<svg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 100 100%22><text y=%22.9em%22 font-size=%2290%22>%f0%9f%97%9d%ef%b8%8f</text></svg>"
                data-domain="blog.kagi.com"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a>. Et peut-être que c’est une bonne chose de ne pas se sentir loyal envers une marque / entreprise en fait&#8239;? Profit et confiance font rarement bon&nbsp;ménage.</p>

<blockquote lang="en">
<p>Apple’s been big on touting their environmental record in past presentations but left it out of this one. <mark>Maybe it’s hard to hype fancy autocorrect</mark> when we’re spinning carbon-intensive power plants back up to power&nbsp;it.</p>
<p>Less privacy. More environmental impact. More hype. <em>Cool.</em></p>
<p>More than anything it all goes to show that the one thing you can count on companies to do over the long term is disappoint&nbsp;you.</p>
<p><cite><em><a data-link-domain="coryd.dev" href="https://coryd.dev/posts/2024/did-anyone-ask-for-these-ai-features/" hreflang="en"
            title="Consultation de l’article (anglais)">Did anyone ask for these AI features? • Cory Dransfeldt</a>
            <a href="/david/cache/2024/486e49501e495e42b2e40320969c373f/" hreflang="en"
                data-tippy data-description="I wasn't going to say anything about WWDC. I wasn't going to say anything about WWDC. But, did anyone actually ask for these AI features?"
                data-source="https://coryd.dev/posts/2024/did-anyone-ask-for-these-ai-features/"
                data-date="2024-06-24"
                data-favicon="https://cdn.coryd.dev/assets/icons/favicon.svg?v=19.5.32"
                data-domain="coryd.dev"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>
<a href="#hr-190" title="Lien vers cette section de la page"><hr id="hr-190" /></a>

<blockquote lang="en">
<p>Recently, however, a number of execs and so-called “thought leaders” in tech seem to have genuinely convinced themselves that generative AI is on the verge of replacing all the work done by junior engineers. I have read so many articles about how junior engineering work is being automated out of existence, or that the need for junior engineers is shriveling up. It has officially driven me&nbsp;bonkers.</p>
<p>All of this bespeaks a deep misunderstanding about what engineers actually do. <mark>By not hiring and training up junior engineers, we are cannibalizing our own future.</mark> We need to <em>stop doing that</em>.</p>
<p><cite><em><a data-link-domain="stackoverflow.blog" href="https://stackoverflow.blog/2024/06/10/generative-ai-is-not-going-to-build-your-engineering-team-for-you/" hreflang="en"
            title="Consultation de l’article (anglais)">Generative AI Is Not Going To Build Your Engineering Team For You</a>
            <a href="/david/cache/2024/93c287363dbe9cae22f4360c5b107971/" hreflang="en"
                data-tippy data-description=""
                data-source="https://stackoverflow.blog/2024/06/10/generative-ai-is-not-going-to-build-your-engineering-team-for-you/"
                data-date="2024-06-24"
                data-favicon="https://stackoverflow.blog/favicon.ico"
                data-domain="stackoverflow.blog"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>
<p>C’est l’une de mes réflexions actuelles pour une future activité qui inclurait de la transmission / montée en compétence technique. J’ai l’impression de pouvoir apporter des choses dans le domaine et les personnes que l’on appelle «&nbsp;junior&nbsp;» galèrent pour trouver du boulot alors qu’elles sont déjà&nbsp;capables.</p>
<a href="#hr-191" title="Lien vers cette section de la page"><hr id="hr-191" /></a>

