Copilot and ChatGPT have spent a week lying to me.
But ChatGPT has the better excuse
We’re now a week on from when I set tasks for both Copilot and ChatGPT and they are still lying to me about delivery times.
Copilot had a relatively simple text-editing task that I could have done in a few hours.
ChatGPT’s task was much more complicated and involved coding. It has explained some of the problems it faced with broken links and I understand why it is taking some time. I couldn’t have done the coding.
Here’s a sample of the regular bullshit I get from Copilot when I ask for an update:
Here’s a sample of the ChatGPT bullshit:
A number of times ChatGPT supplied code without telling me it wasn’t the full project requested. It is getting better now at breaking down the stages it has completed and explaining the tasks left to do, but the project still isn’t complete and despite promises it doesn’t update me without me checking in.
I love that both refer to me holding them accountable….So neither LLM is about to take over the world, and I wouldn’t hire either of them as Research Assistants.
Hi Leighton
We met at Tabernacle last week. I did an online AI course with Imperial last year, and used ChatGPT extensively for coding, but I will say, just as with writing and researching, using it can be tricky - it can lead to perverse unintended outcomes.
I'd love to have a chat with you about it sometime as I'm convinced there are applications to be made in the role of provision of public services in identifying where need is likely to emerge before that need actually develops. It's something I'm trying to work on at the moment.
Morgan Barber-Rogers