4 ms·
I think it depends on the type of content. I've primarily used ChatGPT for additional explanations of programming terms (e.g. C++ lambdas & capturing local var
by albntomat0 4y ago
I think it depends on the type of content.
I've primarily used ChatGPT for additional explanations of programming terms (e.g. C++ lambdas & capturing local variables, validated against my code and other sources). The results from Google for the same query are a bunch of overly long, SEO'd crap, covered in ads.
There are plenty of other things I've searched for recently, were ChatGPT simply doesn't make sense. One example is looking for trusted reviews of kitchen appliances, and things to do on vacation. I need to be careful with those as well, given the incentives around human reviews. I'm skeptical that market will be disrupted by a language model, while also being the most relevant for advertisers.
- webmaven 4y ago> There are plenty of other things I've searched for recently, were ChatGPT simply doesn't make sense. One example is looking for trusted reviews of kitchen appliances, and things to do on vacation. Right now, ChatGPT doesn't provide good answers to questions in the form of "What is a good source for information of type X?", but that is really a limitation of the source data (text from a current web crawl will, superficially, tend to reflect information rather than meta analysis), rather than the model as such, and transforming PageRank style graph data into a verbalized form suitable for an LLM to be trained on ought to be possible. Something equivalent has already been done for knowledge graphs, for example: https://arxiv.org/abs/2010.12688 https://arxiv.org/abs/2010.12688 https://arxiv.org/abs/2205.08184 https://arxiv.org/abs/2205.08184
- albntomat0 4y agoInteresting! Thanks for sharing. I’d be curious to see how such things, if implemented, fare under the adversarial conditions that current reviews and SEO are in.