A profound relational revolution is underway, not orchestrated by tech developers but driven by users themselves. Many of the 400 million weekly users of ChatGPT are seeking more than just assistance with emails or information on food safety; they are looking for emotional support.
“Therapy and companionship” have emerged as two of the most frequent applications for generative AI globally, according to the Harvard Business Review. This trend marks a significant, unplanned pivot in how people interact with technology.
I also play with llms for a living. I use ChatGPT for therapy and to process emotions. I also see a therapist. ChatGPT is there on my time table and at the time I’m trying to process or learn or just have some fun to see where the limits of the model are. I don’t have to wait for a random time slot in 4 days where the thoughts get clouded by time.
I know ChatGPT isn’t real and it can be dangerous as it always looks to normalize and support your point of view. But sometimes people need an outlet that’s not my waifu pillow.
Therapy in real life takes time, effort, you have to build rappprt. You have to find a therapist that meshes well with you; it’s really like dating and finding a matching partner. Many people will take months/years before they’re willing to open up fully to a therapist… Where in 2 minutes they’ll tell ChatGPT their darkest thoughts and closest held secrets. It’s different.
ChatGPT (last time I tried it) is extremely sycophantic though. Its high default sampling also leads to totally unexpected/random turns.
Google Gemini is now too.
And they log and use your dark thoughts.
I find that less sycophantic LLMs are way more helpful. Hence I bounce between Nemotron 49B and a few 24B-32B finetunes (or task vectors for Gemma) and find them way more helpful.
…I guess what I’m saying is people should turn towards more specialized and “openly thinking” free tools, not something generic, corporate, and purposely overpleasing like ChatGPT or most default instruct tunes.