GPT-3.5 vs GPT-4: Which AI is Better at Complex Machine Learning Tasks? 

I was able to get this to work with GPT 3.5, it just took a lot of back and forth and sometimes combining its different outputs, great video! fun exercise. 👍

Keep up the awesome work man! Learning a lot from you

Awesome and unbiased comparison. keep up the good work buddy 👏👏

I only have access to 3.5 but my experience was similar. I wasn't trying to recreate already established algorithms, but advance a couple of my more mad science type projects. After running the code I would get errors that I copy/pasted back to GPT. It generated new code. Wash rinse repeat. After a few rounds of this I found GTP kind of looping back on itself and recreating bugged code that it had corrected previously. While it did give me some hope for the plausibility of my projects (gpt was able to fill in portions of domain specific knowledge that I lack), ultimately I was not able to solve my use cases to any satisfactory conclusion. I have no doubts though that future versions would be more effective, but given the growing amount of backlash to large language models in general I have to wonder whether such models will ever be released to the public. Great video! 👍👍👍

From scratch is Good. People optimize code to better suit natural language. If you handle the natural language processing on the client actually has Tremendous impact on qualify of response.. Its like tree of thought but for the transport layer.

Interesting video!

Great video, very underrated.

Great video. I would suggest in 3.5 to first prompt wisely an expert ML developer role instead of cold start it.

GPT 4 17:57

Awesome!

IMO gpt4 is so great. I had an idea for a IOS app that I wrote halfway 2 years ago before giving up. Its really frustrating learning Swift.. So I passed that code to gpt4 and asked it to do some stuff for me. it sort of did it but it gave me some errors. So I passed those errors back to gp4 and BOOM.. The damn thing works just as I wrote down in the chat.. Amazeballs. BUT the caveat is that it often quits halfway through. So when im in a stressful situation and need to fix some code fast, it's really frustrating..

gpt4 even made the brush a lot better. without the "cracks". Still a bit jittery brush but at least not a lot of cracks in the lines.

What I would like to understand how ChatGPT can generate code. Because I am somehow pretty sure that it does not understand the code it writes so I am guessing it is copy-pasting most of the code from the internet. Or Am I wrong?

I'm all for refraining from using extranious, imprecise English in building prompts. Create simple SOV sentences list the functions. Think it out before you prompt. No need for please and excess junk words and filler words which could set the LLM astray.

What the hell's Nacho doing with python?

That is crazy

You literally told it in your first prompt to implement everything from scratch🤷‍♂️