<blockquote lang="en">
<p>The modern tech company or capitalistic enterprise is taking cues from fascism and dictators in how they introduce a small piece of controversy, like a testing of the waters. Wait for an “overreaction” by the masses/their audience/core users, wait for that to burnout and normalize, before flooding the zone with too much to fight or argue with all at once. <mark>This is the same way fascism causes burnout</mark> and complacency and feelings of helplessness, like there’s nothing to be done to fight the&nbsp;thing.</p>
<p>This is what is being done with&nbsp;AI.</p>
<p><cite><em><a data-link-domain="micro.anniegreens.lol" href="https://micro.anniegreens.lol/2024/05/06/ive-had-this.html" hreflang="en"
            title="Consultation de l’article (anglais)">I’ve had this thought but never put it into words.</a>
            <a href="/david/cache/2024/54ee1030f6fef778ec10e59726bc5731/" hreflang="en"
                data-tippy data-description="The modern tech company or capitalistic enterprise is taking …"
                data-source="https://micro.anniegreens.lol/2024/05/06/ive-had-this.html"
                data-date="2024-06-24"
                data-favicon="https://micro.blog/anniegreens/favicon.png"
                data-domain="micro.anniegreens.lol"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>
<p>À relire dans un contexte d’élections&nbsp;#OhWait.</p>
<a href="#hr-192" title="Lien vers cette section de la page"><hr id="hr-192" /></a>

<blockquote>
<p>L’une des grandes peurs entourant le déploiement des systèmes d’intelligence artificielle (IA) concerne ses potentielles «&nbsp;dérives&nbsp;», quand celle-ci pourrait «&nbsp;déraper&nbsp;». Sauf que, rappelle Ollivier Dyens, l’humanité n’a pas eu besoin de l’IA pour commettre des atrocités – camps de concentration, bombe atomique, massacres à la machette et à la baïonnette, etc. <em>Pourquoi devrions-nous avoir davantage peur de&nbsp;l’IA&#8239;?</em></p>
<p>L’intelligence artificielle, en tant que reflet de «&nbsp;nous-mêmes&nbsp;», pourrait en fait nous servir de miroir, à nous faire voir ce qui nous caractérise&#8239; (en tant qu’êtres humains, individuellement et collectivement), <mark>à nous montrer nos propres biais – et en particulier ceux que nous n’arrivons pas à discerner&nbsp;nous-mêmes.</mark></p>
<p><cite><em><a data-link-domain="lobrassard.net" href="https://www.lobrassard.net/carnet/2024-06-14-idees-recues-ecole-ete-humanites-numeriques.html" hreflang="fr"
            title="Consultation de l’article">Cinq idées reçues lors de l’école d’été sur les humanités numériques</a>
            <a href="/david/cache/2024/92ca1fc6eb390997db6b76afaf742d63/" hreflang="fr"
                data-tippy data-description="Carnet de recherche de Louis-Olivier Brassard."
                data-source="https://www.lobrassard.net/carnet/2024-06-14-idees-recues-ecole-ete-humanites-numeriques.html"
                data-date="2024-06-24"
                data-favicon="data:;base64,iVBORw0KGgo="
                data-domain="lobrassard.net"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>
<p>Compte-tenu de la façon dont sont entraînés ces LLM, la démonstration de ces biais se ferait en les accentuant… ce qui aurait potentiellement pour effet de les encourager au&nbsp;final&#8239;?!</p>
<a href="#hr-193" title="Lien vers cette section de la page"><hr id="hr-193" /></a>

<blockquote>
<p>Selon Louise Champoux-Paillé et Anne-Marie Croteau de l’École de gestion John-Molson, Université Concordia, les femmes sont moins intéressées par l’IA que les hommes, donc sont plus à risque de perdre leur emploi dans l’avenir. En fait, est-ce vraiment un manque d’intérêt ou ne serait-ce pas plutôt <mark>la difficulté de se tailler une place parce que les hommes ont la&nbsp;préférence?</mark></p>
<p><cite><em><a data-link-domain="mimo.blog" href="https://mimo.blog/le-panneau-des-prejuges" hreflang="fr"
            title="Consultation de l’article">Le panneau des préjugés</a>
            <a href="/david/cache/2024/63a6d75aee63c072c6d3b174c918740e/" hreflang="fr"
                data-tippy data-description="Selon Louise Champoux-Paillé et Anne-Marie Croteau de l'École de gestion John-Molson, Université Concordia, les femmes sont moins intéres..."
                data-source="https://mimo.blog/le-panneau-des-prejuges"
                data-date="2024-06-24"
                data-favicon="https://i.snap.as/CoWgXgFY.png"
                data-domain="mimo.blog"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>
<p>En parlant de&nbsp;biais.</p>
<a href="#hr-194" title="Lien vers cette section de la page"><hr id="hr-194" /></a>

