A recent spate of applications built on OpenAI’s GPT-4 API has the crypto community buzzing with designs toward the development of a fully-autonomous, self-correcting cryptocurrency trading bot.
Two such apps, dubbed ‘BabyAGI’ and ‘AutoGPT’, have received particular notice with many users attempting to build crypto trading applications 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 is evidenced by the demonstrable usefulness of the ChatGPT interface, but it has no capacity for memorization.
Applications built on the GPT API are basically limited to single-session use, meaning the model can’t recall information from previous interactions. This has to do with the amount of data (referred to as the number of ‘tokens’) individual queries require, and GPT’s tendency to hallucinate — a problem that becomes increasingly noticeable as token counts rise.
Users are, essentially, starting with a clean slate whenever they query the machine. In terms of building a crypto trading application capable of self-correction and historiographic analysis — adjusting to real-time market conditions while simultaneously keeping short and long-term trends in focus — this means even the most robust bot built on the GPT API would typically require heavy human supervision.
Some clever developers may have discerned a potential method for circumventing these limitations by building applications 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 goal of these particular apps goes beyond simply automating crypto news aggregation or teaching a machine learning agent how to recognize the dip.
AutoGPT, for example, uses GPT-4 to generate code and then exploits GPT-3.5 as what appears to be a virtual artificial memory space wherein information is combined and shuttled between the two.
Another effort, BabyAGI, combines GPT-4 with LangChain, a coding framework, and Pinecone, a vector database, to spawn new agents in order to complete complex tasks without losing focus on the original objective.
Both apps could have the potential to serve as the backbone for a multi-agent, fire-and-forget AI application capable of managing a crypto portfolio from top to bottom based on little more than plain language prompting.
While it appears neither app was specifically designed with the cryptocurrency market in mind, we’ve spotted several efforts across social media and on GitHub to adapt one or both for autonomous trading.
