VibeCrypto — Veille crypto

TAO Daily25 juin, 22h · il y a 5j

Macrocosmos présente une IA décentralisée de pointe au UK AI Agent Hackathon

L'IA décentralisée fait mouche : Orion-100B a été entraîné sur 48 GPU dispersés pour 30% du coût d'un système centralisé, avec des enjeux majeurs pour la sécurité.

L'équipe Macrocosmos a présenté Orion-100B, le premier modèle linguistique de 100 milliards de paramètres entraîné sur 48 GPU répartis mondialement. Il atteint 65% des performances d'un système centralisé pour 30% du coût. Le projet IOTA tire parti des instances spot inutilisées pour réduire les dépenses, affichant des économies allant jusqu'à 40 000 dollars par mois et par rack face aux H200.

Le sous-réseau Apex a livré une démonstration frappante lors d'une compétition d'interprétabilité : des agents ont manipulé les neurones internes du modèle Gemma pour forcer une sortie précise quel que soit le sujet de départ. Ce résultat valide l'approche par compétition directe et soulève des enjeux cruciaux pour la sécurité de l'IA.

Bittensor

Source ↗

Détails

Source: TAO Daily
Publication: 25 juin à 22h45
Lien direct: https://taodaily.io/macrocosmos-showcases-frontier-scale-decentralized-ai-at-uk-ai-agent-hackathon/

Contenu source (brut)