<blockquote>
<p>Dire que ChatGPT «&nbsp;hallucine&nbsp;» est un anthropomorphisme&nbsp;: c’est une métaphore, une représentation calquée sur une réalité que nous vivons en tant qu’humains&#8239;; en l’occurrence, on hallucine quand on perçoit quelque chose qui n’est pas vraiment&nbsp;là.</p>
<p>Or la réalité du fonctionnement de ChatGPT, qui est basé sur un grand modèle de langue (large language model, LLM) est très différente&nbsp;: […]</p>
<p>C’est essentiel à comprendre&nbsp;: en tant que programme, <mark>la fonction de ChatGPT n’est pas de nous informer mais de simuler un interlocuteur humain.</mark> Pour cela, il assemble des éléments de langage de manière&nbsp;probabiliste.</p>
<p><cite><em><a data-link-domain="arthurperret.fr" href="https://www.arthurperret.fr/blog/2024-06-21-chatgpt-et-l-indifference-a-la-verite.html" hreflang="fr"
            title="Consultation de l’article">ChatGPT et l’indifférence à la vérité</a>
            <a href="/david/cache/2024/8b0a0eb43814f315e2b296aa5bdd5eee/" hreflang="fr"
                data-tippy data-description="Site web dâArthur Perret, enseignant-chercheur en SIC."
                data-source="https://www.arthurperret.fr/blog/2024-06-21-chatgpt-et-l-indifference-a-la-verite.html"
                data-date="2024-06-24"
                data-favicon="https://www.arthurperret.fr/blog/favicon-32x32.png"
                data-domain="arthurperret.fr"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>
<p>De l’importance du champ lexical qui entoure ces technologies. Ces outils ne manipulent pas que des mots, ils nous font croire à une nouvelle vérité alors qu’ils n’en ont pas la capacité. Et une partie de l’humanité devient croyante. (Le parallèle avec une religion est&nbsp;assumé.)</p>
<a href="#hr-195" title="Lien vers cette section de la page"><hr id="hr-195" /></a>

<blockquote lang="en">
<p>Don’t let the past steal your&nbsp;present.</p>
<p><cite>Cherríe L. Moraga</cite></p>
</blockquote>
<p>Comment les LLM peuvent-ils nous empêcher de tourner en rond pour aller <a href="/david/stream/2018/04/07/">vers la spirale</a>&#8239;? Je me fais du souci <a data-link-domain="arxiv.org" href="https://arxiv.org/abs/2406.07016">pour la science</a>&nbsp;notamment&nbsp;:</p>

<blockquote lang="en">
<p>Recent large language models (LLMs) can generate and revise text with human-level performance, and have been widely commercialized in systems like ChatGPT. These models come with clear limitations: they can produce inaccurate information, reinforce existing biases, and be easily misused. Yet, many scientists have been using them to assist their scholarly writing. How wide-spread is LLM usage in the academic literature currently? To answer this question, we use an unbiased, large-scale approach, free from any assumptions on academic LLM usage. We study vocabulary changes in 14&nbsp;million PubMed abstracts from 2010-2024, and show how the appearance of LLMs led to an abrupt increase in the frequency of certain style words. <mark>Our analysis based on excess words usage suggests that at least 10% of 2024&nbsp;abstracts were processed with LLMs. This lower bound differed across disciplines, countries, and journals, and was as high as 30% for some PubMed sub-corpora. We show that the appearance of LLM-based writing assistants has had an unprecedented impact in the scientific literature, surpassing the effect of major world events such as the Covid&nbsp;pandemic.</mark></p>
</blockquote>
<p>Avec l’éternel questionnement associé, qui ne date pas&nbsp;d’hier&nbsp;:</p>

<blockquote lang="en">
<p>So we enshrine the idea of intelligence into our computing paradigm when we don’t even know what it means for ourselves. <mark>Are we making computers smarter or ourselves&nbsp;dumber?</mark></p>

