Et voilà, le chiffre est tombé : 95 % des POC d’IA générative échouent (source en commentaire). Comme à chaque nouvelle hype autour de l’IA, on retrouve quatre problèmes majeurs : – À la vitesse où le domaine évolue, le temps qu’un projet se lance en interne, il existe déjà un
@dfintelligence
-
AI in Enterprise Beyond Chatbots
By
–
C’est lundi, petit rappel : l’IA en entreprise c'est pas seulement l’IA Gen. Je vois encore trop d’équipes obsédées par exemple par “avoir un chatbot” (souvent peu utilisé d'ailleurs faute de cadrage) ou par le dernier workflow n8n aperçu sur LinkedIn, qui relève surtout de
-
The recurring question of the best prompt for AI experts
By
–
I don't know a single data scientist, nor a single head of AI, from any company, who hasn't been asked internally, 'What's the best prompt for…' Even I, on every call, get asked this question when I'm doing scoping, sometimes even for trivial things.
-
Jupyter Notebooks in Production: The AI Prehistory New Experts Ignore
By
–
95% of the new "AI experts" on LinkedIn never experienced the good old days when some people wanted to put Jupyter Notebooks into production. That was the real prehistory of AI: explaining to managers that, well, no, we don't do that… Just because MNIST runs.
-
AI Lucie and Albert: Millions of Public Money Wasted, Unused
By
–
And we, for public servants, have the AIs Lucie and Albert, paid largely with public money (several million euros) that were just thrown in the trash because no one uses them. COCORICO!!!
-
Access to true GPT-5 then low-cost version to handle load
By
–
Alright folks, I've been thinking about this GPT-5 story. I think that for a month, I really had access to the real GPT-5… and that currently, it's a "low-cost" version to handle the load. My hypothesis is that maybe, in a few months, everyone will be able to
-
GPT5-Pro the only valid one but very slow
By
–
The only one that holds up is GPT5-Pro, with which I get nearly equivalent results. But good God, it is slow.
-
Post-Launch Disappointment: GPT-5 Generates Buggy Code Instead of Perfect Interfaces
By
–
Well… I can't get results as good as I used to with GPT-5 before the public launch… I'm bummed out. I don't know what's going on. Before the public launch, it used to nail amazing interfaces for me, and now it's generating code with bugs… too weird.
-
Militant for Claude Code for three months
By
–
PS: And I say this while I've been advocating for Claude Code for three months, which I love.
-
Claude Code is not one-shot, contrary to criticisms
By
–
In my last video, all my code benchmarks were done in one shot, in a single sentence. I see plenty of people acting like monkeys telling me: "Yes but Claude Code is better, you know nothiiing". So I say it again: CLAUDE CODE IS NOT ONE-SHOT!!! Yes