Things like this are absolutely necessary for substance use! Limiting the amount of information people can access on these topics only ends up harming them as reliable and sometimes life saving information is withheld. This is a really great use of LLM and I hope that you and your team continue to make ideas like this a reality! The bot works great and gave me some very useful and genuinely well written/structured information so thank you
Amazing bot just asked some questions about nn dmt and my own subjective experience I would have otherwise needed to scrape the internet for and got very solid answers in a very short amount of time, absolutely cannot wait to see how this ai evolves in the future, great works Josie 👽🙏
@@josikinz I was curious as to why 15mg of dmt gave me no visuals and the answer was along the lines of it depends on brain chemistry method of smoking aswell as overall mood and mental state I also asked if altered states of consciousness can really be achieved through means other than psychedelics and the answer was yes, things like deep meditation sensory deprivation and even shamanic drums are all good methods to achieve these states of being, Overall 10/10 ai, does exactly what you need it to and I seriously can't wait to see how far it goes, it's a very useful thing in the community seeing as most Google searches just bring up 1800 hotline and news articles so this ai is super beneficial 👍👍
I definitely enjoyed my trial questions and I must say you all did a great job as far as I can tell. I didn't get into subjective effects, but it handled some pretty challenging questions quite well. It even reminded me to consider tyramine levels in foods when looking to extend the duration of a DMT freebase experience by using MAOIs due to tyramine's negative side effects when not broken down by monoamine oxidase sufficiently. It would be fulfilling and fun as hell to work on this. Josie et al., thank you for your time and effort. I think users will have a blast with this and if it's that good already, you've set yourself up with something really nice here. Peace and farewell.
What a wonderful service to the world. The harm reduction category is an enormous and helpful resource for all humans. Great early and useful application of AI too!
I’m pretty sure this stems from a gap in training data (and arguably research) but I found a significant knowledge gap in its response to “What effects do GLP-1 agonists have on ingesting magic mushrooms?” Specifically, it did mention the well-documented effect of GLP-1 agonists causing delayed gastric emptying, but being an LLM lacking the ability to reason, didn’t elaborate on that having potentially significant consequences on the overall process of orally consuming psilocybin.
this still seems risky, but probably not *as* bad as like, AI-written mushroom foraging guidebooks ...though, I'd wanna put a warning to not ask it for dosage information, just to be safe - you dont want the chance of getting an ai-hallucinated answer for that - and same for drug combos putting "Your go-to guide for informed and responsible exploration. Get accurate, unbiased, and substance-agnostic answers to all your drug-related questions!" as the description worries me somewhat - I wouldn't quite trust an AI to give accurate dosing info, combo safety, or precise technical knowledge, but general information like subjective effects would probably be fine I'd like to clarify that I *do* think this is cool It could be useful to instead put it in a public chatroom, so other people can reply to the answers it gives, just as a safeguard against potential errors - though that would make it harder to privately ask questions, it can probably still be preserved - maybe you DM the bot the question and it links you to an answer in the public chat, where it doesn't say *who* asked it, just that the question was asked?
If the hyperparameters (fancy word for settings) are set properly, you should get no hallucinations aside maybe some marginal edge cases (i.e you actively try to fool the bot). I haven't tried this bot in particular, so I don't know how well it actually performs, but technically you can make a language model that adheres 100% to training data. The reason the likes of ChatGPT hallucinate occasionally is actually a feature, not a bug. It stems from a few different hyperparameters, especially one called repetition penalty, where the AI is penalized for repeating things, as well as the sampling method (how it's decided which token to pick next). Penalizing repetition and adding an element of randomness to the sampling will lead to a more "creative" AI, whereas setting these parameters to always favor the most likely next token without penalizing repetition will lead to deterministic and rigid answers, but also ones that tightly follow the training dataset. ChatGPT and others occasionally lie not because it's impossible to make them not lie, but because a bot that lies is more marketable and cheaper to create. I'm not going to go much further than that in a youtube comment, but there are some other factors that go into it too. Such as, if there is misinformation in the training data (as in, the training data was not properly curated and cleaned, a very labor-intensive project), then there is still a chance that the bot will repeat misinformation. At the end of the day, companies like OpenAI will always end up acting in the interest of profit, so they will always include as much "dirty" data as they can get away with. A specialty bot trained to answer questions in a narrow domain (such as this one) is much easier to make truthful, especially since you don't need to care about stuff like creativity. TL;DR: Hallucinations and misinformation isn't an inherent feature of language models. Therefore, if created and set up well, a bot might never get anything wrong as long as the subject matter lies within its domain of knowledge.
