Assessing AI Options + Perplexity

I’ve been assessing AI options lately. ChatGPT, Gemini, Claude 3, Mistral, and the like, but also more towards how I’m meant to pay for them.

I’ve always been paying for ChatGPT Plus for over a year or two already, and honestly, the thought of paying $20/mo for each service is just not worth it. Even for ChatGPT Plus, because I feel like I’ve been happier when using other services in other ways when GPT-4 couldn’t.

See, GPT-4 is the gold standard and I’ve been very happy with its output since it was publicly available. But at the same time, it’s also slow as hell. And feels even slower when Gemini / Gemini Advanced exists. I realized that I use these AIs more as a digital assistant rather than a research tool, although I use both and are equally valuable. But in most cases, speed and convenience are actually my biggest draws for the service, and not always the quality of responses.

I decided to leave ChatGPT Plus because of this particular pain point, and oddly enough, it’s something I wouldn’t even leave if their webapp and mobile apps didn’t insist using GPT-4 all the time, or if you could at least choose a default or at least remove one tap or click away from toggling between the two options. Because yeah, sometimes, I just want answers fast and even GPT-3.5 can do this very well.

Enter Perplexity#

Now I’ve been trying Perplexity lately and I am loving the service. It took me a while to get the most out of it, but now I’m getting the hang of things.

For one, Perplexity actually goes beyond the usual chat experience and offers something that’s in line with what I want — an interface meant for search and speedy results, while also offering the same chat for follow-ups and such.

But I like it even more since you can choose the model that you want, and even the default Sonar model they have is quite nice. I like that you can rewrite with a different model if you really want GPT-4 or Claude 3 to come in. And unlike my pain points with ChatGPT, I can choose a default, and I have toggles for fast search, or if I want to dive deeper (enable Pro, and switch between focuses)

And they have a decent Android app too, which is a must since I need to be able to query very quickly on the go too.

Things I really enjoy with Perplexity:

  • Focus switching - for their approach on search, or for the classic text generation with Writing mode. I need to explore the other options too, like the Reddit and Academic modes.
  • Rewrite option - if you want a particular query addressed by a model, this button lets you choose.
  • Response layout - I like that sources are up top, in a row, media on the right, and the answer following below, and related options at the bottom. It’s good for making me think deeper with what I’m looking for, and to expand the query further.
  • Collections - I’ll explore more of this feature further, but basically, I see it as a means to have preset system prompts, which is very handy! ChatGPT makes use of Custom GPTs for this kind of stuff which is more powerful.
  • Address bar search + fallback searches - while not a feature of Perplexity directly, I like that you can compose searches via https://www.perplexity.ai/search?focus=[focus]&copilot=[true|false]&q=[query] and I’ve used this to replace my default Firefox search and Alfred fallback searches.

Conclusion#

For now, I think my setup will look like this:

  • Unsubscribe to ChatGPT Plus
  • Subscribe to Perplexity Pro
  • Use Perplexity Pro nearly all the time
  • Use OpenAI API / Claude 3 API / etc for uses beyond what Perplexity Pro can manage, like if I need contexts or features beyond what Perplexity can offer (e.g., for code, for extensive research, and for code interpreter)

There’s stuff I haven’t mentioned above, but I do have other AI stuff still subscribed too for the sake of conveniences:

  • Notion AI, because it’s handy to have in Notion (both the Q&A features and working inline)
  • Gemini Business, because of sensitive data (not negotiable). Although the API can be used for this kind of stuff, I need it running in Google Workspace.

I might axe these both too, if I can just offload things to a proper API workflow once I have one running.

comments powered by Disqus