Hacker News Reader

Comment by dontupvoteme | Hacker News Reader

PLEASE do not do this. The field is far too new and the risk matrix of hallucinations are through the roof, ESPECIALLY with those who have mental illness.

This is the sort of stuff that might get legislation passed on open source models or turn public opinion against them.

Yes the USA's most powerful union is the AMA and yes they're the reason # of doctors are low and doctor salaries are skyhigh (especially compared to doctors in the rest of the world), but this is not the way to address the problem, at least not yet!

View on HN

Not only that, but the model was trained on 172 training examples - and the conversations aren't remotely like the therapist conversations I've had in terms of nuance or topic.

This is a toy project. Hence, large-scale collection of data and pre-processing them into conversational format, also removing sensitive information is not possible. I was able to curate only 172 rows of data.

I have shared a notebook and explained detailed steps in my blog - https://medium.com/@iamarunbrahma/fine-tuning-of-falcon-7b-l.... If you are interested to replicate the steps on medical-domain therapy chat transcripts, you are definitely welcome. If you face any issues during fine-tuning steps, you can connect with me on my blog. Would love to help out!

It is interesting. Sorry to pour so much cold water on the topic, it's more the application than the code that people have issue with.

I wonder if retrieval augmented generation would be appropriate here too. Fast, scalable. Fine-tuning is comparatively pretty expensive.

Thanks for sharing your code, iamarunbrahma.

The purpose of this project was to showcase how can we fine-tune large models on a free-tier GPU provided by Colab. Hence, any individual can utilize parameter-efficient fine-tuning methods to tune LLMs on domain-specific datasets.

RAG and fine-tuning serves a slightly different purpose. Fine-tuning helps LLM in learning a new task/skill such as question/answering task, summarization task etc, and improving reliability at producing a desired output such as JSON format structure thereby reducing dependency on prompt engineering.

On the other hand, RAG provides you with external domain-specific knowledge, which one can leverage to get latest information.

I see, thank you for the explanation. RAG is also appropriate for question/answering but I do see your point. Can see myself returning to your code for inspiration soon.

> I was able to curate only 172 rows of data

Skill issue - focus more on data integrity & less on HN clout

OP is sharing a personal project, he's not claiming he's solved the mental health crisis.

This is Hacker News. These types of projects are half of the reason I come here, what's with the hate?

Oh!! So, you want every individual who post their github repo on HN; to collect dataset like billions of dollars funded startups and make a reliable chatbot which not a single well funded startup or listed NASDAQ companies could have done. I salute you.

It’s just Q&A. To call this a chat is very misleading.

Here is another perspective.

The web is already FULL of bad medical information. Older folks treat it as gospel already. All this technology does is make both good and bad information more accessible.

More importantly, there is NOTHING anyone can do to stop this. Nothing the US government can do will stop the proliferation of AI models that do everything including giving medical advice (or other things that we know it will screw up)

"Dr. P<hil>" is still on TV.

I take a pragmatic view on these things - assuming appeals to morality and ethics will fail, what do we do to educate people about this?

Pretty cynical take. The world is already shit, what's another turd on the pile?

Perhaps the right take is to take every experiment to its extreme even if it's confusing.

People have mostly learned photos can be edited. What we need are outlandish experiments and examples, and then media showing off how wild this technology is.

Attempting to bubblewrap this tech both won't work and will lead to more confusion.

This is why we need pocket universes in which we can spin up and test new technology, like in the peripheral. Too risky to try chaos engineering in a legally risky domain like this

Thank goodness this is the topic of the top comment. Just reading the headline gave me a cold sweat in all the ways this could go wrong.

I get people wanting to be FIRST in a space, but they seem to continuously be the people least needing to be first. Will this ever be something viable? maybe, but it definitely is not now. Yes, it can't get there without going through the growing pains, but let's work those pains out in less detrimental areas first. Areas with much less fallout when it does go wrong

This comment just reeks of gatekeeping and ludditism.

How do you expect something to go from “toy” to “viable” if people are not stepping up to build upon the body of knowledge? A lot of people who have created innovative solutions were likely, by your definition, to be among the “least needed” people in a space.

Maybe you would benefit by trying to understand why a harmless toy and proof-of-concept is so threatening.

maybe not testing with actual human's fragile mental well being with something being developed as toy, eh?

the words toy and mental health should never be mixed together. this is exactly why it is threatening. someone cavalier enough to think that a group of computer programmers throwing together a chatBot to interact with mental health is just, well, mental.

Literally no one is advocating for this as is to be used as a substitute for currently proven treatment. It shouldn’t mean that we can’t try out different solutions, nor will the final hypothetical product derived from this likely even take the same form or possess the same chat interface.

We cannot let fear stall innovation because of someone might use it incorrectly.

So says every person that has ever had a thing they made used in a way never intended. Deny users will do something you've never thought about at your own peril. Unfortunately, in cases like this your peril can have very negative consequences.

This isn't a case of "manning up" and doing something scary. We're not astronauts going to the moon or any other situation where the decisions made will only affect you. This is a matter that when the product is misused or has a "glitch", it is not the maker that suffers.

The fact that you will not even acknowledge this just proves maybe you are not the right person for this.

It's the 7B model, which is understood to be a toy model for fun experimentation across the industry and even hobbyists.

I'm sure OP is not going to go out and treat patients with this. 7B barely produces sensical sentences most of the time.

people are already doing this on Character.ai and other popular AI consumer products, whether the companies want them to or not. I think it's better to accept people want something here and make it good & transparent vs trying to avoid it... it's like alcohol, sometimes people want a thing badly enough they don't care if its bad for them

Reply on news.ycombinator.com