Intel AI Playground

[05032-Mendicant-Bias]

Summary

Intel released a beta version of its local GenANI playground for local LLMs and diffusion models with ARC GPU acceleration and Core NPU acceleration.

 

Quotes

Quote

AI Playground beta is a free to download and use AI PC starter app, that makes AI easy on Intel Arc GPUs. AI Playground provides a variety of easy-to-use AI capabilities to create and enhance images, or get answers from an AI chatbot, all running locally on your own PC powered by either an Intel Core Ultra-H processor or with an Intel Arc dGPU with at least 8GB of vRAM.

 

My thoughts

Intel is ramping up support for local, open-source GenANI models, providing competition to the near monopoly of Nvidia CUDA/Tensor acceleration. This development has several positive implications:

1. Cost Reduction: Increased competition means lower costs for model access and acceleration for consumers.
2. Focus on Local Models: By emphasizing local models, tech giants can invest effort in small foundational models that run on their devices, rather than relying solely on monolithic models that require warehouses full of resources.
3. Openness: Local models are more transparent. If a model runs on your device, you can inspect its weights (unless homomorphic encryption becomes widespread, hopefully NOT!).
4. Privacy and Independence: Local models allow users to maintain privacy and avoid being subject to tech giant censorship. E.g. Bing Chat refuses to tell me when is USA election day!
5. Reduced Latency: With local models, you can achieve real-time performance without the several seconds of latency associated with cloud-based services. E.g. STT TTS live translation that Human tried and didn't work because of cloud latency.
6. Scrutinizing Censorship: Open models enable scrutiny of any censorship applied to the model, allowing for fine-tuned variations (similar to diversity llama models).
7. Predictions: As cloud computing costs rise, tech giants may shift compute costs to local devices. Microsoft plans to integrate local Copilot models with Windows, Apple aims for a local Apple model, and Google may eventually create a comparable Gemini local model that is not useless.

The future looks promising for GenANI assists!

 

Sources

 https://game.intel.com/us/stories/introducing-ai-playground/

 

 

[monopoly_hater]

Stuff like this is intel’s only hope with the current state of things.