Gemini Pro is quite good, probably better than GPT-4, but it’s still terrible at doing a search engine’s job reliably, and Google is undoubtedly using an older, cheaper setup here.
It’s mind boggling that Google shoved this into search the way they did.
Could you recommend any? I’ve been using ChatGPT and it’s been helpful, but I’d love something more specific to the task.
As an example, there was some kind of event happening downtown near me last night. Really bad traffic and roads closed. I tried figuring out what was going on. Checked the city’s site, the Facebook page, googled around, couldn’t find anything. I asked ChatGPT and it told me it was the annual Christmas parade that started at 7:30, it’s route, and where I could catch it from where I was. Say what you will about AI, but that was unbelievably helpful.
Depends on how deep you want to integrate AI. It ranges from learn to rank(AKA AI sorting) and transforming user input(ha! pun!) to vector search(transforming user input to special format) and learned index(AI search). And then there are whole area of AI, LLMs specifically, wchich sits on top of search engines: retrival-augmented generators(search-and-generate).
Speaking as a local LLM enthusiast, it’s still unreliable and is “googling” something behind the scenes to get answers. And it’s not like a human poking around, internally it’s looking at a wall of webpages and “hoping” to hit something right.
But to answer your question Perplexity is similar, and perplexica is the open source equivalent. There are also researcher “agent” scripts that will script LLM calls.
I get that, and for sure always need to verify it’s results, but SEO has turned looking information up into a slog of AI articles and lists.
I very rarely am able to find the answers to my questions anymore. And I say that as someone who has always joked that my job was a professional Googler. Whereas I used to be able to search key words or vague snippets of something from memory or lookup a hyper-specific question and find a solid answer within a couple minutes, now the first 2 pages are ads and “websites” with “articles” that are just vessels for more ads.
Using something like ChatGPT gets me a little closer to that old level of access. The hand that feeds I suppose…
Just remember to take anything AI says with a grain of salt. TBH I’d suggest digging deeper into how LLMs work at a high level so you can understand just how they can be confidently “wrong”
Oh I understand how they work, I actually made an AI chatbot for a client of mine. Now, to be fair, it was mostly just fiddling with and fine-tuning a prompt in a WordPress plugin. It’s called AI Engine by Meow Apps. Super cool plugin, but It was the most frustrating project of my life because despite being given clear and consistent instructions, and having a dedicated database to pull answer from, it still just made shit up. I set the temperature to .1 and it lost all personality and still made shit up.
Finally got it to a usable place where it basically was just walking the user through a quiz but, yeah, I definitely take it with a grain of salt lol
Gemini Pro is quite good, probably better than GPT-4, but it’s still terrible at doing a search engine’s job reliably, and Google is undoubtedly using an older, cheaper setup here.
It’s mind boggling that Google shoved this into search the way they did.
I don’t think goal of Gemini was search. There are better tools for search, including AI tools.
Could you recommend any? I’ve been using ChatGPT and it’s been helpful, but I’d love something more specific to the task.
As an example, there was some kind of event happening downtown near me last night. Really bad traffic and roads closed. I tried figuring out what was going on. Checked the city’s site, the Facebook page, googled around, couldn’t find anything. I asked ChatGPT and it told me it was the annual Christmas parade that started at 7:30, it’s route, and where I could catch it from where I was. Say what you will about AI, but that was unbelievably helpful.
Depends on how deep you want to integrate AI. It ranges from learn to rank(AKA AI sorting) and transforming user input(ha! pun!) to vector search(transforming user input to special format) and learned index(AI search). And then there are whole area of AI, LLMs specifically, wchich sits on top of search engines: retrival-augmented generators(search-and-generate).
Thanks so much for the material! I’m going to comb through these tomorrow.
Speaking as a local LLM enthusiast, it’s still unreliable and is “googling” something behind the scenes to get answers. And it’s not like a human poking around, internally it’s looking at a wall of webpages and “hoping” to hit something right.
But to answer your question Perplexity is similar, and perplexica is the open source equivalent. There are also researcher “agent” scripts that will script LLM calls.
I get that, and for sure always need to verify it’s results, but SEO has turned looking information up into a slog of AI articles and lists.
I very rarely am able to find the answers to my questions anymore. And I say that as someone who has always joked that my job was a professional Googler. Whereas I used to be able to search key words or vague snippets of something from memory or lookup a hyper-specific question and find a solid answer within a couple minutes, now the first 2 pages are ads and “websites” with “articles” that are just vessels for more ads.
Using something like ChatGPT gets me a little closer to that old level of access. The hand that feeds I suppose…
Anyways, thanks for the tip. I’ll check it out!
How capitalism ruined internet
Just remember to take anything AI says with a grain of salt. TBH I’d suggest digging deeper into how LLMs work at a high level so you can understand just how they can be confidently “wrong”
Oh I understand how they work, I actually made an AI chatbot for a client of mine. Now, to be fair, it was mostly just fiddling with and fine-tuning a prompt in a WordPress plugin. It’s called AI Engine by Meow Apps. Super cool plugin, but It was the most frustrating project of my life because despite being given clear and consistent instructions, and having a dedicated database to pull answer from, it still just made shit up. I set the temperature to .1 and it lost all personality and still made shit up.
Finally got it to a usable place where it basically was just walking the user through a quiz but, yeah, I definitely take it with a grain of salt lol