My partner is ok with it *
2,253 karma · joined July 8, 2015
I'm shocked that GPT-5 or Gemini can code so well, yet if I paste a 30 line (heated) chat conversation between my wife and I it messes up about what 5% of those lines actually mean -- spectacularly so.
It's interesting to ask it it analyze the conversation in various psychotherapeutic frameworks, because I'm not well versed in those and its conclusions are interesting starting points, but it only gets it right about 30% of the time.
All LLMs that I tested are TERRIBLE for actual therapy, because I can make it change its mind in 1-2 lines by adding some extra "facts". I can make it say anything.
LLMs completely lose the plot. They might be good for someone who needs self-validation and a feeling someone is listening, but for actual skill building, they're complete shit as therapists.
I mean, most therapists are complete shit as therapists but that's besides the point.
What's the birth rate in the civilized world?
How many men under 30 are virgins or sexless in the last year?
I'm sorry, but I find this intelectual dishonesty and moving the goal posts.
Speaks more about our inability to recognize the monumental revolution about to happen in the next decade or so.
My definition of a complete AGI is: an AI that can read JIRA tickets, talk with non-programmers and do all my job and get me and all/most software engineers fired and proves sustainable.
But in general, it's an AI that can do any remote-work just as good as humans.
Indeed there's all kinds of Montessori.
I can vouch for my daughter's .
If anybody wants to give it a go, my benchmarks are:
1) find reviews of parents, especially no abuse, shouting, kids in the last year should LOVE the place.
2) observe even for few minutes a class in their focus time -- you will feel almost shocked if you haven't seen this before -- like you entered Santa's workshop -- children should be deeply engaged in their activities. If you haven't seen it before you might suspect abuse (that's why point 1 is so important), no way kids love to wipe the floors, lay tables, prepare food and so on, but they actually do.
And all that done in almost complete silence.
Proper Montessori with good, empathetic, dedicated educators is amazing!
The famous Chinese Room Translator -- silica is irelevant, you could probably implement LLM-like algorithm with pen and paper, do you still think the paper could suffer or be "conscious"?
And you'd be correct.
It doesn't matter if they kiss to talk or pass on their memories as DNA or they exist in a permanent stat of ego death.
Because we received a modulated signal across the Galaxy, it tells me they:
1) are willing to talk using sequences 2) are technologically inclined, hence they know maths and physics
But we find PLENTY of common grownds when we talk to the smartesr of those groups and races, across milenia and continents via groups, scientific forums, discussion books.
We find very little common grounds when we have forced encounters with the uneducated trouble makers up to no good, in systems designed for high trust abused by said individuals.
A finite intelligence, willing to talk across the galaxy, talking in finite sequences, using engineering and maths?
I'm sure there's a lot of universal aprioris
True, but even the boy who cried wolf too many times eventually got his sheep eaten by the wolf.
I have my own personal anecdotal benchmarks and I never hyped LLMs before GPT-5.
Things that simply did not work before GPT-5 no matter how many shots I gave them, GPT-5 breezed through.
For me, it would take at least 2 generations of no felt progress in the models to call for diminishing returns, and I'm not seeing them.
GPT-5 is not the final deal, but it's incredibly good as is at coding.
Anecdotal, but it's something completely else in terms of capabilities, ignore it at your own peril, but I think it will profoundly change software development.
There are other concepts like space fountains, orbital rings and sky hooks that seem more doable -- especially the sky hook seems close to do-able, especially on the Moon.
It depends what you mean by "up there". ChatGpt tells me you'd free fall from 1000 km to 100km in about 8 minutes. It also did the math that you'd need 1.65G of sideways thrust to reach orbital speed. That's quite a bit of force for spacecraft sized objects.
If you have an actual space elevator, sure, you can go to close to geosynchronous altitude and by that time you'd have enormous sideways velocity just by being dragged sideways by space elevator and indeed it would be easy to propel yourself to orbit (above a certain altitude my intuition tells me you could let go of the rope and while you'd end up on an eliptise you'd still be in orbit)
I missed the first one, when will the second one be?
Not an agent, but I've seen people choose doctors based on asking ChatGpt for criteria and the did make those appointments. Saved them countless web interfaces to dig through.
ChatGpt saved me so much money by searching for discount coupons on courses.
It even offered free entrance passwords on events I didn't know had such a thing (I asked it where the event was and it also told me the free entrance password it found on some obscure site).
I've seen doctors use ChatGpt to generate medical letters -- Chat Gpt used some medical letters python code and the doctors loved the result.
I've used ChatGpt to trim an energy bill to 10 pages because my current provider generated a 12 page bill in an attempt to prevent me from switching (because they knew the other provider did not accept bills of more than 10 pages).
Combined with how incredibly good codex is, combined with how easily chat gpt can just create throw away one-time apps, no way the whole agent interface doesn't eat a huge chunk of the traditional UX software we are used to.
I'm not trying to find a gotcha, but wonder how that works with paying by phone nfc and then by card?
Strong emotions can override reason (and wrong reason can create strong emotions that create tight loops/patterns one cannot escape), and there's no better way to train yourself out of a pattern then by rehearsing fighting the pattern and overcoming it.
One could be excused into thinking that that's all there should be about psychotherapy, but there exist meta-loops that spawn bad loops ad infinitum and you can spend a lot of time fighting the bad thing that can be seen and not understand why no real life progress can be made.
So for me, there are many many successful point therapies, and much much fewer higher level successful therapies where chatgpt or low skill therapists won't really help.
> The question of the existence of the unconscious is crucial, because if CBT starts to take into account the possibility of unconscious meanings and logic, then it would become another branch of psychoanalysis.
