[Quiz] Who knows how OpenAI GPT's work?

Here’s a GPT quiz for you:

1. Who can train / fine tune a published GPT?
a) Only the GPT developer while using the GPT
b) Only the GPT developer while editing the GPT instructions and.or files.
c) Anyone using the GPT.
d) Automated fine-tuning by OpenAI.
e) Something else.

2. What happens when a GPT builder uploads new knowledge files?
a) The GPT knowledge is expanded without losing previous knowledge and fine- tuning
b) The GPT knowledge is reset and starts from scratch based on the new instructions. Any fine tuning is lost.
c) Something else.

3. What happens when a GPT builder edits the GPT instructions field
a) The GPT knowledge is expanded without losing previous knowledge.
b) The GPT knowledge is reset and starts from scratch. All fine tuning is lost.
c) The new and old instructions both remain in the GPT’s logic, including any fine tuning on the old or new instructions.
d) Something else.

4. When I use someone else’s GPT, it displays a thumbs down icon under the response for negative / constructive feedback. What happens if I click the thumbs down icon? Does it change that GPT developers GPT in any way? If not, what happens as a result?

If you don’t know the answers please respond confirming that, so we can all get confirmation of how this works.

1 Like

These questions are nonsensical.

  1. There is no “fine-tuning” a GPT, there is no “training” a GPT.
  2. A published GPT is a static snapshot of the GPT when it was published, as there is no training or fine-tuning, nothing can be lost.
  3. The thumbs down does not affect the GPT in any way—GPTs are static. The feedback goes to OpenAI, just like it does with all other ChatGPT feedback.
1 Like

While I totally agree with @elmstedt in defense of OP, I perceive the ‘linguistics’ around consumer facing AI products as completely confusing/non-sensical. I always need to translate the technical sounding everyday language into something that actually makes sense.

But this really doesn’t help the quiz, which suffers from exactly this problem.

I read online that OpenAI GPTs were going to allow fine-tuning, hence the questions. In that context the questions are totally, umm, sensical. You could have responded with option e (“something else”) and provided polite feedback. I’m trying to locate the information I saw previously.