After all the verified success with ChatGPT, there’s now a new solution open for it…anyone can ever get it to work.
This week, Philip Wang, a programmer who has actually reverse engineered the Meta AI system in the past, Enabled The new PaLM + RLHF, a cross-AI text generator that works in a format similar to OpenAI’s ChatGPT – with the peculiarity of being completely open and free.
The system combines Google’s PaLM with a technology known as Reinforcement Learning with Human Feedback – RLHF – to create a system capable of operating in a format similar to ChatGPT, including the ability to write emails and even create small programmatic content. Through AI device learning.
However, unlike the OpenAI solution, PaLM+RLHF does not contain any training models. As such, downloading its code won’t offer the same ultimate experience you’d find in ChatGPT, unless someone can get models to train with high precision.
Like many other AI systems, PaLM+RLHF works by analyzing input models, which can be from various sources, that are used to train the AI to perform its tasks. And this is exactly where the problem lies for anyone who intends to use it, as it will be necessary to “feed” it with enough information so that it can provide tangible details that even resemble what is in ChatGPT.
The problem is that most users don’t have these models available for collection, and they also don’t have the hardware to make them work properly. The task of feeding models to the program is very demanding in terms of resources, and far from what most users have in terms of processing power for such a task.
A study by AI21 Labs, conducted in 2020, estimated that training a model with “only” 1.5 billion parameters could cost about $1.6 million—and these amounts are relatively small if you consider all the information that needs to be processed to create something consistent.
So, while the solution is interesting to analyze the capabilities of the system, at the same time it would be something that hardly creates any kind of competition with what’s in OpenAI’s ChatGPT.
