PR and list, I am replying with a non sequitor I began working this with with a user named kta from #anarchism:matrix.org to work on an AI anarchy bot ( https://github.com/9K3G27BXx5xkYy6G/anbot ). The original mission statement was to “answer political and economic questions with song lyrics in the style of Chumbawamba.” I have succeeded at making 3 separate parts in the codebase: - example code that connects to matrix in a swappable manner (chat.py) - example code that connects to a swappable language model service and answers questions anarchically (open_ai.py) - example code that downloads the pages of anarchistfaq.org, loads them into an embedding database, and extracts knowledge from them that relates to a query (anarchistfaq.py) I would likely next use the qa features from langchain ( https://langchain.readthedocs.io/en/latest/use_cases/question_answering.html ) to tie the behavior of answering questions with song lyrics, to the knowledge in anarchistfaq.org . I am presently quite burned out, confused, etc, although that could change. I haven’t been speaking with the person who owns the project. If anybody would like to try to download the vector database i generated of anarchistfaq.org in a format loadable by the example code, i am presently magic-wormhole sending it from a firewalled system with wormhole code 6-microwave-pupil . Likely this is not very useful but exists! Here are example outputs from the codebase: $ python3 -m anbot.open_ai # answers question without specific knowledge What is the best fiscal policy? Burn all the money, kill all the kings, It's the only way to make everything right, Burn all the money, kill all the kings, It's the best fiscal policy tonight. $ python3 -m anbot.anarchistfaq # extracts knowledge from website 4 portions that relate to "the future": [ Document(page_content='"We have learned that instead of a jump into the realm of the Children of the Future, we can hope for no more than a steady advance, in which the healthy new overlaps the sick old structure, with the new slowly outgrowing the old." [Children of the Future, pp. 38-39]', lookup_str='', metadata={'source': 'https://www.anarchistfaq.org/afaq/sectionJ.html'}, lookup_index=0), Document(page_content='The future workplace would be an expression of the desires of those who worked there. It would be based around a pleasant working environment, within gardens and with extensive library, resources for education classes and other leisure activities. All this, and more, will be possible in a society based upon self-realisation and self-expression and one in which individuality is not crushed by authority and capitalism. To quote Kropotkin, the future workplace would be "airy and hygienic, and consequently economical, factories in which human life is of more account than machinery and the making of extra profits." [Op. Cit., p. 197] For, obviously, "if most of the workshops we know are foul and unhealthy, it is because the workers are of no account in the organisation of factories". [The Conquest of Bread, p. 121]', lookup_str='', metadata={'source': 'https://www.anarchistfaq.org/afaq/sectionI.html'}, lookup_index=0), Document(page_content='"The notion that human beings discount the future certainly seems to correspond to everyone\'s subjective experience, but the conclusion drawn from it is a non sequitor, for most people have enough sense to want to be able to exercise consuming power as long as fate permits, and many people are in the situation of having a higher income in the present than they expect in the future (salary earners will have to retire, business may be better now than it seems likely to be later, etc.) and many look beyond their own lifetime and wish to leave consuming power to their heirs. Thus a great many . . . are eagerly looking for a reliable vehicle to carry purchasing power into the future . . . It is impossible to say what price would rule if there were a market for present versus future purchasing power, unaffected by any other influence except the desires of individuals about the time-pattern of their consumption. It might will be such a market would normally yield a negative rate of discount . . .', lookup_str='', metadata={'source': 'https://www.anarchistfaq.org/afaq/sectionC.html'}, lookup_index=0), Document(page_content='Ultimately, what the state and capital gives, they can also take away. What we build by our own self-activity can last as long as we want it to and act to protect it:\n\n"The future belongs to those who continue daringly, consistently, to fight power and governmental authority. The future belongs to us and to our social philosophy. For it is the only social ideal that teaches independent thinking and direct participation of the workers in their economic struggle. For it is only through the organised economic strength of the masses that they can and will do away with the capitalist system and all the wrongs and injustices it contains. Any diversion from this stand will only retard our movement and make it a stepping stone for political climbers." [Emma Goldman, Vision on Fire, p. 92]', lookup_str='', metadata={'source': 'https://www.anarchistfaq.org/afaq/sectionJ.html'}, lookup_index=0) ] For a specific response, the langchain qa feature would prompt the model to discard irrelevent information from the extracted knowledge and summarize points that relate to the conversation. It could then form e.g. song lyrics or anything else, based on the information in the vector database and document store. It is simple how it does this, if one looks at the langchain source. It just tells the language model to do it. The code is written to use openai; for me that’s just so i psychologically clamp up slower. My openai key is included in the source. The saass backend is swappable for a local one like huggingface or a community compute network like koboldai. Whoever you are who has read this far, the remaining parts of our consciousnesses thank you for your attention and tolerance.