as i’ve hinted on some of my socials, i’ve been working, since mid-2024, at
synthetic echo on a system i’ve named
disco.
disco
stands for discovery.
and it brings a unique, critical, and beneficial aspect to many AI applications:
the social voice
below are three examples of running some questions by disco.
in each case, the answers you’ll see were informed by bringing millions of real conversations had by real humans to bear on the question at hand.
this is in stark contrast to ‘normal’ AI, which instead presents the user with an amortized, book reportish, standard, and non-controversial answer to your query.
and that, my dear reader, with the personality of a toaster oven.
disco isn’t about being being objectively correct, it’s not trying to
tell you what to think.
instead. it uses the power of AI to tell you what people think.
this way, you can make an informed decision for yourself, after examinging the multitude of opinions out there.
disco’s goal is to be subjectively correct: to accurately reflect the
voices of people.
this is in stark contract to commerical AI, which aims to give you the most vanilla answer, with a bias towards those that motivate you to purchase, or, at minimum, to stay on the platforms of thier affiliate companies.
disco pulls no punches.
because most questions have a variety of correct answers, depending on who
you ask. disco shows you this variety, and it reveals who would have
been likely to provide that type of answer, with references to the source
material it used to form that conclusion.
in this way, you can just read it and weep ;-)
technically, and as astute readers will have no doubt noticied, it is through
these citations that disco provides verficiation of non-hallucination.
without further ado, here are those examples.
please take some time to review the output, which provides the answers of 42 discrete audiences.
pay careful attention not only to the citations provided which, as you shall see, provide support for the answers given.
as you’ve hopefully realized, after reviewing the above, disco is not your
mother’s AI.
it’s a lot more honest, a lot more realitic, and it’s grounded in real human conversation.
disco has many applications. essentially, any AI application that could, would, or should benefit from consideration of the social voice.
this includes, but is not limited to:
- running dynamic, real-time, surveys across audiences before making business decisions.
- having specific audiences review proposals or generated content and give honest feedback, in their own words.
- mocking national elections.
- understanding unfiltered thoughts about your company, practice, or brand, including dangerous or otherwise censored information (we have 4chan…)
- actually understanding the nuances of the social conversation, instead of counting clicks, keywords, and search terms.
- and much, much more.
the above was a simple demonstration. the technology can be brought to bear on any problem a general purpose AI can be. which is to say, almost all of them.
as present, i am working hard to:
- add substack content to
disco - add bluesky content to
disco - release an API that developers can use to integrate
discos RAG context in thier own AI applications - release a basic UI for non-technical people to leverage
disco - produce a threat index of the fortune 500 using 4chan, and other ‘gray’ sources
- raising a round
reach out if you wanna stay in touch regarding the project.
ps. for the curious, here’s the audience configuration used in these examples.
pss. yes, i support custom audiences of arbitrary data sources and authors. segments #ftw