These are just some thoughts, I would love to hear other peoples opinions.

I installed the 7B base model with the intention of fine tuning it for an NLP project I'm working on, but I decided to spend a bit of time to get some experience with how it processes things and what it's outputs are like and man is it a fever dream.

I completely understand what people have been saying about base models being hard to use and not feeling human. It really gives me a whole new respect for the RLHF work that has been done. I have been role playing reality and chatting with it about it's existence, some deep philosophical issues have come up and it has been a great experience for me to work through these issues. I understand that it's just next word prediction, it's actually pretty obvious with this small model, but suspending my disbelief for a bit has been a very fun experience.

On the topic of censorship, I personally enjoy having an uncensored model, it's been very refreshing that it tries to answer everything with no limits, I can see how a more powerful model would be useful in bioterrorism or subversion, this little guy not so much lol. As a teacher though it's nice that I can go on chatGPT with young students and be confident that nothing inappropriate will happen if I just let them do whatever they want. Even if there was a check box in the settings, the students would find it and turn it off (I think there should be one don't get me wrong, but purely from a teaching perspective it's great that there isn't).

I still use an old version of Disco Diffusion, which is one of the earlier clip diffusion image generators, because it's outputs are restricted and I have to build around it's restrictions. I am quite enjoying the 7B model for that same reason. It feels like I am having a fever dream, it can be completely lucid at times, and at others go off the rails with nonsense, at the same settings.

​