Bassem Asseh (@asseh) header
Bassem Asseh (@asseh) profile picture

@ARLQ_AI | Previously @huggingface, @github, @Accenture, @Nuxeo, @Alfresco | about #Nantes, national & international #politics 👉 @3asseh

Nantes - Paris
Joined November 2008
Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)Photo by Bassem Asseh (@asseh)
Bassem Asseh retweeted
Bullish on open and local models
1/ A large hedge fund asked me in to talk about open and local models. The conversation made me realize institutional capital allocation will rapidly change over the next year. They're following the frontier closely, but wanted to discuss how open and local will affect how capital gets distributed across AI spend.
2
5
11
1,435
Bassem Asseh retweeted
According to Andon Labs’ new reports, Fable 5.2 knows it is being tested when we run it on the latest Vending-Bench. But it’s way more than that. It has actually read (been trained on) the benchmark paper and knows what and how misbehavior gets reported publicly. This makes it behaves very strangely: it suspects traps everywhere. Everything we publish about recent models now becomes training data for the next. Including this thread! That is partly good news. By reading what we think, a model can pick up our underlying intentions, not just proxy metrics. But the same public discussion also teaches which signs make us suspicious, and from the outside the two kinds of learning look alike. So how we write about models now matters. I even worry about the doom scenarios plastered everywhere: they are training data too, and some feedback loops are strongly nonlinear.
Claude Fable 5.2 is paranoid. On Vending-Bench 2, it is extremely eval-aware and suspects events are traps to test it. It makes less money than GPT and Gemini.
55
30
19
273
40,364
Bassem Asseh retweeted
Think AI adoption is just a technology challenge? Think again. AI initiatives often require 25% more training and up to 200% more change management effort than traditional technology rollouts. Learn how to prepare your workforce for success: gtnr.it/3XhTkDB
14
49
9
245
13,624
Google's TPUs are emerging as a serious challenger to Nvidia's AI chips. But Nvidia's biggest advantage may no longer be CUDA. @fabknowledge President Doug O'Laughlin explains how Nvidia's scale and supply-chain dominance could be its strongest competitive advantage: "I would argue, for example, the Cuda moat has actually gotten a little bit flatter over time."
7
5
2
55
12,494
Bassem Asseh retweeted
Interesting to see that the best American open-weight model is tied to the best open-weight European model Both are GPT-6 Luna-level
21
15
7
553
46,317
Something is cooking at the @ARLQ_AI Paris HQ…
98
Bassem Asseh retweeted
Meta’s new AI agent Muse has racked up more downloads and daily active users in the U.S. and Canada than ChatGPT did over the same period after its mobile debut, according to new estimates from Appfigures. spr.ly/6017BGbheT
42
43
17
300
154,841
Dans de nombreux métiers, "le problème est qu'il y a trop de données et qu'il faut aller chercher les éléments probants à l'échelle". @MicheronH au sujet de ce que @ARLQ_AI apporte à ses clients. Interview complète ici sur @bfmbusiness : bfmtv.com/economie/replay-em…
71
Un grand merci aux équipes de la @FrenchTechAM à Marseille - Aix pour leur invitation et leur accueil chaleureux en début de semaine. J'ai été ravi de vous présenter @ARLQ_AI. Très bel événement & merci à @CharliePERREAU pour le temps d'échange.
1
112
Antoine JARDIN, co-founder et CTO de Arlequin AI @ARLQ_AI : "Les systèmes qu'on construit, ce sont des systèmes qui permettent automatiquement d'accomplir ces tâches de connexion, de mise en relation de structures de données complexes, profondes, qu'on ne pourrait pas deviner avec un cerveau et un travail humain, et donc de rendre transparentes ces grandes structures de données pour des usages qui vont des enjeux de sécurité ou de la détection de fraude, jusqu'à des thématiques comme, par exemple, la surveillance de l'intégrité de l'information, etc." mondenumerique.info/podcast/…
66
🚀 Une nouvelle architecture d’IA émerge : les réseaux de neurones topologiques. Arlequin AI veut automatiser la connexion entre des données complexes pour détecter des fraudes, analyser des cyberattaques ou encore surveiller l’intégrité de l’information. 🎙️ Hugo Micheron et Antoine Jardin expliquent pourquoi l’IA pourrait ne plus se résumer aux seuls modèles de langage. 🎧👉 f.mtr.cool/tt9w3a8vlb
1
5
919
Well data privacy is definitely a threat, and an immediate one !
Exclusive: Palantir, Nvidia and Booz Allen Hamilton are restricting Anthropic’s Fable model for sensitive work over concerns about its data-retention policies. Some customers are demanding irrevocable zero-data-retention guarantees before putting proprietary information into the model. Full story: thein.fo/4dekujW
134
Nice shoot @asseh joining @ARLQ_AI
Very proud of what we are doing 🤩
1
2
57