I don’t think any of us were ready for how quickly productivity could multiply when we had the right AI tools and the right hardware to use them. But a lot of what we do routinely now would be fantastic only a couple of years ago.

Intelligence, both human and artificial, flourishes through collaboration, enhancing and empowering those it interacts with.

One area that we are particularly concerned with is allowing our customers to stay free of the cloud and all the fees and connection challenges involved with that. Instead, we are looking at low-cost hardware enhancements that we can provide to customers that can’t afford an A100 GPU but wouldn’t mind paying for a small accelerator unit which they would buy once and own forever. The challenge is in gaining the computational improvements of a GPU without requiring a truckload of expensive graphics memory.

Toward this end, we are investing in a customized accelerator that is sufficient for small LLMs and doesn’t break the bank for the customer.

Another area that we are exploring is the use of older GPUs for running our LLMs. We will be looking at technologies such as HACK which can smooth the use of key VDRAM memory by paging Q, K, and V multiplication blocks.

SENTIASYS IS ALL ABOUT BRINGING AI TO VERY CONVENTIONAL COMPUTER SYSTEMS

SentiaSys is committed to following the latest research in hosting small LLMs and leveraging that technology for the benefit of customers.