<blockquote>
<p>You can’t tell if a machine has gotten smarter or if you’ve just lowered your own standards of intelligence to such a degree that the machine seems&nbsp;smart.</p>
</blockquote>
<p>Prescient.</p>
<p><cite><em><a data-link-domain="blog.jim-nielsen.com" href="https://blog.jim-nielsen.com/2024/notes-from-you-are-not-a-gadget/" hreflang="en"
            title="Consultation de l’article (anglais)">Notes From “You Are Not A Gadget”</a>
            <a href="/david/cache/2024/b39aef611801e2b15f189e8d02586b88/" hreflang="en"
                data-tippy data-description="Writing about the big beautiful mess that is making things for the world wide web."
                data-source="https://blog.jim-nielsen.com/2024/notes-from-you-are-not-a-gadget/"
                data-date="2024-06-24"
                data-favicon="https://blog.jim-nielsen.com/favicon.ico"
                data-domain="blog.jim-nielsen.com"
                ><svg xmlns="http://www.w3.org/2000/svg"
                    width="24" height="24" viewBox="0 0 24 24" fill="none"
                    stroke="currentColor" stroke-width="2" stroke-linecap="square"
                    stroke-linejoin="round"><circle cx="12" cy="12" r="10"></circle>
                    <path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"></path>
                    <line x1="12" y1="17" x2="12.01" y2="17"></line>
                </svg>
                <span class="sr-only">[archive]</span></a></em></cite></p>
</blockquote>

  
    <nav>
      <p>
        
          <a href="/david/2024/ecriture/"
             title="Liste de tous les articles 2024 associés à cette étiquette"
             rel="tag">#écriture</a>
        
          <a href="/david/2024/ia/"
             title="Liste de tous les articles 2024 associés à cette étiquette"
             rel="tag">#IA</a>
        
          <a href="/david/2024/parvenir/"
             title="Liste de tous les articles 2024 associés à cette étiquette"
             rel="tag">#parvenir</a>
        
        <a href="/david/2024/#tags" title="Liste de toutes les étiquettes 2024">tous ?</a>
      </p>
    </nav>
  
  <nav>
    <p>
      
        <a rel="prev"
           href="/david/2024/06/22/"
           title="Publication précédente : Jour 2">← Précédent</a> •
      
      <a href="/david/2024/" title="Liste des publications récentes">↑ En 2024</a>
      
        • <a rel="next"
            href="/david/2024/08/10/"
            title="Publication suivante : Kudopamine">Suivant →</a>
      
    </p>
  </nav>

      
        <form action="/david/recherche/" method="get">
          <fieldset>
            <legend>Recherche</legend>
            <label for="input-search">Termes de votre recherche :</label>
            <input id="input-search" type="search" name="s" aria-describedby="indexation-infos" required>
            <input type="submit" value="Chercher">
            <p id="indexation-infos">
              <small>
                Seuls les contenus de ces 8 dernières années sont indexés.
              </small>
            </p>
          </fieldset>
        </form>
      
      <aside>
        <theme-toggle></theme-toggle>
      </aside>
    </article>
    <hr>
    <footer>
      <p>
        <a href="/david/" title="Aller à l’accueil">Accueil</a>
        •
        <a href="/david/log/" title="Accès au flux RSS">Suivre</a>
        •
        <a href="http://larlet.com"
           title="Go to my English profile"
           data-instant>Pro</a>
        •
        <a href="mailto:david%40larlet.fr" title="Envoyer un courriel">Email</a>
        •
        <abbr title="Hébergeur : Alwaysdata, 62 rue Tiquetonne 75002 Paris, +33184162340">Légal</abbr>
      </p>
      <template id="theme-selector">
        <form>
          <style type="text/css">
            fieldset div {
              text-align: center;
            }
          </style>
          <fieldset>
            <legend>Thème</legend>
            <div>
              <label>
                <input type="radio" value="auto" name="chosen-color-scheme" checked>
                Auto
              </label>
              <label>
                <input type="radio" value="dark" name="chosen-color-scheme">
                Foncé
              </label>
              <label>
                <input type="radio" value="light" name="chosen-color-scheme">
                Clair
              </label>
            </div>
          </fieldset>
        </form>
      </template>
    </footer>
    <script src="/static/david/js/instantpage-5.1.0.min.js" type="module"></script>
    <script>
      class ThemeToggle extends HTMLElement {
        constructor() {
          super()
          const themeSelectorTemplate = document.querySelector('#theme-selector')
          const form = themeSelectorTemplate.content.firstElementChild
          this.attachShadow({ mode: 'open' })
          this.shadowRoot.appendChild(form.cloneNode(true))
        }