Do you have any plans to release the model eventually, so that anyone can run it on their own machine, or is this a strictly for-profit endeavour? Are there going to be more access (web UI, stand alone app) and payment (bitcoin or other crypto) options in the future?
A few suggestions. Please list all source of information that generates the LLM. I realize this might be protected to prevent someone else from profiting on your work. But we need links to scientific research to back up some of this advice. Maybe links to published articles or writings. That way people can judge the sources for themselves. Should we trust them. I realize many sources of info may not have under peer review. Also consider the impact this could have on private health information. I understand you are not recording conversations. But as you alluded to Discord could be. A P2P model in a indy app might be safer. Or some kind of hard encryption In between bot and user. If it done locally on a device interest and doesn’t go to the Internet that would be better. All kinds of man in the middle etc hacking could collect very personal and exploitable information. Not protecting PHI violates the US HIPA laws. Please consider that some orgs like FDA in the US may consider this app providing medical advice without a license. There are US state orgs that govern this also. I not a lawyer nor a physician. But you and your collaborators need to secure the services of one thats likely a MD and attorney. With experience in law in multiple countries. You are on the forefront of tech that many very large and wealth med companies and orgs are trying to innovate in. So be careful. Cheers and good luck! I hope your work keeps many people happy and safe!
If the bot were claiming to generate, link to, or regurgitate medical advice, then HIPAA would be relevant. The bot does not offer or promise or offer medical advice.
Just watched this video and think it's pretty uplifting given the subject matter "Soft White Underbelly interview and portrait of Richard, a DMT user in Los Angeles."
I often consider the topic of addiction. Addiction is probably thought of as a mostly or entirely physical thing, depending on who you ask. One thing I have been recently considering is the possibility of what I would refer to as habituation to distinguish it from addiction. Essentially, you become interlinked with whatever it is your addicted to at an almost cultural, subconscious level. Is smoking weed dangerous? Maybe but probably no more than drinking alcohol or smoking cigarettes. Is smoking weed everyday dangerous? Not immediately, but probably to some degree in the long-term.
@@ancwhor Definitely more trustworthy, but I also wouldn't just take any random paper at face value. There's a spectrum here, and I think this bot sits somewhere between published biochemists and crackhead Dave from down the street.
I would really like to join your discord, but when I click the link it doesn't take me to any specific discord page. Could you give me the name of it? Or maybe send me an invite?
The world is blessed to have people with talent
Absolutely amazing! Great cross of AI and pschedlic education
out here doing the gods work
thank you
literally lol
Important work. Thank you. ❤
good bot 💜
woaaaaaaah, thank you so much for this contribution to collective intelligence again and again 🙏
Wsp stingray can wait to see the next replication, been with you for a while now and really admire your work in the community 👽🙏 keep it up
Fascinating as always Josie.
Thankyou and have great yuletide.
Happy Solstice 💚
Things like this are absolutely necessary for substance use! Limiting the amount of information people can access on these topics only ends up harming them as reliable and sometimes life saving information is withheld. This is a really great use of LLM and I hope that you and your team continue to make ideas like this a reality! The bot works great and gave me some very useful and genuinely well written/structured information so thank you
it better always recommend boofing over insufflation
Why?
@@craigd123 mainly a joke but boofing just works well
Amazing! This is a great idea
Thank you :]
Amazing bot
just asked some questions about nn dmt and my own subjective experience I would have otherwise needed to scrape the internet for and got very solid answers in a very short amount of time, absolutely cannot wait to see how this ai evolves in the future, great works Josie 👽🙏
Woah, what did you ask it?? I’m really glad it helped!
