Blanca
715 posts, incept 2020-07-25
2025-01-27 07:58:02
I was wondering when you were going to weigh in on DeepSeek. Yesterday afternoon I was talking with a friend about the implications for the traditional AI companies and the need for all of the power hungry GPUs if DeepSeek's claims are true.
Some years ago I was concerned about the slow execution speed of what should be a simple translation from one format to another using commercial generic software. The architect's answer was we'll just throw more processors at it. I wrote an equivalent translator that only required one CPU and one thread of execution and it was an order of magnitude faster than the commercial software using many CPUs. I gained no traction with the architect. Nah - can't use this, it isn't standard, and the company paid a lot of money for the commercial software and is in a special relationship with the company that produced it, etc. I was the bad guy for exposing their crummy software. I got into arguments with the architect. Eventually, I found another job and left the company.
I suspect that there will be a lot of resistance to DeepSeek by the traditional companies because it exposes them for what they are (if DeepSeek proves to be the game changer as claimed). Stay tuned.