I read quite a few books, from Transactional Analysis, to DBT therapist manuals, CBT manuals, and the one I loved the most and which actually helped me the most cure my BPD -- the Schema Therapy manual.
CBT seems very well suited/tooled for point therapies, very through, asks you to do homeworks, note down thoughts, I can absolutely see how it works very well for a broad spectrum of point issues.
Schema Therapy I find like it is a 3rd generation/refinement of the whole "unconscious" thing -- for me it made a lot of sense coming from TA with the Parent-Adult-Child, but I found ST approach to modes a lot more actually therapeutical -- what's the point to have mythological "Unconscious"/theory of mind if it's not actually therapeutical?
Reading Jeffrey Young's ST handbook and watching him speak actually made me cry -- I bet he would be an awesome therapist NO MATTER the formation, because he's a great human being and archetype of a THERAPIST: humble, empathetic, curious, calm, scientific, systematic and riguros, no-nonsense, results oriented -- and that's what EVERY therapy system out there should strive to achieve in creating therapists that actually help people.
BUT the question is what is "help" ? And for me, beyond anecdote the benchmark of therapies is curing BPD/NPD :)
Looking for other reasons behind procrastination is very important, indeed.
There can be many, many core beliefs that hold you down.
This will sound cliche and 70's pop-psy self-help, but people think about themselves as an adult of age XX and don't realize many core ideas about themselves are not those of an adult, but those of themselves at age 7.
My example is that since my daughter was born I was using on her a blessing my grandma was always using on me, and I did not realize I was miss gendering her -- I was using the masculine form and my daughter eventually asked me about it -- why was I using the masculine form on her -- it then struck me I heard the blessing from my grandma when I was very young and it just became a core part of me.
That's cute, until you realize you internalize A LOT of stuff by the time you're 7 and unfortunatelly it's not always positive stuff.
My father did a lot of good things for me, but he was very competitive, he almost NEVER let me win at anything to the point he became visibly distraught when I was about to win against him, so I struggle to capitalize on my insights, especially when I have strong "about to win" feelings which turned into a life long self-inflicted "Cassandra curse".
Do you think by the time GPT-9 comes, we'll say "That's it, AI is a failure, we'll just stop using it!"
Or do you speak in metaphorical/bigger picture/"butlerian jihad" terms?
EDIT: clearly better than Claude or any other model that I tried before. I had a bonus benchmark -- add a narrow triangle on the head of the snake that indicates the direction of movement, after a single iteration GPT-5 fixed it whereas Claude could never get the rotation of the triangle right, nor could o3 the last time I tried.
I think the plan is not just words, if it was, you could read a book on how to ride a bike.
Because we communicate in language and because code output is also a language we think that the process is also language based, but I think it's not, especially when doing hard stuff.
I know for certain in my case it isn't -- when tracking a hard problem for a junior after 2 hours of pair programming the other week, I had to tell him to commit everything and just let me do some deep thinking/debugging and I solved the problem myself. Sure I explained my process to him in language the best I could, but it's clear it was not language, it was not liniar, I did not think it step by step.
I wish I could explain it, but when figuring out a hard problem, for me it takes some time to take it all in, get used to the moving parts, play with them. I'm sure there are actual neurons/synapses formed then, actual new wires sprawling about in the brain, that's why it takes time. I think the solution is a hardware one, not a software one.
That's why we can sleep on it and get better the next day and that's why we feel the problem. There are actual multiple paralel "threads" of thinking going at the same time in our heads and we can FEEL the solution as almost there.
I think it simply is that hard problems can occur in a combination of code, state, models that simply cannot be solved incrementally and big jumps are necessary.
I'm not saying the problem cannot be solved incrementally, but it's possible that by going in small steps, you either reach the solution or a blocker that requires a big jump.
My go to benchmark is a 3d snake game Claude does almost flawlessly (or at least in 3-4 iterations)
The prompt:
write a 3d snake game in js and html. you can use any libraries you want. the game still happens inside a single plane, left arrow turns the snake left, right arrow turns it right. the plane is black and there's a green grid. there are multiple rewards of random colors at a given time. each time a reward is eaten, it becomes the snake's new head. The camera follows the snake's head, it is above an a bit behind it, looking forward. When the snake moves right or left, the camera follows gradually left or right, no snap movements. write everything in a single html file.
EDIT: I'm not trying to shit on GPT-5, so many people here seem to be getting very good results, am I doing something wrong with my prompt?
I got open eye visuals from smoking some dubious hashish, 2 other people smoking the same stuff had no issues. I really enjoyed the experience, except for a small incident where I woke up in the middle of the night and started slapping my snoring friend -- I hallucinated he was in fact chocking on his own vomit and was going to die and in my mind I was saving his life.
I also got mild HPPD that got better in a few years -- mainly I could not look at certain grid patterns because bright flashes of light would come from the intersection -- this mostly happened from vent iron grids and some shirts of mine.
The problem is not the agentic architecture, the problem is the LLM cannot really add knowledge to itself after the training from its daily usage.
Sure, you can extend the context to milions of tokens, put RAGs on top of it, but LLMs cannot gain an identity of their own and add specialized experience as humans get on the job.
Until that can happen, AI can exceed algorithms olympiad levels, and still not be as useful on the daily job as the mediocre guy who's been at it for 10 yers.
The israelis used the Spike missile for DECADES (they even had mock tanks launching it) and kept it quiet.
The ukrainians immediately uploaded videos of them using FPVs, I was sure at that time that the russians will in no time catch up and surpass ukraine in their FPV drone usage.