… and to add additional credibility, he’s “on the Spectrum”.
intercst
4 Likes
That was a great explanation. I had to watch it twice. So Deepseek was trained on public LLM’s that they combined together to train their model. But they have bigger problems with halucinations than other models. Very interesting. Not the best but maybe good enough.
2 Likes
As I said in another thread, McDonald’s made a fortune selling “not the best but maybe good enough” burgers, and a lot of applications for AI (responding to customer complaints, monitoring, etc.) might be just fine with a higher error rate than, say, the Pentagon using it to search for drone swarms in the midst of battle.
4 Likes
I agree but if they can do it then a lot of other companies are going to do it. This really commoditizes the lower end. Unless all the big LLM’s make all their open models, private. But it might be to late for that already.
1 Like
That was a good explanation of what it is.
Here is another one:
Thinking people are naturally suspicious but we should not confound that with the general environment that promotes suspicion of everything out of China uncritically.
I don’t think we should react to DeepSeek like that.
It is an innovation. It is a disruption. But I would be short of calling that a Sputnik moment. That hasn’t come…yet? We should copy them and learn from them like they have done with ‘our stuffs’ instead of repeating mantras that are tired and meaningless. The US itself, Japan and the Asian Tigers went through this same economic phase as China did. There is nothing to see there.
DeepSeek was imposed different constraints so it had to play a different game. We should ask what can the result of that enable? DeepSeek shows a different way that could be beneficial to many other businesses than the usual AI suspects.
We should be thankful for such an innovation instead of trying to negate a fact.
tj
4 Likes