There are no doubts, AI has been active in many fields and having its own fans and enemies. Regardless of our agreement to give it a voice to express itself or not, there are still other ways to use AI such as just extracting existing info directly as they are with no change, which can be hugely beneficial .
There is this general view about AI that they hallucinate which is true but it is avoidable. They are often big models which are holding lots of info, so the hallucination can happen. But there are ways to avoid it if you ātrain a small model for specific tasksā.
There are thousands of Suttas and we may often have difficulty to find the ones which would match the situation we are encountering. It is a common difficulty most Buddhists have with suttas. A fine-tuned ragged AI model can give you exactly what you inquire . I hope someone who is into training models and a buddhist as well, sees this someday and does something. Weād be all thankful.
[quote=āPasanna, post:3, topic:40913, full:trueā]
Just a heads up. Folks around here have some pretty strong views about AI for the suttas.
Hmmā¦. I wonāt say too much, but people often project their own fears for that which they donāt understand yet. It reminds me of people who used to oppose Cars when they were first invented. They were calling it artificial monsters that shouldnāt replace soulful horsesā¦
There is so much AI hatred these days and this is a period of history that has risen and would pass away as well. It is quite ironic to me as well that we buddhists all talk about Anatta, yet some of us are strictly looking for a self in AI and when we donāt find it we get mad about it. No matter we want AI or not, the technology is happening. So better to train it and master it so it would be used in more accurate and beneficial way than something else
Thatās what their page says ā This is a hybrid search system that combines semantic vector search with traditional full-text search ā itās not an LLM wrapper that generates responsesā
Just checked it and tried it now. It is not AI, but nevertheless looks great and its responses are reasonably good! Going to use it more often.
I did read actually and i see too much phobia and enmity in what is said. Just to say, AI is in baby-stage yet and should not be fully relied on which we would agree on, does not need such degree of demonizing towards the technology the way it is done with those posts and specifically their clickbait titles.
I would call those people prescient. Artificial Monsters sounds about right.
According to the World Health Organization (WHO), road traffic injuries caused an estimated 1.35 million deaths worldwide in 2016. That is, one person is killed every 26 seconds on average.
Wow! 1.35 million. I knew it was large, but that figure surprised me when I looked it up. There are many other problems with cars such as local pollution, both from exhaust emissions and noise of their engines and of course we also have the cars impact on climate change.
Some people can asses the impact of new technologies weighing up the benefits and dangers, and others canāt see the dangers even when those dangers are evident.
Iām definitely going to call cars Artificial Monsters from now on
This line of thinking completely overlooks other details - such as, how many people have managed to find the medical needs met just because they were able to travel to a hospital faster? How has the society improved with the speed of knowledge traveling faster with cars?
Some people can assess the impact of new technologies weighing up the benefits and dangers, and others canāt see the benefits even when those benefits are evident.
Well yes, I am definitely someone who thinks that all cars should still have a man with a red flag walking ahead of them for the safety of pedestrians. But yeah, the unbearable lightness of being and all that.
There isnāt a ācontrolā planet Earth where we successfully discouraged the use of cars, so weāll never know the answer to your questions I guess.
I do own a car and I have been using it recently to ferry cancer patients to hospital, so Iām definitely aware of this potential benefit. But I also think that caring is more important than curing, and I think that maybe this aspect of humanity has waned in the advent of the need for speed.
In addition local services have been moved further away in the rush for āefficiencyā, so for some (e.g. some poor people in rural areas) I would hazard a guess that their medical needs are being less well served.
I have benefited greatly from (fairly recently) moving to an area with relatively few people, few cars and a life lived at a slower, more human pace, with more time to address peopleās needs and build relationships. But yes, you are right, there is no longer a doctor in the village, so to get even the most basic medical care requires transport.
Thatās a loaded question. Has society improved? Dukkha in the EBTs seems pretty much the same as dukkha today. On the whole people seem as deluded today as they were in my youth (in the 1960s) and as deluded as they were in the suttas.
Which overlooks even more details like the enormous amount of damage roads and motor vehicles do to wildlife every single day ā¦ The mindless slaughter is incredible. I still remember the time I was driving at night and drove by a raccoon laying maimed in the road, not dead yet mind you ā¦
As far as LLMs go, I believe what the OP was asking about has been done already, and it didnāt work very well. The usual sort of preposterous stuff, making up sutta references, and so forth. Narrowly trained LLMs I think are for much small data sets that the Pali canon, which is a huge corpus of text in one language, much less two or more.
The search engine linked to is an interesting use of the technology to allow a more natural way to input parameters, though. Itās a way to avoid the trouble of LLMs and get some use out of them.
AI right now is pretty good at this, and a lot more. I asked the latest ChatGPT to analyze the points of contention between Ven Thanissaro and Ven Brahmaliās positions on jhana, and eventually it ended up cross referencing Pali sutta references against the Chinese Agamas and doing deep dives into the Chinese and Sanskrit phrasings and grammar, and refined its own position at every step.
Directly extracting passages as you said would not be a problem, even on a free tier of your favourite AI service. Even on a locally hosted LLM that is fairly well configured. It could be set up to provide a link to the original source on SuttaCentral, so that the user doesnāt have to take its word for it.
As with any technology that advances at such a rapid rate, people have ethical concerns and that is understandable. Some things however are objectively true and not a matter of opinion:
AI is happening, whether you like it or not. You can reject it wholesale or adapt it to your needs.
AI is improving at an astonishing rate and will only continue to do so.
There are over 2 million AI models on huggingface.co and they canāt all be horrible. And even if they are, you can train your own.
AI is a tool, albeit a fascinating one, and it is a poor craftsman who blames his tools. We use it regularly at work to bounce off ideas, never to take its output at face value and move on to the next task. And if anybody at work were to ever do that, they would have some sharp feedback pointed their way.
I would call those people prescient. Artificial Monsters sounds about right.
