I wrote this note in an effort to make the AI usage policies in this app transparent to all users of the app.
This app uses AI, specifically LLMs (Large Language Models) to create personalized tasks that are specific to your goals, your life, and your challenges. I (the developer) do not support training LLMs on user data, especially if they didn't knowingly opt in.
Please know that I have chosen to disable training on inputs (aka I tell the LLM providers “do not train your model on anything sent by this app”) and only use models with Zero Data Retention (aka “do not save anything sent by this app”). You can be sure I have no plans to change these policies now or in the future.
Obviously it’s still not ideal that we are sending data to large LLM providers at all. When I originally had this idea, I wanted to use a local model hosted on the user’s phone, but the quality of results generated would be low and it would put a lot of strain on people’s phones in terms of memory and calculation (if the phone were even able to handle it). I ultimately decided to use an external provider with the above settings as a tradeoff. Also, I plan to add options for people to choose their own models in the future.
All your data (goals, plans, tasks, notes) is stored in Supabase. This is encrypted — see their security info, quoted below:
All customer data is encrypted at rest with AES-256 and in transit via TLS.
This is hard to address concretely because there isn’t much reliable third-party data publicly available about the true energy and water usage costs of AI. I think more transparency from large companies about this would be very helpful. I personally believe the environmental impact of AI is an important issue, and to be clear: I am very skeptical of many uses of AI especially given the negative impacts it can have on society and the environment, so I have approached this project with caution.
Based on the info available online, the environmental cost of generating text is pretty small, far less than what’s needed for images and especially videos (which you can think of as a lot of images). Additionally, the design of the app limits the number of “generation” calls that are needed to create effective plans. I intentionally steered the design away from being a chatbot. A typical user will not be making many queries per month - much less than you would need with a chatbot.
Furthermore, the app is not designed to encourage you to keep generating more and more content. With these considerations in mind, I believe that using Achievable would constitute a very minimal portion of your carbon footprint.
The choices I’ve described above are what make me feel okay about using this app myself. I have found that the ability of LLMs to break down my own big goals and overwhelming problems one of its best, most empowering uses, and privacy-wise I can trust my own data policies, as a small business which will never face investor pressure to do anything nefarious with people’s data.
Let me know at any time if you would like more info about this!