“It’s good at answering questions related to Bitcoin and the economy — at least better than GPT-4,” Aleksandar Svetski told Cointelegraph at the bustling Amsterdam Bitcoin Conference.
The entrepreneur, author and founder of the new artificial intelligence (AI) Large Language Model (LLM), Spirit of Satoshi, begins to describe his small startup’s journey to create a Bitcoin-centric AI chatbot.
The model is the result of a time-consuming training process that generates responses based on reputable Bitcoin resources, Austrian School of Economics and liberal ideals.Still in its infancy, The Spirit of Satoshi reflects the ideals of a “carefully curated Bitcoin corpus” that includes resources such as Saifedean Ammous’ best-selling book Bitcoin Standard.
Svetski explained that the main difficulty in building the model was not only gathering relevant sources of information from books, research papers, and podcasts, but also guiding the model to generate responses through an exhaustive training process. He added that a common misconception about LL.M.s is that they source information like a search engine:
“They’re just probabilistically stringing words together in a way that represents patterns within the model. So, it’s not even buying anything.”
Svetski explains that this is part of the reason why AI chatbots sometimes experience “hallucinations” and why developing an LL.M. requires focusing on training its answering style. The Satoshi spirit is by no means perfect, at least in its current version.
“Our model will also hallucinate. It will also say some nonsense, but it will say something more like what a Bitcoin enthusiast would say.”
Svetski’s team built a broad but targeted base of Bitcoin-centric information and data and began using programmatic methods to feed the model tens of thousands of question-and-answer pairs. However, a human element is still needed to help the Satoshi spirit generate the kind of reaction that might come from its namesake.
related: Bitcoin Amsterdam: Bitcoin shines in the depths of crypto bear market
As such, continued development of the model relies on the broader Bitcoin community. Spirit of Satoshi uses an incentivized process to allow the public to validate, build, and validate the model’s data.
Using credentials from the Lightning Network, Nostr, or email addresses, the “proof of knowledge” mechanism allows users to be paid satoshis for helping train models.
The process uses a consensus model that automatically imposes penalties if users create “junk data.” Svetski describes this as the key “human” element that enhanced Satoshi’s mental output:
“It’s producing incredible content; it’s the final piece that takes your content from 80% good to 95% good. This has a huge impact on the quality of the model.”
Svetski said the difference between the responses generated by Spirit of Satoshi and ChatGPT was noticeable. The latter was trained in mainstream thinking on concepts such as Bitcoin (BTC) and inflation:
“If we asked ChatGPT about inflation, it would tell you that it’s a sign of a healthy economy. Well, that’s not the case; inflation is a sign of systemic problems, such as a decrease in your purchasing power.”
Svitsky said that this situation is Purpose Behind the Spirit of Satoshi Nakamoto, retrain the LL.M. to reflect the nuances that embody the type of thinking behind the Bitcoin movement:
“If you ask about inflation, our model should say, ‘No, inflation is actually bad for the economy because it discourages saving,’ or ‘Saving has a knock-on effect on people’s time preference.'”
According to its founders, the future of the platform is quite open. The spirit of Satoshi Nakamoto can be a learning tool or online tutor embedded in an educational platform or online university. It may also become the basis for the “ultimate Bitcoin influencer” through its BTC-centric output:
“I’d like to see it become a destination for the next 100 million or 500 million people who want to learn about Bitcoin — that’s their first step into understanding Bitcoin.”
The Satoshi ethos is built on an existing open source model with inherent English proficiency and “Wikipedia-like bias.” The latter was addressed by modeling responses to its Bitcoin and Austrian Economic Principles data sets.
Magazine: “Elegant and Left Behind”: Jameson Lopp’s First Impressions of Bitcoin
Svlook
