The crypto community is excited about recent applications utilising OpenAI's GPT-4 API. Specifically, two apps named "BabyAGI" and "AutoGPT" are receiving attention for their potential to develop a self-correcting crypto trading bot.
The focus of these apps is to enhance GPT-4's natural language processing capabilities with task management and memorisation abilities to improve crypto trading applications.
Currently, GPT-4 can process language effectively, but it lacks the capacity for memorisation.
Applications that use the GPT API can only be used for one session at a time, and the model cannot remember information from previous interactions. This is because the amount of data, or "tokens," required for each query is limited, and GPT has a tendency to generate unrealistic information as the token count increases.
Therefore, users have to start from scratch every time they use the machine. This limitation makes it challenging to build a fully autonomous and self-correcting crypto trading bot which can analyse market conditions and trends in real-time while maintaining a long-term perspective.
A bot built on the GPT API would still require significant human supervision.
Some developers have come up with an idea to bypass the limitations of GPT API by creating apps that utilise its ability to generate code and connect to external sources.
While there are already trading bots in existence, these apps aim to achieve more than just automating crypto news gathering or training a machine learning agent to recognise market dips.
AutoGPT and BabyAGI are two applications that leverage GPT-4 and other coding frameworks to develop agents that can complete complex tasks and retain information over multiple sessions.
AutoGPT uses GPT-4 to generate code and creates an artificial memory space that allows information to be shared between GPT-4 and GPT-3.5.
Meanwhile, BabyAGI combines GPT-4, LangChain, and Pinecone to create new agents that can manage a crypto portfolio through plain language input.
Although neither app was initially designed for crypto trading, some developers are attempting to modify them for autonomous trading purposes. These apps could potentially serve as the backbone for a multi-agent AI application that manages a crypto portfolio with little human intervention.
The writer is the founder at yMedia. He ventured into crypto in 2013 and is an ETH maximalist. Twitter: @bhardwajshash