TherapyAI

Inspiration

We both have recently been part of the UTD Deep Dive Ai program and after getting expirence on the benefits of developing your own models and using these models through our own projects we realized the importance of AI and how it can ebnefit people. We beleive that all people should have access to this technology without being discriminated against

What it does

Our Model takes a Kaggle datset that was about Therapy whcih had little data about those who struggled with disabilites. We then used a gpt 3.5 turbo model to convert some of the queires to relate more to disabled people which was then written into a csv file. We then trained a gpt 2 model on this csv file and use that to make a better chatbot in order to simulate the expected output we would have gotten with Quality Diverse Algorithms

How we built it

We built it using lamgchains open ai llm to diversify our data and add it to our dataset. Afterwords we used transformers tokenizer to tokenize our data and the gpt2 model to train it on. We did this all in Google collab to take advantage of their GPU.

Challenges we ran into

One of the biggest challenges in our development was getting out dataset after wards using langchain to prompt for more diverse datasets was a problem due to needing specific prompt.

Accomplishments that we're proud of

This is our second time training a gpt 2 model and we are extremely proud of how it turned out. We are also extremeley proud of the impact that this idea could have as there has been a myriad of research conducted on this tppic and we beleive that with the assitence from mentors we could take this idea to the next level and attemtpt to develop Quality diverse Algorithms to make these diverse datasets.

What we learned

We learned about how big an issue bias was within AI due to the fact that AI has had a history of providing faulty information that is based of popular internet stereotypes and we also learned about how organizations like Open Ai take steps to counterract these issues by diversyfing datasets and creating condtions that make sure that the AI doesn't say something that is faulty or inaccurate

What's next for TherapyAI

We believe that the next steps would be conducting research on the effectiveness of Quality Diverse Algorithms over other algorithms that do a similar task and learn more about how to implement these into our own programs so that we can an efficient algorithms that can easily diversify datasets for others to use to eliminate bias in AI and improve accessibility for all.

Built With

• langchain
• pandas
• python
• transformers