Background

This year is shaping up to be the year of the AI PC. Major manufacturers are focusing on offline and private AI, launching their own open-source software ecosystems (with a host of pre-trained models) to encourage end-users to buy more graphics cards and CPUs so they can train their own AI.

Computex in Taiwan is mostly a showcase for hardware manufacturers, with fewer software and AI solutions on display. If you want to see software solutions, you should probably go to CES in the US.

I think the cost of training your own models is still too high and can't compare to what OpenAI offers. However, there are more opportunities to use them in specific, limited scenarios or offline environments. One day, when open-source models are mature enough, we'll be able to take them, fine-tune them, and put them to use.

Offline AI Solutions

I spent a lot of time talking with hardware vendors and have put together a summary of the offline pre-trained language models and software ecosystems that were prominent at Computex.

• 1. Intel - OpenVINO => After training, models can be deployed to more lightweight IPCs. Related Link
• 2. Nvidia - NGC, ChatRTX Related Link
• 3. Ollama - Open-source and free. Can be used with language models from Google, Facebook, and OpenAI. Related Link
• 4. Fine-tuning language models Related Link

Photos from the Show Floor