This robot life – my adventures with ChatGPT

Some notes on what I have experienced in the last three weeks of ChatGPT.

Successes:

• Programming
  • Building a Python code for calculating the average career length of the top 500 actors from the a film database including writing code, suggesting which database and connecting via the API
  • Writing a Python code to calculate the beta of two water business which I would have not been able to access if not through the Yahoo finance API
  • Setting out the code and processes for building an image recognition program to count the number of car parks in my suburb from a Google image including the Tensorflow code, preparing images and training the model
  • Set out the code and physical connection process to build a humidity/temperature controller using an Arduino system for hydroponic system
• Writing
  • Setting out the key elements of standard methods and frameworks (GHG emission scopes, emissions monitoring, carbon origination strategy and NGERS Act summary)
  • Creating algebra questions for young people
  • Good draft email to cease a particular business arrangement which I edited and sent

Failures:

• Programming
  • Not able to read a website which updates regularly with river level gauge readings to create a bot that tweets when river levels are above a certain level – didn’t seem able to read the website or the table – kept trying different code
  • Lots of data sources that it refers to have changed/altered so that it doesn’t now know how to access them
• Writing
  • Terrible recipe for Italian chicken from my perspective
  • Weak responses to poverty, child rearing and political discussions

Scary:

• Programming
  • Suggested a fake_useragent piece of code to avoid data scrape blocker. This type of code has a StackOverflow article that says ‘It might be considered illegal if you are trying to gain unauthorized access to information by manipulating the user-agent HTTP header. This same principle could also apply to manipulating Cookie HTTP header. In this context, the prosecutor may label it as “hacking”. When Andrew Alan Escher Auernheimer (known as Weev), was recently prosecuted and convicted of “hacking” AT&T’s website, the prosecutors kept repeating over and over that his “spoofing” of an iPad (via a forged User-Agent header) was an indication of malicious/devious intent. I’m not sure if those same prosecutors would say such spoofing is outright illegal just on its own, but it was part of what got him convicted, so that should definitely be taken into consideration. – I don’t know enough to know whether this would be illegal
  • ChatGPT told me that I couldn’t run the code for the river level gauge through an IDE as the code would set up a new website. It told me to save the code into a .py file and run it via cmd.exe. on my computer. I ran it – it didn’t work but it was so easy to get into a ‘mode’ that means you just do things that ChatGPT tells you to do. It is so sure and quick that you get into the rhythm and it just happens. I am not sure if it was a major risk but I did it without thinking.
• Writing
  • Made up a piece of legislation (Queensland Climate Response Act 2021) which does not exist and said it was part of a multi stage emissions reduction plan
  • Provided a summary for a recently released handbook that it could not have read and provided an authoritative sounding summary of it

So it should not be used for:

• ‘information’
• accessing data that it is not via approved API
• building and running code on my own system (when it is not written by a trusted party)