A recent series of applications built on OpenAI's GPT-4 API has thrilled the crypto community with designs towards developing a fully autonomous and self-correcting cryptocurrency trading bot.
Two of these apps, dubbed “BabyAGI” and “AutoGPT,” have received particular notice, with many users trying to build crypto-trading apps on top of them.
The big idea behind both apps involves task management for GPT-4. Currently, GPT-4 excels at natural language processing, as evidenced by the demonstrable utility of the ChatGPT interface, but it has no memorization capability.
Applications built on the GPT API are essentially limited to single-session use, meaning the model cannot recall information from previous interactions. This has to do with the amount of data (called the number of "tokens") that individual requests require and GPT's tendency to hallucinate - a problem that becomes increasingly noticeable as the number of tokens increases.
Users essentially start with a clean slate every time they query the machine. In terms of creating a crypto trading application capable of self-correction and historical analysis - adapting to real-time market conditions while simultaneously keeping a focus on short-term and long-term trends - this means that even the most robust bot built on the GPT API would generally require heavy human oversight.
Related: Tech giant Alibaba to deploy rival ChatGPT AI
Some smart developers may have discerned a potential method to work around these limitations by building apps that take advantage of GPT's ability to generate code and connect to external sources.
We've seen our fair share of trading bots, but the purpose of these particular apps goes beyond simply automating the aggregation of crypto news or teaching a machine learning agent how to recognize the downside.
AutoGPT, for example, uses GPT-4 to generate code and then leverages GPT-3.5 as what appears to be a virtual artificial memory space in which information is combined and transferred between the two.
Another effort, BabyAGI, combines GPT-4 with LangChain, a coding framework, and Pinecone, a vector database, to generate new agents to accomplish complex tasks without losing sight of the original purpose.
Both apps could have the potential to serve as the backbone of a multi-agent, fire-and-forget AI app that can manage a top-down crypto portfolio based on little more than prompts. in plain language.
While it appears that neither app was specifically designed for the cryptocurrency market, we spotted several efforts on social media and GitHub to adapt one or both of them for standalone trading.