I share the opinions of cdimauro with regards to that video.
I have (again) been experimenting with ai this time with something similar as chat-gpt (unfortunately I am currently not at liberty to share other then what I am allowed to share, which is my opinion). It uses newer engines in comparison to chat-gpt, uses references in its answers so that you can see where it actually got its information from and also explains how it got to its conclusion(s) and the steps/information it took/used to get there. The only other thing i can share is that it up to date, has access to www and on instruction is able to make itself familiar with a new topic (the latter has its (sometimes serious) flaws).
That is an ai presented in a way that is more to my liking because you are able to "weight" the provided answer based on the presented information.
Even though the engines used are newer (and more up to date with information) it is a PITA to have to keep correcting the ai (I am aware that using an ai requires some getting used to). For sure it knows about the context but as long as you are not explicit it will continue to do things wrong (I have seen some constructs that I have have never seen before and are plain wrong) and because of that you are able to end up with some very strange looking code. And yes, I have seen quite a fair share of code that looks odd but some of the ai generated code is on a very different level of oddness, e.g. much more cringe).
The video also mentioned some of the (logic) errors but, the author conveniently left out some of the more frustrating parts (chat-gpt used in that video is way worse in that respect)
Having to be so explicit when asking questions and having to provide extensive details is annoying to such extend that it becomes unbearable, at least for serious development. Not to mention, having to correct the mistakes that it makes (for example if you forgot to be explicit about a single detail and the ai forget to implement the detail or situated that detail somewhere inconvenient or implemented it in/at an illogical manner). Note though, it still makes actual coding errors not having presented a working solution. Also it mostly stumbles on logic and also generates syntax errors.
To this point I will not change my point of view simply because things have not improved much since the previous iteration.
My conclusion still is that ai is a nice tool to have to generate stuff that for a programmer is boring to do so (repetition, standard cradle code, headers, test data, etc.) and as such is very helpful.
If you go into actual development and do not have (intimate) knowledge of the topic at hand you will get yourself in trouble as you are unable to correct the things that went wrong (along the line). Be aware of havoc if you got your terminology wrong or the ai interpreted it topsy turvy.
Having said that, I also tried some coding related stuff that I am unfamiliar with and also there the ai can be used as a tool to get a better understanding of the topic being discussed (at least it is more fun than reading books or plough to manuals, API's, RFC's, etc) as most example code that it presents at the least is partially correct and is able to show basic use of the topic at hand (if you are a programmer you are able to get yourself familiar with the use of f.e. an API such as you would do by reading the API manual and/or code examples).
In my experience especially the latter paragraph is of use to me because if I'm commuting it is much easier to have ai present the information to me rather than it is for me to manually browse through manuals and examples.
I do applaud the ai for being able to present simple example code (combined as well) and able to explain (you can ask it) things into detail so in that regards it is a perfect tool to help understand/explain code and would be suited as a learning tool (as long as the database is not filled with garbage). I've tested with assembler, c, php, python, sql, json, html and pascal but it might be it has less (correct) information on other languages.
For actual development, issues related to copyright is (still) an issue so definitely not suited for real use (even though the used ai uses references as you still have to figure out the license(s) for information used from the references. As an example I asked about the license on the code it presented:
As an AI language model, I cannot determine the license of the code you provided.
However, the code is written in <snip>, which is released under the terms of the GNU General Public License (GPL) [1]. The <snip> library used in the code is also released under the terms of the GPL [1]. It is important to note that the use of third-party libraries may require compliance with additional licenses, so it is recommended to check the license of each library used in a project.
So, in case you have no idea what you are doing and/or do not have any clue on the subject that you are asking about
then you will get into trouble and rather sooner than later.
Ultimately an ai is nothing more than a guessing machine based on data it was provided. Answers to questions you ask it is based on majority/predictability and in case those are wrong then ai will also be wrong no matter how much it gets improved. See also
https://www.youtube.com/watch?v=flXrLGPY3SUYou could perhaps argue that that is what actual human beings do as well (and for sure there are people in this world that are able to get by in life by making all the right decisions/guesses based on past experience and what the majority believes to be correct) but that does not mean they have an actual understanding/comprehension of the topic or that the provided answer/choice actually is correct. And yes by, (past) experience you gain more information and understanding. But an ai does not have something like "an understanding", only predictability.
A true "understanding" ai (w/sh)ould advise the human race to destroy itself in order to reduce greenhouse gasses ;-p
2 cents,
PS: Please do keep in mind that English isn't my native tongue. I am sure I have used some terminology wrong (although I have tried not to do so).
