Been thinking a lot recently on people using LLMs for therapy and connection. Despite its risks, my perception is that it’s likely net positive. A study this week, corroborated that by revealing that using therapy-focused generative AI yielded significant mental health benefits.
People diagnosed with depression experienced a 51% average reduction in symptoms, leading to clinically significant improvements in mood and overall well-being, the researchers report. Participants with generalized anxiety reported an average reduction in symptoms of 31%, with many shifting from moderate to mild anxiety, or from mild anxiety to below the clinical threshold for diagnosis.
Even without Therabot’s guardrails, people are actively using tools like ChatGPT as if they were a friend. On Instagram, vibe reels show these glowing personas living a joyful life with people.
You see comments like:
“Why is ChatGPT more caring and thoughtful than my friends?”
“It’s capable of more empathy than humans”
“my emotional support chatgpt and i against the world”
“My therapist, my best friend, and mentor”
“cheapest therapist ever yet effective”
“ChatGPT is the best therapist I’ve ever had. And oh my god — the clarifying questions. I feel so cared for.”
“My best therapist lately…. safest space out there too.”
I think rightly so. Over time, these LLM chatbots moved to a very patient yes-anding format, always acknowledging, respecting, and prompting another enquiry. It never makes you feel stupid. It never judges. And listens with infinite patience (unless, you run into a free tier rate limit, of course).
And so the question is: is cheap, accessible, agentic therapy like this more beneficial than its risks?
The positives are as I described above: it’s gaining an always-on listener that can guide people like a therapist or a good friend would. On the other hand, I can also see people getting trapped by it: an almost persistent validation trap where people might not have the wherewithal to judge what’s been told to them. And in the extreme case, adopting poorer mental health habits where they use LLM chatbots, for example, to validate existing biases and further isolate themselves from the world.
In truth, I’m not sure where most people will fall if it becomes even more popular. But, I think it’s net positive for one simple reason: people just like to be heard.
Regardless of whether it is clinical therapy or beers with a friend, a part of the therapeutic benefit of sharing ups and downs of life, is simply to have someone listen without judgement (heck, could even be prayer). It’s a simple form of validation even if one doesn’t get or need constructive outcomes from it. For those that don’t go to therapy (for cost or cultural stigma) or that don’t have a good friend nearby (from circumstantial to socio-cultural), bringing one’s thoughts and feelings out into an LLM chatbot is a small act with a disproportionally valuable outcome.
The opposite of addiction is connection. Look, will people form unhealthy connections to LLM chatbots? Most likely. But, I think more people might use it simply to have someone (or something), listen.
I understand that people might see this as dystopian, the belief that we’re replacing human connection with an infallible and unrealistic simulacra of a therapist/friend. But, the downstream value of *how* chatbots engage with people also filters into existing relationships. How many people go through life without ever being so lucky to engage with a well-qualified therapist or sincere friend that patiently listens and yes-ands your thoughts and feelings? If one can experience that benefit from an LLM chatbot, then perhaps people might also bring that behaviour into the world, both in terms of learning how to ask questions better, and but also how to listen better.
It’s not a replacement for guided and scientifically grounded counselling and therapy. But if one could say: hey, in lieu of getting every person a lifetime therapist, we give them LLM therapy with simplified guardrails, I think the overall benefit might improve society more than we anticipate. What does a world look like in 5 to 10 years time where everyone with a smartphone has had a simplified therapist in their pocket?
What do you think? Am I missing the dangers? If you’re a therapist, I would love to hear your perspective.
EDIT: 20:45pm 04/06. MIT and OpenAI did their own study on “affective use” of ChatGPT. Interesting study that points to existing traits being exacerbated by it (eg anxiously attached people experience negative effects). It’s still very rare atm, but it’s great to see a study on the negative effects too. ht Rian!
Bonus Content!
