I previously in Jan 2023 made a small post here about OpenAI and ChatGPT. It was at the time v3 of ChatGPT, the first one made publicly available by OpenAI.
Since Facebook regularly bring “memories” back, essentially older posts, made by me on my profile, it showed me the above post, and I thought… hmm.. how would that go today in ChatGPT 5.
So I asked:
create delphi code to draw a pepsi max bottle on screen
And it responded with code (after 3 code fix requests) that generated this form:
Not perfect, but much better than before. The level of details have increased significantly, but the overall understanding of the form of a Pepsi MAX bottle seems still not to be complete today, nearly 3 years later.
This is a good indicator to show that, yes AI has gotten more refined and have a larger detail level (probably a higher level of internal dimensions) in its understanding of things, but it is still not “intelligent” in the way that it, despite its vast learning material (much exceeding what a human digests in a lifetime), is still not able to figure out the minimum needed for representing a Pepsi MAX bottle in a way that would make that representation immediately recognizable by humans.
Instead it focus on the artistic details.. the gradients, the droplets, and other such stuff, probably learned from watching too many commercials, while still not actually producing the correct shape of the bottle.
That is something to be aware about when using AI.
To use AI in a good way for coding, you absolutely need to understand the context in which the code is to be used, to be able to discern what AI generated code is useful or what to ask for to make it useful.
Just accepting the AI generated code without understanding how it works and in what context it should work, will result in you producing a project that WILL go wrong.
For that reason, AI is currently still not a replacement for good programmers. It is however a tool in the toolbox that good programmers can use, to speed up the coding process, by generating code that would otherwise be tedious to write or that represent scientifically well known algorithms.
The good programmer will know how to apply the produced code and how to test and verify its usability and reliability.
Ive not used Claude… I’ve good experience using Gemini for code generation and ChatGPT for reformatting and special cases where Gemini refuse to understand its in the wrong 😉
ChatGPT actually produce pretty good Delphi code and understands most of Delphi, but all AI’s I have tried do sometimes get confused and include syntaxes and features from C#.
When one points it out, they usually correct themselves.
Chatgpt doesn’t fully support Delphi code, I believe that Claude is the only AI that claims to know Delphi, have you tried it there?
Ive not used Claude… I’ve good experience using Gemini for code generation and ChatGPT for reformatting and special cases where Gemini refuse to understand its in the wrong 😉
ChatGPT actually produce pretty good Delphi code and understands most of Delphi, but all AI’s I have tried do sometimes get confused and include syntaxes and features from C#.
When one points it out, they usually correct themselves.