270 million (not billion) parameters! https://
huggingface.co/google/gemma-3
-270m
…
INNOVATION
-

Google releases Gemma 3 with 270 million parameters
By
–
-
GPT-5 GB200 FP4 Precision and Token Speed Analysis
By
–
So GPT-5 is running on GB200, to take full advantage of it, it could be in FP4. So I wonder if this change is hiding the size of the model somewhat. If it was FP8 and on H100, it would be now running at something like 10 tokens per second rather than current 50 tps.
-

Google Releases Gemma3 270M Tiny LLM on KerasHub
By
–
Google just dropped a new tiny LLM with outstanding performance — Gemma3 270M. Now available on KerasHub. Try the new presets `gemma3_270m` and `gemma3_instruct_270m`!
-
Block Wrappers Simplify Component Architecture Mental Load
By
–
That’s a new thing: instead of individual components, you craft block wrappers (which contain) the components so then it simplifies the mental load – you think in blocks, instead of components.
-
Simplify reasoning effort adjustment in AI interfaces
By
–
Could we make it cleaner to change the reasoning effort? I always forget how to do it and hard to find
-
NotebookLM Video Overviews: Gemini’s Multimodal AI Analysis
By
–
When the @NotebookLM team was building video overviews, they wanted to combine the best of Gemini's multimodality into one feature. The AI host "sees" your sources, processes the information, and then is able to discuss what is truly unique about them.
— Google AI (@GoogleAI) 14 août 2025
For example, we uploaded… pic.twitter.com/xa9rTSoKBCWhen the @NotebookLM team was building video overviews, they wanted to combine the best of Gemini's multimodality into one feature. The AI host "sees" your sources, processes the information, and then is able to discuss what is truly unique about them. For example, we uploaded
-
MCP Gateways: Critical Infrastructure for AI Agent Deployments
By
–
MCP gateways are becoming critical infrastructure for AI deployments, but what's in a gateway and why do you need one? Sharing highlights from a recent blog post on this. Every AI agent needs access to tools: databases, APIs, internal services. The Model Context Protocol (MCP)
-
Comet Enterprise Pro Features for Advanced Users
By
–
Learn more about Comet for Enterprise Pro users:
-
Guardrails AI Launches: Long-time Follower Shares Support
By
–
I have followed @guardrails_ai since they were founded and retweeted their previous launches.
-
Snowglobe: New Simulation Engine for Testing AI Chatbots
By
–
Check out @snowglobe_so , the new simulation engine for testing chatbots from @guardrails_ai https://t.co/hGG5fK6CdI
— Ian Goodfellow (@goodfellow_ian) 14 août 2025Check out @snowglobe_so , the new simulation engine for testing chatbots from @guardrails_ai shreya rajpal (@ShreyaR) Introducing ❄️ @snowglobe_so, the simulation engine for AI chatbots. Magically simulate the behavior of your users to test and improve your chatbots. Find failures before your users do. — https://nitter.net/ShreyaR/status/1956023326721368337#m
→ View original post on X — @goodfellow_ian, 2025-08-14 17:09 UTC