FYI Pangram is a coin toss, i've got crazy false positives with it
And this text does not smell AI-generated anyway (believing something just because there's a number on it is a bit midwitical tbh)
@aymericroucher
-
Pangram false positives; skepticism about trusting numbers
By
–
-
BitTorrent-like crowdsourced training using user compute
By
–
if data transfer was much cheaper and operations were much more expensive, we'd probably have bittorrent-like crowdsourced trainings?
An app where any individual signs in with their phone/computer, which uses their compute at a cost of X cents/hour to participate in the training -
Fable, a watered-down version of Mythos, refutes criticism of access and cost
By
–
Indeed, Fable is a version of Mythos (Mythos is a class of models) that has guardrails. This only reinforces my points: saying that Mythos is 1- restricted access, 2- poorly evaluated, 3- too expensive, is false, since we have a watered-down version that is 1- used.
-
Distinction between the Mythos class and the Mythos-Preview model
By
–
Mythos is a class of models, Fable is a version of it (cf: https://anthropic.com/glasswing)
Babinet seems to confuse the Mythos class and the Mythos-Preview model. -
Mythos: model class above Opus, Preview and Fable versions
By
–
Mythos is not a model, it's a class of models (cf this for example: https://anthropic.com/glasswing) above Opus, of which Mythos-Preview was the first commercial version, and Fable is the first public version. In Babinet's assertions that I note (1-usage
-
Babinet contradicts: Fable used by millions
By
–
La première minute de Babinet est un concentré de carabistouilles :
— m_ric (@AymericRoucher) 15 juin 2026
"Mythos [Fable] n'est utilisé que par quelques très rares entreprises"
-> Babinet vit dans une grotte? Des centaines de millions d'utilisateurs ont pu essayer Fable, dans le chat ou dans Claude Code.
"on est… https://t.co/82NaoKXcVKThe first minute of Babinet is a concentrate of nonsense: "Mythos [Fable] is only used by a very few rare companies"
-> Does Babinet live in a cave? Hundreds of millions of users have been able to try Fable, in chat or in Claude Code. "we are -
Data underappreciated in training models; watch out for these cracks.
By
–
Data is the most underappreciated topic in training models + These guys are cracks => watch out for what they'll do !
-
Aymeric Rouche: AI as a Tutor, Using Claude Code for Clear HTML Explanations
By
–
AI-as-a-tutor:
– Use Claude Code
– Have it generate explanations in HTML to claritymaxx -
AI consciousness: The pressing philosophical question of our age, debated on Hackernews
By
–
Intense discussion at the moment on Hackernews, reacting to Ted Chiang's article. "Can AI be conscious" is a pressing question, and could become the prime philosophical question of our age.
-
Analyzing LLM tokenization and cognitive competence
By
–
Ils peuvent faire des jeux de mots: "penser" au niveau du token n'empêche pas de maîtriser un ensemble plus large, tout comme écrire lettre par lettre ne vous empêcherait pas d'écrire tout un livre. Le problème est davantage un problème de compétences mais la compétence des LLM