Correction: Gemini Advanced (i.e. Ultra) is as good as ChatGPT4
It is worth the price if you pay for it
Last month, I wrote that Gemini Advanced (also known as Gemini Ultra, Google’s paid version of its ChatGPT competitor) is not worth the price. Turns out, I was wrong. I’ve been using it alongside ChatGPT4 for about a month; in most cases, I’ve been giving the same prompt to both of them and comparing the results. And, except for the fact that it still doesn’t read uploaded PDFs, it is as good as ChatGPT4.
In the first few days, I gave too much importance to the lack of PDF uploads, but over time I’ve realized that I don’t use that capability as much. Also, in the early days, Gemini was refusing too answer too many questions, but Google tweaked the settings so this doesn’t happen as often now.
So overall, in the last month, sometimes Gemini Advanced has done better than ChatGPT4, sometimes ChatGPT4 has done better, but most times, their answers have been roughly comparable. And there were tasks in which both of them gave me wrong answers: so always be careful.
Vishnu found that Gemini Advanced does a much better job of summarizing a document than ChatGPT4.
I haven’t done much summarization recently, but yes, ChatGPT4’s summarization isn’t great. So, if you are going to be doing a lot of summarization, I suggest trying out Gemini Advanced.
So overall, I would say:
Gemini Advanced is as good as ChatGPT4
If you can afford it, you definitely must use a paid version, either ChatGPT4 or Gemini Advanced
As far as I can tell, right now there is no need to take the paid version of both them: they are roughly comparable, and having just one of them is good enough. But all the free versions are worse; so don’t base your opinions of what Gen AI can or cannot do based on the free versions.
They can both can make mistakes, so always be careful
More exciting stuff coming: Anthropic’s Claude3 (not yet generally available) is also supposed to be as good as ChatGPT4/Gemini Advanced. And Gemini 1.5 can ingest 1 million tokens at the same time—which will solve one of the most important shortcomings of the current state-of-the-art Gen AI’s the limited context window
In general, stay tuned.
Time to test Claude 3. I'm looking forward to your article on comparing the paid models with other, Navin.
Any use case or two, which are your top of the list to try out with a longer context window?