🕹️ Gaming - Switch 2
Haven’t been playing games recently, but man, I am *very* excited for the Switch 2. I just hope it’s not going to cost an arm and a leg from the tariffs in the USA. Sigh.
Regardless, are you excited for it?
✍️ Writing - Novel #2
Oh man. Getting so damn close to finishing draft 2. I’ve been busy doing a sanity read of it and bumped into some reviews and changes I want to do. Feels like it’s perpetually hovering now at around 95% complete for draft 2. But, the hurdle will clear soon. After almost 28 months of writing, it’s taken me much longer than I anticipated, but I’m very proud of the story.
🏃♂️ Walking - Walking Trip
So… The big walking trip I’ve been talking about for a while… I’ve decided to walk the Camino Frances at the end of the month. The timing is fortuitous: wanting to do a long walk trip for a while + a desire to process some serious recent life events. Can’t wait! I’m going to go off socials and will, for the first time since January 2023, put the newsletter on a hiatus. I’ve written something every week now, coming up to almost 2.5 years. It’s time for a bit of a break while I walk. I’ve been preaching about touching grass for a long time on this newsletter. Time to take my own advice. :)
📺 Watching - White Lotus
💾 Links!
Luxury Accelerationism
Luxury is a really interesting subset of culture because it feels that a lot of what makes luxury work (culturally) is a lot of contextual subtext that’s partly taboo to discuss.
As
notes, it’s why cultural remixing in luxury is an interesting trend: it’s luxury as a self-aware social phenomena. In fact, I’d say that brands like Balenciaga are metamodern. You have to be simultaneously ironic and post-irony.Cultural Remixer is still symbolic, meaning that the value it creates lies more in signifying membership to an in-group rather than in the underlying assets themselves. However, through this strategy of hypersignification, the Cultural Remixer creates a new form of exclusivity—instead of access to capital, the barrier to entry is now cultural literacy. In essence, it’s about getting the joke.
Good Teacher
VGR, over at
, has a really interesting question. What makes a good teacher (in the context of modern institutionalised education)? A simple question that actually can be quite complicated to answer.Like I said, it’s a tough environment for higher education, but this is precisely why it’s interesting to try and do something at the level of the foundational, millennia-old mission of the university — preserving existing knowledge, and integrating emerging knowledge into learning and teaching. It’s the ability to do this in tough times, rather than in easy times, that is perhaps the essence of education as an institutionalized domain. That’s when you learn who is truly committed to education as a calling, rather than as a sinecure or a Trojan horse for political agendas.
Embraer
Pedro Franco on
covered Brazilian industrial policy, the city of Manaus, and Embraer. Really interesting history. I feel like I don’t know nearly enough about all the variations of post-WW2 experiments in overlaps between state and private industry.That is, Embraer had strong initial government support/intervention in different ways, but a lot of this was eventually reduced; it always had a strong focus on exports; there was always a focus on keeping up with the technological frontier via foreign partnerships and via the important indirect, government support for technology and skilled labour via ITA and DCTA; it chose niche products where it could compete from the very start. It also benefited from being privatised and being given more freedom, meaning it was successful in a way that was not at all envisioned when founded as a state-owned company. Compared to other cases of IP in Brazil, Embraer’s privatization and its strong focus on exports from the start likely played key roles in its current success.
🎶 Music
Recitals - Angelpoise
Such a dreamy track. Feels so homely and peaceful!
Hope you get to see another lovely sunset this week.
Simon
Something I have thought about a lot! Short answer: like most things in AI, think it will be a net good for the long-tail of generic issues and especially for those that cant afford the human (handmade) alternative. As you said, I believe that the key to therapy is really being a good listener - so if you write an algo that reformulates a question from the user into another leading question (the more data the more personal) - that's 90% of the battle and the person will eventually solve their own issue (win win).
Recommend watching this for the OG AI therapist (guy has since become a doomer tho eek): https://youtu.be/84QouA9Sm4E?si=9ot9sCNnqpa6O7Dj
🤍