@@josikinz I was curious as to why 15mg of dmt gave me no visuals and the answer was along the lines of it depends on brain chemistry method of smoking aswell as overall mood and mental state
I also asked if altered states of consciousness can really be achieved through means other than psychedelics and the answer was yes, things like deep meditation sensory deprivation and even shamanic drums are all good methods to achieve these states of being,
Overall 10/10 ai, does exactly what you need it to and I seriously can't wait to see how far it goes, it's a very useful thing in the community seeing as most Google searches just bring up 1800 hotline and news articles so this ai is super beneficial 👍👍
wowsie 🙏🏻❤
This is great! Thank you! ❤
Hi Josie hi Emily hope all is well,SAFE TRAVELS
I definitely enjoyed my trial questions and I must say you all did a great job as far as I can tell. I didn't get into subjective effects, but it handled some pretty challenging questions quite well. It even reminded me to consider tyramine levels in foods when looking to extend the duration of a DMT freebase experience by using MAOIs due to tyramine's negative side effects when not broken down by monoamine oxidase sufficiently.
It would be fulfilling and fun as hell to work on this. Josie et al., thank you for your time and effort. I think users will have a blast with this and if it's that good already, you've set yourself up with something really nice here.
Peace and farewell.
What a wonderful service to the world. The harm reduction category is an enormous and helpful resource for all humans. Great early and useful application of AI too!
❤
Thank you!
good shit mate
I’m pretty sure this stems from a gap in training data (and arguably research) but I found a significant knowledge gap in its response to “What effects do GLP-1 agonists have on ingesting magic mushrooms?”
Specifically, it did mention the well-documented effect of GLP-1 agonists causing delayed gastric emptying, but being an LLM lacking the ability to reason, didn’t elaborate on that having potentially significant consequences on the overall process of orally consuming psilocybin.
Wow, cool. I tried it tonight, it even gives advice on how to stop smoking.💙😎
Hay Josie busy,busy busy rock n roll .
this still seems risky, but probably not *as* bad as like, AI-written mushroom foraging guidebooks
...though, I'd wanna put a warning to not ask it for dosage information, just to be safe - you dont want the chance of getting an ai-hallucinated answer for that - and same for drug combos
putting "Your go-to guide for informed and responsible exploration. Get accurate, unbiased, and substance-agnostic answers to all your drug-related questions!" as the description worries me somewhat - I wouldn't quite trust an AI to give accurate dosing info, combo safety, or precise technical knowledge, but general information like subjective effects would probably be fine
I'd like to clarify that I *do* think this is cool
It could be useful to instead put it in a public chatroom, so other people can reply to the answers it gives, just as a safeguard against potential errors - though that would make it harder to privately ask questions, it can probably still be preserved - maybe you DM the bot the question and it links you to an answer in the public chat, where it doesn't say *who* asked it, just that the question was asked?
Good idea
If the hyperparameters (fancy word for settings) are set properly, you should get no hallucinations aside maybe some marginal edge cases (i.e you actively try to fool the bot). I haven't tried this bot in particular, so I don't know how well it actually performs, but technically you can make a language model that adheres 100% to training data.
The reason the likes of ChatGPT hallucinate occasionally is actually a feature, not a bug. It stems from a few different hyperparameters, especially one called repetition penalty, where the AI is penalized for repeating things, as well as the sampling method (how it's decided which token to pick next). Penalizing repetition and adding an element of randomness to the sampling will lead to a more "creative" AI, whereas setting these parameters to always favor the most likely next token without penalizing repetition will lead to deterministic and rigid answers, but also ones that tightly follow the training dataset. ChatGPT and others occasionally lie not because it's impossible to make them not lie, but because a bot that lies is more marketable and cheaper to create.
I'm not going to go much further than that in a youtube comment, but there are some other factors that go into it too. Such as, if there is misinformation in the training data (as in, the training data was not properly curated and cleaned, a very labor-intensive project), then there is still a chance that the bot will repeat misinformation. At the end of the day, companies like OpenAI will always end up acting in the interest of profit, so they will always include as much "dirty" data as they can get away with. A specialty bot trained to answer questions in a narrow domain (such as this one) is much easier to make truthful, especially since you don't need to care about stuff like creativity.
TL;DR: Hallucinations and misinformation isn't an inherent feature of language models. Therefore, if created and set up well, a bot might never get anything wrong as long as the subject matter lies within its domain of knowledge.
