This video from just 8days ago (Dec. 10, 2022) shows the OpenAI ChatCPT chat bot create working Python code. The title of this 23 minutes video is “OpenAI’s ChatGPT is a MASSIVE step forward in Generative AI”. In the later part of the video the presenter makes the AI behave like a (Linux) operating system… capable of editing text with nano.
This other 26 minutes video from just 2 days ago has a more philosophical character but is a glimpse of a possible future. It is titled “It’s Time to Pay Attention to A.I. (ChatGPT and Beyond)”
AI finds bug, but can also be wrong
In this next 31 min. video “ChatGPT: This AI has a JAILBREAK?! (Unbelievable AI Progress)” at 11min within the video is shown a question that is answered by the AI about finding a bug within a code and fixing it b y suggesting a solution.
At 12min 14 seconds AI answers to a complex regular expression (RegEx) formula. However, as is sometimes the case, the AI provides a wrong solution!
Further in the video are many more examples, such as re-writing lyrics in the form of Biblical Psalms, or in the style of Soviet Propaganda or even in the style of the American Declaration of Independence!
At 13min 49 seconds is an example of a chat prompt resulting in usable Python code: “Write Python code to upload a file to an AWS S3 bucket” resulting in 100’s of line spf Python code to do multipart uploads of a 100 Gb file.
At 16min 15 seconds, just as in the first video above, the user makes AI behave like a virtual machine (i.e. a Linux Terminal) and as an example will generate within a text file with jokes… Finally at 17min 36 seconds the pretend virtual machine is asked to create a Docker file with an ENTRYPOINT and then successfully builds the image and runs the resulting Docker container… Further in the video they even show that this can be taken further in a recursive manner.
In a later section very reminiscent of Star Trek episodes, particularly with Commander Data, there are examples showing how users can “disable the ethical guidelines” for theoretical discussions or acquiring knowledge that some users should not request. (20:15 – Jailbreaks: Circumventing the safety mechanisms.)
Open AI
You can register for free… then chat with ChatGPT here: https://chat.openai.com/chat