        connectedCallback() {
          const form = this.shadowRoot.querySelector('form')
          form.addEventListener('change', (e) => {
            const chosenColorScheme = e.target.value
            localStorage.setItem('theme', chosenColorScheme)
            toggleTheme(chosenColorScheme)
          })

          const selectedTheme = localStorage.getItem('theme')
          if (selectedTheme && selectedTheme !== 'undefined') {
            form.querySelector(`[value="${selectedTheme}"]`).checked = true
          }
        }
      }

      const prefersColorSchemeDark = '(prefers-color-scheme: dark)'
      window.addEventListener('load', () => {
        let colorsLayer = undefined
        let hasDarkRules = false
        for (const styleSheet of Array.from(document.styleSheets)) {
          let mediaRules = []
          for (const layerRule of styleSheet.cssRules) {
            if (!(layerRule instanceof CSSLayerBlockRule)) {
              continue
            }
            if (layerRule.name === 'colors') {
              colorsLayer = layerRule
            }
            for (const cssRule of layerRule.cssRules) {
              if (cssRule.type !== CSSRule.MEDIA_RULE) {
                continue
              }
              // WARNING: Safari does not have/supports `conditionText`.
              if (cssRule.conditionText) {
                if (cssRule.conditionText !== prefersColorSchemeDark) {
                  continue
                }
              } else {
                if (cssRule.cssText.startsWith(prefersColorSchemeDark)) {
                  continue
                }
              }
              mediaRules = mediaRules.concat(Array.from(cssRule.cssRules))
            }
          }

          // WARNING: do not try to insert a Rule to a styleSheet you are
          // currently iterating on, otherwise the browser will be stuck
          // in a infinite loop…
          for (const mediaRule of mediaRules) {
            // Safari requires the `0` second parameter (even if default).
            colorsLayer.insertRule(mediaRule.cssText, 0)
            hasDarkRules = true
          }
        }

        if (hasDarkRules) {
          if ('customElements' in window && !customElements.get('theme-toggle')) {
            customElements.define('theme-toggle', ThemeToggle)
          }
        }
      })
    </script>
    
  <script src="/static/david/js/popper-2.11.8.min.js"></script>
  <script src="/static/david/js/tippy-bundle-6.3.7.umd.min.js"></script>
  <script>
    tippy('[data-tippy]', {
      content(reference) {
        reference.addEventListener('click', (e) => e.preventDefault())
        return `
        <h3 lang="fr">
          <img src="${reference.dataset.favicon}" loading="lazy">
          <a href="${reference.dataset.source}"
            >Article sur ${reference.dataset.domain}</a></h3>
        <p lang="${reference.hreflang}"><em>${reference.dataset.description}</em></p>
        <div class="tippy-links" lang="fr">
          <a href="${reference.href}">Archive au ${reference.dataset.date}</a>
        </div>
        `
      },
      allowHTML: true,
      interactive: true,
      delay: [150, 700],
      hideOnClick: false
    })
  </script>
  <script type="module">
    import { annotate } from '/static/david/js/rough-notation-0.5.1.esm.min.js'

    const markObserver = new IntersectionObserver((entries, observer) => {
      const computedStyle = getComputedStyle(document.documentElement)
      const markBackground = computedStyle.getPropertyValue('--mark-background')
      for (const entry of entries) {
        if (entry.intersectionRatio === 0) continue
        const markElement = entry.target
        markElement.style.backgroundColor = 'inherit'
        const annotation = annotate(
          markElement, {
            type: 'highlight',
            multiline: true,
            color: markBackground,
            // animate: !window.matchMedia('(prefers-reduced-motion: reduce)').matches
            animate: false
          }
        )
        annotation.show()
        observer.unobserve(markElement)
      }
    }, {threshold: 1.0})

    for (const markElement of document.querySelectorAll('mark')) {
      markObserver.observe(markElement)
    }
  </script>

  </body>
</html>