On 2/28/24 23:18, Hannes Reinecke wrote:
Now there's a thought. Maybe we should do exactly the opposite, and
posting _more_ ChatGPT generated content on the web? Sending them
into a deadly self-enforcing feedback loop?
That's a rather negative view. Besides concerns about the energy spent
on training LLMs, in my opinion LLMs can be very useful. If I need to
look up information about a kernel subsystem that I'm not very familiar
with, I type in the question in the user interface of an LLM. While it
can be very hard to look up information in mailing list archives with
a traditional web search engine, LLMs can give accurate answers quickly.
Here is an example of such a question: "Why does the Linux kernel cgroup
implementation propagate statistics to parent cgroups periodically
instead of immediately?" Both ChatGPT and Gemini produce reasonable
answers to this question but the web search engines that I tried did not.
Bart.
