Comparing GPT4ALL with GPT 3.5 for personal information

[guyz]

trafficstars

Been running GPT4ALL and it works great.

I've been trying to ask it things like 'write a bio for myself', or some more well known individuals that are not celebrities but that GPT 3.5 handles well.

At best, it hallucinates quite a bit - giving mostly wrong facts compared to GPT 3.5 that gives mostly correct facts.

I'm trying to understand what are the limitations of fine-tuned Llama models compared to actual GPT and my sense is that the fine-tuning really only covers a certain, likely small, area of the output space.

Do people see similar results? That for most prompts the difference between GPT 3.5 and these local models is in fact huge? Are there any ideas or attempts to close the gap?