Cyberesia - AI Ecosystem

06/04/2026

Ne laissez plus l'IA penser à votre place. Inscrivez-vous sur Ikigenia.com, commencez à dialoguer avec les Shibanions et bénéficiez d'un accès priviligié à Aisance Cloud dès son lancement.

05/04/2026

Retrouvez déjà Les Shibanions sur Ikigenia.com et bientôt sur Aisance Cloud dans un nouveau format !

05/04/2026

Ikigenia.com, l’IA qui te permet d’apprendre plus vite que ceux qui prennent l’ascenseur.

03/04/2026

Le Shibanions rejoignent Aisance Cloud!
Les plus jeunes pourront découvrir les époques, les grandes découvertes et les personnages qui les ont façonnées dans un format nouveau, immersif et spectaculaire, pensé pour éveiller leur curiosité et leur donner envie d’en apprendre toujours davantage.

22/12/2025

19/09/2025

Swiss or not swiss? That's the question 😏..

19/09/2025

Funny enough: Swiss AI models speak better Chinese than Romansh…

⚡ Even a simple yes/no can send an LLM into a parallel universe.

hashtag : The issue isn’t stupidity… it’s lack of context.

I asked the new open-source LLM (Apertus-8B, running locally in bf16) a simple question in French:
“AI for Good, c’est un projet onusien, non ?”
Instead of a clear “Oui, c’est une initiative de l’ONU (UIT, Genève)”, it replied:
❌ “Non, c’est mené par swiss-ai.org.”
❌ Then claimed it was personally involved.
❌ Later contradicted itself: “l’ONU a aussi une initiative du même nom.”

Amusing to watch. But risky if you rely on it blindly.

The problem isn’t that the model is dumb.
It just runs without context.

Small models -> free-associate. Without grounding, they mix partial truths (swiss-ai.org exists) with contradictions (AI for Good is UN-led).
👉 Add retrieval (RAG), and the nonsense disappears.
The model grounds its answer in ITU/UN sources.
swiss-ai.org is recognized correctly as Switzerland’s AI initiative.
No contradiction. No invention.
Takeaway: Open-source models don’t need to be perfect out of the box.
They need context. With retrieval, even a small model becomes a trustworthy copilot.

Marcel Salathé – happy to share. The quantized models are showing their limits.
Have you seen LLMs create “parallel universes” like this? I do!

02/08/2025

AI Sentinel... The AI feed that’s faster, cleaner, bullsh*t-free, and speaks your language.

02/08/2025

AI news another way... no ads, no tracker, no scam

05/06/2025

Clone core workflows into autonomous agents that cut spend and return time back to your teams—without touching your current stack or compromising data sovereignty.

* 30 % operating cost drop
* 30-200 % first-year ROI Software
* 20 % FTE capacity released