[rrsteer]

I am a heavy user of generative AI tools. I use it at work and also as an aid while teaching my daughter.

At work I mostly use it to synthesize and summarize text, translation, create text from images, help in creating charts and for research. While teaching/helping my daughter with her studies, I mainly use it to cross-check answers to math problems, help with languages and grammar.

I wanted to get the advanced paid version of one of the popular Gen AI tools. Basically either the OpenAI's ChatGPT or Google's Gemini. Having used both of them for a considerable amount of time, I came to the sad conclusion that ChatGPT >> Google Gemini. Sad because, I have alway harboured a soft corner for Google products. I generally feel their products are amongst the best and easiest to use. Hence, have made an effort to stay in Google's ecosystem. In contrast, I have nothing but hatred for Microsoft and its products. As S.Jobs said, and which I feel is very true, the company and its products lack soul. Even though ChatGPT is more Google than Microsoft, but since it's backed by Microsoft now, I equate it to being a Microsoft product.

I will share two examples to demonstrate my point - one from each of my main uses cases.

Work - capturing data from a picture based chart.

The picture for which I wanted the AI tools to create the data:




Gemini's response:




ChatGPT's response:





Teaching Aid - checking the answer for a math problem

The problem:

A garden is 357m and 50cm long and 76 m and 25 cm wide. It is to be paved with square shaped tiles of same size. Find the largest size of tile that can be used and the least number of such tiles required.


Gemini's response:



ChatGPT's response:




As you see above, Gemini either couldn't do the task properly or its answer was incorrect. I have rarely seen ChatGPT fail at what I throw at it, in contrast, Gemini flounders quite a bit. Hence grudgingly I ditched Gemini and got a ChatGPT Plus account.

Disclaimer: I am not a technology person, just a consumer of technology. So my comparison is only based on how well the two platforms performed for me. It maybe that Gemini is more advanced than ChatGPT or vice versa, I don't know and haven't invested time on reading about that.

Final thoughts: The more I engaged Gemini, the more I got the feeling of using a Microsoft product and the more I used ChatGPT, I got the feeling of using a Google product. E.g. if I ask for a synonym of a word - ChatGPT gives a list of words - thats it, and that's what I need. In contrast, Gemini will give me a list of words and then additional list like informal synonyms, more specific synonyms depending on context and ends with a question - Would you like synonyms for a specific context? - No and Thank you very much!

[Fillmore]

Originally Posted by DigitalOne

AI is not replacing you (yet)!!

https://Youtu.be/vB9dJt9j-5M?si=_8bEwPgxqCZ-LLTg

This was not what I got when I tried the same question

[Thad E Ginathom]

Oh good grief, that's weird reasoning.

I'm innumerate, thrown out of the maths class at school. Here's my poor-at-maths inteligence anser:

nine-tenths is bigger than one[-point-one] tenth. That's it.

[DigitalOne]

Originally Posted by Fillmore

This was not what I got when I tried the same question

When I saw the YouTube video on 13th Aug and tried it, this was the response I got :



GenAI tools are still inconsistent with the responses.

----

Here is a frustrating attempt, a few days back, to get ChatGPT to help me solve the daily Wordle puzzle. Given that these are "Large Language Models", I expected it to better than this.

[onwheelengineer]

Originally Posted by DigitalOne

GenAI tools are still inconsistent with the responses.

----

Here is a frustrating attempt, a few days back, to get ChatGPT to help me solve the daily Wordle puzzle. Given that these are "Large Language Models", I expected it to better than this.

This is very true, infact the models would remain inconsistent and the constant rising fear would be subsided when we all look at how these things work. In normal terms it converts sentences into a vector or a sequence of numbers and end up predicting the next in the sequence. That's why it makes such mistakes that seem like its incapable to say the least.

Although the new OpenAI model (OpenAI o1) seems interesting and in contrast to this. Its developed with a different sort of training, its still a model that predicts the next in the sequence but what it does now is that it divides the one big task asked to it as a set of small tasks then follow that and solve it one by one arriving at the final answer.

This is called Chain of thought prompting. It's shown some good results especially in math and research application, even scoring 83% for the qualifying exam for international math olympiad, but still since its a next step predicting model, it goes wrong in one step, the entire sequence may mess up. Infact if you look at the earlier models by OpenAI or google itself you can see that sentences generated by them would seem quite bad (it would generate two words relevant to the context, then the sequence would go bad and there'd be random words that doesnt even make sense).

AI has come a long way, but it still has a long way to go, with some news suggesting that there are improvements trying to be brought about to make sure the sequence doesnt get badly messed up by just a wrong word prediction, although there's a lot of algorithms for the same, none is foolproof and will require human monitoring to rectify. Humans aren't going anywhere soon, repetitive and menial jobs would though.