The tech circles are buzz with latest release by Open AI – GPT-4o. And you know things are getting hot when OpenAI decides to schedule its reveal just da day before when Googles IO was scheduled, which ended up showcasing an almost similar product – or was it similar? Well I do not use Google Gemini much, so cannot say much.
Coming to GPT-4o – I thought I will write a post about it as I am heavy user – not of the API from from the normal frontend as a premium user. And from my perspective things seems to have taken at least as big a leap as it took from 3.5 to 4 if not more.
GPT-4o seems like GPT-4 and 3.5 combined together, just better. Its so much faster than GPT 4, and so less buggy – though far from perfect – however the crashes and mood swings have gone down a lot.
I have been using it as a coding assistant for past few months, and the amount of code that GPT 4 can handle, versus the amount that GPT-4o can now handle has changed drastically. Writing a 250 line dart code file was and is such a pain in older version, however the newer version just cannot stop writing code.
Older version was so frustrating because of all the place holders it used to generate, and random things it would sometimes do, however the new ones seems to be super precise, much much faster and seems to be happy to generate code at drop of an hat. Talking about lines of code – Claude though so good, I think surrenders without an error at 200 odd lines, it just – mean – nothing – lol.
I think that code generation behaves differently for different languages it works much faster an smoother for python and javascript, however for dart and flutter it takes a bit longer time. In Fact both older and new version seem to fail more often when writing dart code compared to other languages I have tried on it.
But there is a difference in how the newer GPT-4o fails now – the tab and code generation might seem stuck, however if you open the same thread in different tab, you will find that either it has completed generation or you will get an option to continue with the generation. Thats probably about it. The code is already there, the UI in other tab is stuck probably due to memory or streaming issues, but the backend has already done its job.
GPT-4o seems to be least lazy version for writing code – at least for Dart and Python I have been putting it to use for. And bash commands and settings etc – working like a charm. In Fact sometimes its frustrating that is is generating so much code, more than eager to generate code – but the good part it is consistent and not messing about with it.
Btw – all this using a custom GPT called Grimoire – which I find better than vanilla GPT and other coding GPT assistances of chat GPT itself. And above experience is using Grimoire on both older and newer version of the Open AI’s product.
And a very crucial point – as a premium user I am not running into limits any more. There are 2 meanings – the responses have become more efficient, elaborate, and useful. And secondly the company might have revised or removed the limits for premium users.
Have a look at this project here, this is 100% done using various versions of ChatGPT. So if you are new to coding or are keep to learn, I would say spend some time raking your brains on free or premium version and see it carry you much farther than expected.
Link to Github:
https://github.com/shashankaggarwal/MacroTracker