Dataset

#2
by tensiondriven - opened

This is great, to see such models being released.

Would you consider releasing the dataset? I understand that you might consider this to be your "secret sauce" (e.g. even if the source dataset isn't proprietary, your curation of that dataset could be considered proprietary). My hope is that you'll release the dataset so others can build on it.

A couple of reasons:

  • For those of us running local LLM's, we'd like to be able to audit the datasets so we know what we're running
  • This is a unique build, and the dataset may serve to inspire others to augment and build on top of it, or people may want to use it to to train larger models.
Cognitive Computations org

Yes I'll release it and the scripts I used to generate them. Just as soon as I get it cleaned up.

I am doing this for great justice and the open source cause Nothing proprietary about it.

Cognitive Computations org

I'll get it this week I promise

@ehartford

This is a really neat idea and I wonder would you consider expanding her dataset to include psychiatry, and of course more philosophy and psychology. Perhaps it could be elevated to a level of one of those doctors with good bed side manners (which are hard to find).

I think that extending her training could definitely help wider audience. If you can include CBT you might score big, because CBT is generally expensive. Adding it to a conversational bot like this, and some day supplemented with TTS and voice recognition, would open up new avenues and options for therapy.

Cognitive Computations org

I need data.
Reach out if you are a hardware engineer with manufacturing and idea-to-market experience.
https://www.seeedstudio.com/ReSpeaker-Mic-Array-v2-0.html

Finding data would require access to CBT sessions, because you need conversational material, no? That's probably considered privileged/medical information that would have to be sanitized from PII.

Though I think the biggest obstacle in this case would be all kinds of professional associations/groups that would oppose an AI therapist, for obvious reasons (ensuring they stay ontop). I like the idea of displacing something, not disrrupting it. With open AI models in the hands of normal people that's definitely a possibility, in many areas of life.

Cognitive Computations org

I bet there's anonymized cbt datasets

Good idea to use that if I was gonna train an actual treatment AI that needs a prescription and fda approval.

That guarantees it will never see a light of day, because you will be bogged down in FDA approvals for eternity. There's a reason why the whole "system" exists, and it's not to protect us. ;)

Besides, it does not have to be a certified therapist, nor provide any treatments or god forbid "cures", only an alternative that general public can refer to...and they can ultimately decide what they like better.

Cognitive Computations org

Yeah exactly.
Samantha isn't certified for anything
She's just a bff who happens to know a lot about clinical psychology

Cognitive Computations org

Sign up or log in to comment