$Datadog(DDOG)$ 在JPM会议上这段对话算不算LLM reality check

Analyst：
Okay. Now so let's take a few moments to get into the topic of generative AI. This has really become the single biggest topic for us. I keep pointing out it's evolving at light speed. And I think some of the capabilities, they're amazing, they're a bit frightening. The -- we look at this -- so as it relates to the monitoring market, the bulls are saying, well, generative AI thrives on large data sets, right? Observability players have some of the largest data sets that we're aware of out there. So it's going to be positive. 
And then you listen to a bear and they would say, well, just take all these logs and metrics in traces and just stream it directly into -- whether -- into ChatGPT, but stream it directly into one of these large language models and ask it which apps are running low on memory, right, or where servers are low -- something is low on storage, and then ask it to actually go and remediate and see what will happen, which is a hypothetical discussion. But Datadog is known for being well ahead of the curve, what do you think generative AI is going to be in for Datadog?
Olivier Pomel
So first of all, just to put the numbers in perspective to trying to send observatory data directly into large language models would be orders of magnitude more expensive than running the application, the policies that data to start with. So I think our largest customers would probably have to spend the GDP of California just to get to make that meat happen. And I don't even -- as of today, that doesn't even work, right? That doesn't even give you the answer you want.
So from a technical perspective, the answer is still going to be you have to gather the data as smartly [ in West ] as low impact as possible. You have to understand how to summarize it in a smart way, then you have to understand which models you can train that are very, very specialized and very, very cheap to run so that you can then maybe combine that with some of the more expensive and more general purpose models to try and tie things back together.
...