Can AI fix your mental health?

Of course not, don’t be absurd. Chatting with an algorithm that is designed to parrot back a simulation of human understanding, compassion and insight is obviously not the answer to the current mental health crisis.

Nevertheless, when an old aquintance approached me with a software development project, an AI-based mental health companion (think “Her“) is exactly what I agreed to try to build. I could use the money, what can I say? Besides, learning a bit about large language models and playing around with some shiny new tech didn’t sound too bad. Back-end, web app, iPhone app, Android app, all in 5 months. The development, architechture, UI, design and platform team was me. Just my kind of project.

I had a great couple of months digging into Llama2, Mistral, Hugging Face, llama.cpp, etc. but obviously it’s not a one-man job. And as this was an attempt to save a an already failed project, you could add: minimal funding, no real team, no real business case, no requirements (except building an app that made money), no ownership and no contract. Life as a contractor in the start-up lane, I guess. Building a chat on top of ChatGPT and charging for that is pretty pointless, a blatant violation of any privacy and will not make you money (unless of course you somehow make it go viral on TikTok). Running your own LLM’s on Nvidia H100’s will quickly bankrupt you. I love tech, I really do, but in the end, the biggest issue for me was of course the simple question, why? Why build something I did not believe would help anyone?

The project was funded and owned by a Danish startup accelerator/VC. I have succesfully dodged contact with the local startup/VC scene for most of my career. The core problem with this scene (in my experience) is, that unlike the US startup/VC environment which is heavily influenced by product and tech people, the European scene seems to be almost completely populated by “business” people, and is saturated with a fake-it-till-you-make-it attitude.

In the end I successfully delivered what I had set out to achieve – a working beta, and we amicably parted ways. While the beta may not solve any mental health issues, the code could still be valuable to many and, as with my other projects, will be released as open source. So long VC-world, it was interesting.