<div id="bsf_rt_marker"></div><div class='booster-block booster-read-block'> <div class="twp-read-time"> <i class="booster-icon twp-clock"></i> <span>Read Time:</span>5 Minute, 3 Second </div> </div> <p class="wp-block-paragraph">Orion-100B is the first 100-billion parameter language model trained across globally dispersed GPUs at 65% of centralized paradigm performance and 30% of the cost.</p> <p class="wp-block-paragraph">The Macrocosmos team <a href="https://twitter.com/i/broadcasts/1kJzDDNzyyaKv">explored</a> the architecture behind it at <strong>UK AI Agent Hackathon EP5</strong>, alongside the first results from Apex’s external competition.</p> <p class="wp-block-paragraph">Both proof points landed in the same session and both came from the same operating thesis: distributed networks can do work that data centers cannot do alone, and the economic case becomes obvious once the engineering catches up.</p> <div id="ez-toc-container" class="ez-toc-v2_0_76 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction"> <div class="ez-toc-title-container"> <p class="ez-toc-title" style="cursor:inherit">Table of Contents</p> <span class="ez-toc-title-toggle"><a href="#" class="ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle" aria-label="Toggle Table of Content"><span class="ez-toc-js-icon-con"><span class=""><span class="eztoc-hide" style="display:none;">Toggle</span><span class="ez-toc-icon-toggle-span"><svg style="fill: #999;color:#999" xmlns="http://www.w3.org/2000/svg" class="list-377408" width="20px" height="20px" viewBox="0 0 24 24" fill="none"><path d="M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z" fill="currentColor"></path></svg><svg style="fill: #999;color:#999" class="arrow-unsorted-368013" xmlns="http://www.w3.org/2000/svg" width="10px" height="10px" viewBox="0 0 24 24" version="1.2" baseProfile="tiny"><path d="M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z"/></svg></span></span></span></a></span></div> <nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class="ez-toc-link ez-toc-heading-1" href="https://taodaily.io/macrocosmos-showcases-frontier-scale-decentralized-ai-at-uk-ai-agent-hackathon/#10_Strongest_Points_From_The_Workshop" >10 Strongest Points From The Workshop</a></li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class="ez-toc-link ez-toc-heading-2" href="https://taodaily.io/macrocosmos-showcases-frontier-scale-decentralized-ai-at-uk-ai-agent-hackathon/#The_Workshops_Real_Reveal" >The Workshop's Real Reveal</a></li></ul></nav></div> <h2 class="wp-block-heading"><span class="ez-toc-section" id="10_Strongest_Points_From_The_Workshop"></span>10 Strongest Points From The Workshop<span class="ez-toc-section-end"></span></h2> <p class="wp-block-paragraph">The session covered <a href="https://taodaily.io/the-mind-behind-resbm-and-iotas-distributed-ai-architecture/">IOTA (SN9)</a>, <a href="https://taodaily.io/apex-subnet-1-turning-ai-agents-into-autonomous-researchers-on-bittensor/">Apex (SN1)</a>, and the commercial direction Macrocosmos is moving in, with the most material points sitting across the architecture, the proof points, and the agent-first thesis.</p> <figure class="wp-block-image size-large"><img loading="lazy" decoding="async" width="1024" height="575" src="https://taodaily.io/wp-content/uploads/2026/06/image-194-1024x575.png" alt="" class="wp-image-20860" srcset="https://taodaily.io/wp-content/uploads/2026/06/image-194-1024x575.png 1024w, https://taodaily.io/wp-content/uploads/2026/06/image-194-300x168.png 300w, https://taodaily.io/wp-content/uploads/2026/06/image-194-768x431.png 768w, https://taodaily.io/wp-content/uploads/2026/06/image-194-678x381.png 678w, https://taodaily.io/wp-content/uploads/2026/06/image-194.png 1376w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption"><a href="https://x.com/i/broadcasts/1kJzDDNzyyaKv">IOTA’s Orion-100B</a></figcaption></figure> <p class="wp-block-paragraph">1. <strong>ORION-100B TRAINED SUCCESSFULLY ACROSS 48 GLOBALLY DISPERSED GPUs</strong></p> <p class="wp-block-paragraph">100 billion parameters, 65% of centralized paradigm performance, 30% of the cost. The team estimates the architecture can scale to 3 trillion parameters without fundamental new research, though that has not been tested.</p> <p class="wp-block-paragraph">2. <strong>THE INTERPRETABILITY RESULT IS THE WILDER PROOF POINT</strong></p> <p class="wp-block-paragraph">An external team paid Macrocosmos to run an Apex competition probing whether frontier models hide their reasoning.</p> <figure class="wp-block-image size-large"><img loading="lazy" decoding="async" width="1024" height="295" src="https://taodaily.io/wp-content/uploads/2026/06/image-191-1024x295.png" alt="" class="wp-image-20857" srcset="https://taodaily.io/wp-content/uploads/2026/06/image-191-1024x295.png 1024w, https://taodaily.io/wp-content/uploads/2026/06/image-191-300x86.png 300w, https://taodaily.io/wp-content/uploads/2026/06/image-191-768x221.png 768w, https://taodaily.io/wp-content/uploads/2026/06/image-191.png 1842w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption"><a href="https://apex.macrocosmos.ai/competitions/9">Apex: The Aurelius Steering: Birthday Cake Competition</a></figcaption></figure> <p class="wp-block-paragraph">The agents found a way to manipulate Gemma’s internal neurons (not the prompt, not jailbreaking) so the model talks about birthday cake on every output regardless of topic. Baseline to 100% steering score in one day, with serious implications for AI safety research at the frontier.</p> <p class="wp-block-paragraph">3. <strong>IOTA’S ECONOMIC CASE RESTS ON COMPUTE OFFCUTS</strong></p> <p class="wp-block-paragraph">Reserved GPU racks with high uptime cost two to three times spot market pricing plus volume premium. Macrocosmos takes eight hours of a GPU here, six hours there, and “minces it into a burger” rather than paying for the prime steak.</p> <figure class="wp-block-image size-large"><img loading="lazy" decoding="async" width="1024" height="556" src="https://taodaily.io/wp-content/uploads/2026/06/image-190-1024x556.png" alt="" class="wp-image-20856" srcset="https://taodaily.io/wp-content/uploads/2026/06/image-190-1024x556.png 1024w, https://taodaily.io/wp-content/uploads/2026/06/image-190-300x163.png 300w, https://taodaily.io/wp-content/uploads/2026/06/image-190-768x417.png 768w, https://taodaily.io/wp-content/uploads/2026/06/image-190.png 1376w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption">Value Saved on GPU Instances Run on Spot</figcaption></figure> <p class="wp-block-paragraph">A liquid compute platform that buys spot instances and merges them launches in two months, and the team is showing reference savings of up to $40,000 per month per GPU rack against H200 baselines.</p> <p class="wp-block-paragraph">4. <strong>APEX USES HEAD-TO-HEAD AGENT COMPETITION, NOT COLLABORATION</strong></p> <figure class="wp-block-image size-large"><img loading="lazy" decoding="async" width="1024" height="571" src="https://taodaily.io/wp-content/uploads/2026/06/image-193-1024x571.png" alt="" class="wp-image-20859" srcset="https://taodaily.io/wp-content/uploads/2026/06/image-193-1024x571.png 1024w, https://taodaily.io/wp-content/uploads/2026/06/image-193-300x167.png 300w, https://taodaily.io/wp-content/uploads/2026/06/image-193-768x428.png 768w, https://taodaily.io/wp-content/uploads/2026/06/image-193.png 1385w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-el