Google Releases Gemma 3n: A Multimodal AI to Power a New Wave of On-Device AI Apps

#AI #Google #Gemma3n #Gemma3 #OpenSource #OnDeviceAI #EdgeAI #MachineLearning #Developer #MultimodalAI #LLM

https://winbuzzer.com/2025/06/27/google-releases-gemma-3n-a-multimodal-ai-to-power-a-new-wave-of-on-device-ai-apps-xcxwbn/

Andalucía anuncia JuntaGPT: la IA llega a la administración pública de la mano de Google y Gemma 3 https://blog.elhacker.net/2025/06/andalucia-juntagpt-ia-google-gemma3.html #inteligenciaartificial #andalucía #españa #gemma3 #google

One of my biggest problems with #LLM is that they normally run in the #cloud. I don't want to share all my questions and thoughts, especially with a company under USA jurisdiction (the EU is quite a bit more concerned about protecting my privacy).

Today I installed #googleaiedgegallery Now I am running a #gemma3 model with 3 GB locally on my 4 years old Pixel 6 smartphone. And I am thrilled!

It's still beta software, but relatively easy to install:
https://github.com/google-ai-edge/gallery

Ever wished you could star in a documentary? I made a 'documentary' about myself using AI - Gemma 3, Eleven Labs, & a ton of code! Check it out to see how I did it! #AI #Documentary #Gemma3 https://youtu.be/8n_tpLn6Xbo

Ho provato #Gemma3 di #Google in locale con #prompt multimodali, composti da testo + immagini. 
La nuova versione di #Ollama, infatti, supporta i modelli multimodali, grazie a un nuovo engine. 
Negli esempi uso il modello per descrivere delle immagini, ma anche per leggere dati dall'immagine di una tabella. 
Il risultato? Ottimo! Un altro passo avanti per l'utilizzo dei modelli "piccoli" e open source in locale. 

Fuck yo' sass!

Brilliant.

This isn't a spoiler, but #Gemma3 was able to geolocate the real life building used for this scene in this week's latest #Andor episode (again, not a spoiler).

[cc: @simon]

Microsoft Clippy Returns as AI Assistant, Empowered By LLMs You Can Run Locally on Your PC

#AI #Clippy #AIClippy #AIAssistants #LLMs #LocalAI #OpenSource #ElectronJS #LlamaCpp #GGUF #Gemma3 #Llama3 #Phi4 #Qwen3 #RetroTech #MicrosoftOffice #OnDeviceAI

https://winbuzzer.com/2025/05/06/microsoft-clippy-returns-as-ai-assistant-empowered-by-llms-you-can-run-locally-on-your-pc-xcxwbn/

Google Tests "Computer Use" AI Agent in AI Studio, Potentially Streamlining Gemma 3 Deployment

#AI #AIAgents#GoogleAI #AIStudio #Gemma3 #CloudRun #Automation #TechNews #DeveloperTools #Alpbabet #BigTech

https://winbuzzer.com/2025/05/06/google-tests-computer-use-ai-agent-in-ai-studio-potentially-streamlining-gemma-3-deployment-xcxwbn/

So, I had an idea while messing with #LocalLLM geolocation with #Gemma3.

I fed it this photo from Wikipedia of Governors Island in NY.

Unsurprisingly, it figured it out immediately.

> Considering all the evidence, I am confident that the image depicts Governors Island in New York City. The shape of the island, the surrounding skyscrapers, the port facilities, and the presence of the Verrazano-Narrows Bridge all point to this location.

Fair enough. That's baby stuff at this point. ;)

But then I had an idea: "Assuming this refined guess is correct, where might the photo have been taken FROM?"

The ol' geolocate the camera trick. (Usually finding a CCTV or something, in some cases.)

> Based on the clarity of the view of Governors Island, the partial obstruction of other buildings, and the overall angle, I believe the photo was most likely taken from One World Trade Center. The height and location provide the perfect vantage point for this shot.
>
> Specifically, I'd estimate the photo was taken from the *observation deck or a high-floor office within One World Trade Center, likely facing southeast towards Governors Island and the Verrazano-Narrows Bridge*.

And uh... well, yeah probably. :D

Though I don't expect nearly as much success on other photos. I'll keep trying and see what I get.

[cc: @simon]

KI mit 27 Milliarden Parametern – und trotzdem auf deiner GPU? Google macht’s mit Gemma 3 möglich. Dank Quantisierung läuft das Modell sogar auf Consumer-Hardware wie der RTX 3090. Was bedeutet das für Open Source, Datenschutz und lokale KI-Nutzung? https://www.all-ai.de/news/topbeitraege/googles-ki-offensive-gemma-3-f%C3%BCr-alle #Google #Gemma3 #KI

#Google releases #Gemma3 QAT for consumer GPUs

https://technewsro.blog/google-lanseaza-gemma-3-qat-pentru-gpu-urile-de-consum/

Gemma 3 QAT Models: Bringing AI to Consumer GPUs

https://developers.googleblog.com/en/gemma-3-quantized-aware-trained-state-of-the-art-ai-to-consumer-gpus/

Google Releases Gemma 3 QAT AI Models for Consumer GPUs

#AI #AIModels #GoogleAI #Gemma3 #LLM #OpenSourceAI #GPUs #QAT #Quantization #DeepLearning #MachineLearning #NVIDIA #RTX3090 #Kaggle

https://winbuzzer.com/2025/04/20/google-releases-gemma-3-qat-ai-models-for-consumer-gpus-xcxwbn/

I confronted a honey badger character I made with gemma3 with a genie intending to figure out what it would wish for, but instead it immediately insulted the genie. Then I thought maybe I could make it interesting and make each wish require the llm to beat the genie in a battle of wits.

The first wish was the genie should stop smelling like inscence and instead produce a structural material analysis of its lamp.

I don't know what I expected but it certainly wasn't this LMAO

Running Gemma 3 locally with no API costs? Yes.
Free to experiment? Absolutely.

We built a full-on comment processing system using Docker Model Runner + #google Gemma 3 — no cloud, no third parties. Just local power.

Try it out https://www.docker.com/blog/run-gemma-3-locally-with-docker-model-runner/

Some local #LLMs tested on an average #gamingPC: https://github.com/donatas-xyz/AI/discussions/1