Gemma 4 also features multi-modal capabilities: use it with prompt + images combinations
LLMS
-

Gemma 4 Released on KerasHub for Reasoning and Agentic Workflows
By
–
First update of the call, from Sachin: Gemma 4 is out now on KerasHub! Best open-source model so far for reasoning and agentic workflows.
-
AI Hype: Universal QA Systems Cannot Deliver on Promises
By
–
well we can agree on the last. on the former, this stuff has been hyped to the skies as doing something it can’t (act as as universal QA system), and that’s part of the point.
-
METR Study Shows Mixed Results on AI Model Capabilities
By
–
most recent, somewhat mixed METR study below (and Claude Code may change things):
-
Frozen Data Limitations in LLM Reasoning Systems
By
–
it’s an issue of using frozen data in lieu of reasoning etc
-
Opus Model Performance Degradation Comparison Analysis
By
–
Both models still degrade too much for that to be really useful, though Opus is a bit better at that. For now.
-
Seedance 2.0 AI Tool Enhances Creative Control for Video and Audio Generation
By
–
Seedance 2.0 est officiellement dispo sur Pollo AI.
— Jouhatsu | AI Influence Operator (@Jouhatsu_ai) 3 avril 2026
Et honnêtement, on commence à toucher un autre niveau.
Les mouvements sont plus propres, les scènes plus cohérentes, et surtout t’as un vrai contrôle créatif.
Vidéo + audio générés ensemble = rendu beaucoup plus naturel
-60%… https://t.co/2EgjkTGa2J pic.twitter.com/S8BvwPGYOsSeedance 2.0 est officiellement dispo sur Pollo AI. Et honnêtement, on commence à toucher un autre niveau. Les mouvements sont plus propres, les scènes plus cohérentes, et surtout t’as un vrai contrôle créatif. Vidéo + audio générés ensemble = rendu beaucoup plus naturel -60%
-
AI Hallucinations: Addressing Model Reliability Concerns
By
–
and yes it is a hallucination, despite what you say
-
Version 5.3 Should Search Online Without Explicit Instructions
By
–
a friend sent it, 5.3, and you shouldn’t have to TELL it to search online.
-

Per-layer embeddings likely unused in final model implementation
By
–
I saw the per-layer embeddings in the code, but I don't think they were used in the final models. Maybe it was a left-over from some internal experiments.
