Gemini Deep Research Source Selection

So this evening I’m beginning my research on where an LLM might work well and where it might not. Of course, based on my first post on this topic one might think AI is terrible at everything, however I feel like that might be too great a generalization. I have not been impressed with output for software development, but that doesn’t mean it’s useless. So as I start my research tonight, I began by just going straight to Gemini 1.5 Pro Deep Reasearch, because why not? Obviously it’s output is going to be considered “sus” until proven otherwise, however it does also just barf a ton of sources at me that I can review.

In that vein, I have learned something however: if I want good sources, instead of an article on Forbes written by the CEO of an AI company who bought his way into the role of Forbes Tech Contributor, I need to ask for them. It appears Gemini is really receptive to this kind of prompt too. If I ask for science journals and non-profit sources, it favors those over things like Reddit posts. This may not apply to all LLM or AI systems – certainly if I try this on my Open-WebUI setup on my desktop, it’ll probably see a low level of success. But for the cloud providers, it’s definitely worth a shot and certainly for Gemini, I need to remember to be more explicit in my prompts for what I want. Generally I tend to be more open-ended. That helps me rubber duck by letting the LLM do a lot of riffing, but I do also need to consider constraining it in more important areas such as source quality.

So, if you’re using these systems for research or assitance or whatever and it’s pulling in low quality sources for what you’re trying to do, try asking it to restrict its source material. For example: if you’re asking for beauty tips, maybe suggest certain TikTok creators or if you want recipes ask it to limit the search to dedicated recipe sites. This should work across all kinds of diciplines or needs so I would say don’t be shy about asking for these restrictions. I have been but I’ve also humainzed these systems a bit too.

Obviously don’t treat them poorly, on account of the upcoming robot uprising – be a friend to our robot overlords. But also, be sure to be explicit in your needs!

Other LLM Benefits

One thing I also ran across tonight in my initial quest for research on this topic is one of the sources pulled in by Gemini was in Korean. I don’t speak Korean, but the LLM can speak anything! So that could be a neat and useful aspect to doing research with an LLM or AI system. Such a system could source information in any language and provide that to the user in their preferred language. Perhaps another thing I’ll need to review for accuracy.