Is it really worth the panic?
I didn’t think I would ever quote Trump on a subject such as this, but he did say it best: “I view that as a positive, as an asset… Instead of spending billions and billions, you’ll spend less and you come up with, hopefully, the same solution.”
I have used DeepSeek throughout this week and it is good, but not alarmingly good enough to force me to use it over ChatGPT.
The big news – and unfortunately one we cannot independently verify – is that it took much less to train. There are some reasons for that:
– It used a lot of the data from ChatGPT to train (that was verified) and
– It compromised on the search algorithms, meaning if ChatGPT will use a higher probability to get to an answer, DeepSeek will get to 60% and call it good enough.
As probabilistic computers have shown us, we can approximate answers and be right most of the time. This means a smaller model will come to the same conclusions as ChatGPT got, but with less compute power. And BTW, if you optimize for these AI benchmarks, you will look just as good as the big boys.
Only real workloads will show a difference.
Whether this news is worth a 15% drop in Nvidia’s stock is highly unlikely. As we have seen today with the major bounce back up, many others agree.