Small refinement: the underlying model isn’t stochastic at all. The forward pass is a deterministic function of the weights and input, it just produces a probability distribution over the next token. The stochasticity is an optional sampling step layered on top, not something inherent to LLMs. Greedy/argmax decoding (or temperature 0) makes the whole thing deterministic.
So “purely stochastic” overstates it a bit: the distribution is computed deterministically, and you choose whether to sample from it or not.
Maybe dumb question but how do I just take a sat image and create the scene? The scripts in the repo are all about training which I assume requires you to have the 3d data too.
These sort of projects always look cool but I think the real "wow factor" would be a file upload where you can see the result on your image. I assume there are reasons why this isn't done.
Is there anyone monitoring US government websites for deletion of pages beyond what people may notice in an ad hoc way? I feel like this should be automated.
Thanks for this! It means a lot. It can be a little discouraging to hear criticisms when you've put a lot of thought and time into something that you hope can help people.
But I fully understand and appreciate the comments made by other posters.
I have put it back online with a disclaimer that must be acknowledged, and pointed out this is a research app not a substitute for therapy.
I hope that does enough to mitigate some of the risks, whilst allowing me to continue to work at making this a safer and more valuable resource. Whenever that may be, or not :)
Thanks so much for your thoughts. I tried my app with a friend and it took a good amount of digging deeper to actually get to their root thoughts about their difficult event.
So to be successful with such a tool, a good level of self-awareness is required.
Yes a tool as an adjunct to regular therapy is a good opportunity. I discussed this actually with my own therapist and he thought it was a good idea, with the NHS being so overwhelmed here in the UK to be able to have a first-line resource-cheap option is appealing.
I suppose the thing to figure out is how this is best done in a safe and useful manner.
It can be hard to find a therapist who you gel with. I've been through a few in my time. The real work is done by oneself ultimately. Be it with CBT homework or mindfulness between sessions. Obviously it's very individualised though.
Robert Burns (of CBT fame) is currently involved in an app which incorporates AI so I'm interested to see how they've dealt with the sensitive issues.
I've taken it down. It was meant as a quick demo of how one might leverage GPT4 in this area but there's a risk posting here people might think it's a serious product.
From your comment however, you don't rule out applying AI in this field entirely?
I have taken it down, thanks for the advice. Definitely not worth taking risks. As someone who has struggled with mental health this is something that I built mainly for myself. I will keep it as a personal assistant.
So “purely stochastic” overstates it a bit: the distribution is computed deterministically, and you choose whether to sample from it or not.