I have recently relied heavily on a certain chatbot for online research, accessing information across various websites efficiently.
Currently, I use the chatbot 80% of the time and a search engine only 20% of the time for information retrieval. I am curious about others’ experiences:
Are Techlore members shifting from search engines to llm bots to find information on the internet?
Perhaps a Techlore member has a fried or family member who is in the same situation as I am. If so please let me know.
I don’t know if this particular chatbot is storing my searches and inquiries. I can’t say if it’s privacy orientated or not. But because it’s so effective, I can’t help but not use it. Any comments or questions concerning my questions would be greatly appreciated.
As for me - no. For several reasons.
- privacy (unless you’re running your own)
- chatbot is usually quite out of date
- different llms for different areas
- I need to form everything into a sentence vs just 2, 3 keywords
That said, I use phind.com from time to time with various degree of success. Sometimes it help me solve the problem, sometimes doesn’t understand and runs in circles and sometime gives me an idea or point me in the right direction.
This is a very valid consideration. But is this meaningfully different than a search engine? In either case to use the service you must necessarily share with the service your queries.
- chatbot is usually quite out of date
True in many cases, but not a fundamental limitation. There are publicly available ‘online’ LLMs
(as an example, Perplexity.ai was able to find this thread, which is just 11hrs old)
I think there are both pros and cons to this.
Search engines tend to be good at shorter queries, but tend to be not very good for longer more specific or nuanced queries. LLMs are better at those longer queries (often), not sure about the shorter queries, I tend to write prompts for an LLM in more natural language which tends to be longer, so I haven’t tested many short queries. (I did just try 3 short vague queries and got satisfactory answers: eurusd and baltimore bridge collapse and weather today)
I especially like being able to ask LLMs followup questions or challenge it/refine the question. Of course there are many frustrations and shortcomings to LLMs as well.
I’m finding that there are some queries and topics that I’ll default to traditional search, and some queries and topics where I’ll default to an LLM. I would never want to rely fully on an LLM, but I like using it as a complement to traditional search.
I don’t personally feel either a traditional search engine or an LLM is objectively better or worse than the other, and I don’t foresee one replacing the other for my workflow, both have their place, both have pros/cons. Also the line between the two is getting more blurry (traditional search engines have been integrating AI-like features for some time, and some AI companies (e.g. Perplexity) are building services that keep the UX and features of a traditional search engine and incorporate those into their AI/LLM services.
I’d say I’m somewhere between 90/10 and 70/30 (traditional search / LLM). Though for some subcategories/topics, that ratio is flipped (some technical topics for instance).
You can (and should) read the privacy policy (you could even use your favorite chatbot to help you analyze it). It is almost certainly not privacy-centric, I’m not aware of any prominent AI services that are. But there can still be meaningful differences between non-privacy-centric services. Short of self-hosting, which is not cost effective for most people, the only public/hosted LLM I know of that could be considered remotely privacy-centric would be Brave Leo, but that has its limitations and caveats.
It really depends on what you ask for, how you ask and which bot(not) you are using. Without these any comparison will be incomplete.
For instance, I heavily use perplexity and it completely changed my search behaviours. Especially, if I am looking on information from Reddit, I select focus mode and it generally provides good answers than search engine.
LLMs enable you add context or examples, so they can find relevant information. With all caveats, most LLMs are better than search engines on privacy questions, especially without internet scanning. Because it eliminates the seo garbage.
I use different LLMs for different purposes such as ChatGPT 4 for work, perplexity for internet search and huggingface without an account for sensitive questions. You can use perplexity without an account but I have one because at the end of the day, I have very few sensitive data or questions there.
I have to say i’m quite solidified into search engines for now and honestly i don’t even use LLM / AI tools in general. Not because i consider these not good enough but because i just don’t find enough reason to make the switch.
I do track AI news and they do sound interesting and exciting to learn them but with the life i live, i just don’t see that much value in using them (at least for now). I guess i do sometimes use Brave browser and read on occasional results from it’s website summaries, although that sounds a bit like a hybrid search in my eyes.
Coming back to searching information, i mostly watch YT videos from my subscriptions and have the homepage videos completely blocked and disabled. When i use search engines, it’s quite easy for me to spot a bad quality results or i default to sites i already have trust in, like for example Healthline for many diet or health related info.
So yeah, in short, what i’m using right now serves me well but if i will have enough reasons in future to try Chatbots. etc. more, i likely will.
Interesting that you’re writing about YT. I took the title from a YT channel and then entered it into Perplexity, asking it to describe what the video was about. I could tailor the response to be an extremely long or very short version. I could also find out if the video was clickbait or not.
It takes a person on average just a few minutes to read through an article than to look at a video that could last as long as 20 minutes, which may contain a lot of fluff. Alternatively, I could use a bot to get the key information I need in roughly a minute.
I do get that approach and for example sometimes i see comments doing TLDR’s that boil down the video chapter to a succinct sentence or two and these sometimes get quite a bit of likes.
I can see that working for certain people but in my experience that might condense the video too much, leaving behind potential context and valuable extra around the topic.
As an example i might sometimes watch entertainment content like reactions to my favourite shows and some videos are 1/3rd reaction to show itself and 2/3rd’s are the reactors expressing thoughts, feelings, theories etc. that could be considered extra fluffy as hell but are still enjoyable as they give fresh perspective on things i might’ve not even thought about. This can similarly apply to educational stuff for me.
In the end i think an important thing is to choose what we think is best for us and which we’ll regret the least.