New research shows that giving artificial intelligence systems an “internal monologue” makes their results significantly better. Essentially, artificial intelligence has been taught to think before responding, much like humans think about what they should say next before speaking. This is different from how popular AI language models such as ChatGPT behave. The latter do not “think” about what they are writing and do not provide various possibilities for next steps in the conversation.
The new method, called Quiet-STaR, instructs the AI system to generate many internal arguments in parallel before responding to a query. When the AI responds to clues, it generates many options and outputs the best answer. After all, artificial intelligence learns by discarding options that turn out to be incorrect. Essentially, the learning method gives AI models the ability to predict future conversations and learn from current ones.
Researchers from Stanford University and Notbad AI applied the Quiet-STaR algorithm to Mistral 7B, a large open-source language model, and published the results on arXiv. The Quiet-STaR-trained version of Mistral 7B scored 47.2% on the reasoning test versus 36.3% before any training. The model still failed the school math test, scoring 10.9%. But this is almost double the 5.9% result in the initial version.
Models like ChatGPT and Gemini don't relate data to common sense or context, so they don't actually understand their own responses by simply generating words. Previous attempts to improve the “thinking” ability of language models were very specialized and could not be applied to different AI models.
The self-learning algorithm STaR, which the researchers used as the basis for their work, is one example of such learning, but it is also constrained by these limitations. The scientists who developed Quiet-STaR named the method that way because STaR's work happened in the background. This can work with different models, regardless of the original training data. Now they want to explore how similar techniques can bridge the gap between neural network-based artificial intelligence systems and human reasoning capabilities.
GPT-4 studies GPT-2 logic in OpenAI study: AI creators don't know how it works
Source: Live Science
The competition for ITS authors continues. Write an article about the development of games, gaming and gaming devices and win a professional gaming wheel Logitech G923 Racing Wheel, or one of the low-profile gaming keyboards Logitech G815 LIGHTSYNC RGB Mechanical Gaming Keyboard!