💚
Do you have any plans to release the model eventually, so that anyone can run it on their own machine, or is this a strictly for-profit endeavour? Are there going to be more access (web UI, stand alone app) and payment (bitcoin or other crypto) options in the future?
based
Oh thank god, I was worried it would be cringe
It seems similar to an upscale version of “Sims City”!❤
the computer game from the 90s?
@@quillclock Yes, indeed I have not played that game years ago but I saw my younger brothers play😎
hey wait, what did you use to make this? it can't be chatgpt since that's censored?
VEry good
A few suggestions. Please list all source of information that generates the LLM. I realize this might be protected to prevent someone else from profiting on your work. But we need links to scientific research to back up some of this advice. Maybe links to published articles or writings. That way people can judge the sources for themselves. Should we trust them. I realize many sources of info may not have under peer review.
Also consider the impact this could have on private health information. I understand you are not recording conversations. But as you alluded to Discord could be. A P2P model in a indy app might be safer. Or some kind of hard encryption In between bot and user. If it done locally on a device interest and doesn’t go to the Internet that would be better. All kinds of man in the middle etc hacking could collect very personal and exploitable information. Not protecting PHI violates the US HIPA laws.
Please consider that some orgs like FDA in the US may consider this app providing medical advice without a license. There are US state orgs that govern this also.
I not a lawyer nor a physician. But you and your collaborators need to secure the services of one thats likely a MD and attorney. With experience in law in multiple countries.
You are on the forefront of tech that many very large and wealth med companies and orgs are trying to innovate in. So be careful. Cheers and good luck! I hope your work keeps many people happy and safe!
If the bot were claiming to generate, link to, or regurgitate medical advice, then HIPAA would be relevant. The bot does not offer or promise or offer medical advice.
Links and citations would be clutch. Look how the ai search app perplexity works.
Just watched this video and think it's pretty uplifting given the subject matter "Soft White Underbelly interview and portrait of Richard, a DMT user in Los Angeles."
Can you please do a video about psychedelics' affinity towards cardiac receptors and its significance?
Psilosbye Mexicana, aztecorum as well as zapatecorum dossage information?
I am excited for this. What is the base model?
GPT
I often consider the topic of addiction. Addiction is probably thought of as a mostly or entirely physical thing, depending on who you ask. One thing I have been recently considering is the possibility of what I would refer to as habituation to distinguish it from addiction. Essentially, you become interlinked with whatever it is your addicted to at an almost cultural, subconscious level.
Is smoking weed dangerous? Maybe but probably no more than drinking alcohol or smoking cigarettes. Is smoking weed everyday dangerous? Not immediately, but probably to some degree in the long-term.
We are all just reflections of our higher self you are in a dream, wake up. There is no way out no way in because you are both you are everything ❤
Does NAC effect psychedelics, can If I take NAC every day is it safe to take psychedelics?
😮
Is PsyAI still up and running? I'm not getting responses
Make a salvia trip replication
Is that just in telegram ?
I made an AI chatbot of Terence McKenna.
How'd ya do that? Can I use it??! O-O
can we make it for whatsapp?
Say a cuss lol
ngl i would rather speak to someone with a body mind and soul then a computer screen but hay any awareness about safety with pyschedelics is key
mildly concerned trusting unreliable ai with safety information
You shouldn’t trust the AI. But it‘s basically the same as asking a youtuber.
@@hanfo420 which you still shouldn't do. When it comes to things that can be dangerous 😅
@@ancwhorWho should you ask that you can unequivocally trust?
@@KalebPeters99 chemists. Aka research papers.
@@ancwhor Definitely more trustworthy, but I also wouldn't just take any random paper at face value.
There's a spectrum here, and I think this bot sits somewhere between published biochemists and crackhead Dave from down the street.
I'M NOT EVEN THERE AND I CAN HEAR IT. THE TV. I CAN HEAR IT THROUGH UA-cam. AAAAAAAAAA
REALLY???
Ayo what happened in here 👀
@@josikinz yes i swear
I would really like to join your discord, but when I click the link it doesn't take me to any specific discord page. Could you give me the name of it? Or maybe send me an invite?