According to a new research, scientists have created an AI system that can solve math problems at the grade school level.
While language manipulation is one of AI models’ strong points, the multi-step reasoning required to solve arithmetic issues has always been a step too far.
The AI system, on the other hand, was capable of solving around 60% of the problems it encountered.
In fact, researchers at OpenAI (the firm behind language model GPT-3) claim to have developed a method to recognize its own errors, allowing it to reassess until it finds a viable answer. The children’s test results were 60 percent from the OpenAI database and 55 percent for the AI system.
Maths is a discipline at which AI excels.
The advent of artificial intelligence (AI) algorithms capable of solving even basic math challenges, as projected by OpenAI, is a significant step forward and will offer a number of possibilities, according to the organization.
“One significant challenge in mathematical reasoning is the high sensitivity to individual mistakes,” explained the researchers. “Autoregressive models, which generate each solution token by token, have no mechanism to correct their errors. Solutions that veer off-course quickly become unrecoverable.”
To address this issue, OpenAI developed a verifier tool that allowed the AI model to be used without human intervention.
Instead of relying on humans to evaluate the answers given by the AI model, OpenAI employed a set of verifiers who were responsible for evaluating those responses.
These verifiers were given 100 possible solutions generated by the algorithm and then assigned with determining whether any of them was correct.
“Providing correct arguments and recognizing incorrect ones are key challenges in developing more general AI,” OpenAI added.
“[Grade school] problems are conceptually simple, yet one subtle mistake is enough to derail an entire solution. Identifying and avoiding such mistakes is a crucial skill for our models to develop.”
Paradoxically, according to the company, as AI is applied to solve more complex problems, the verification procedure that allows its AI systems to solve simple arithmetic questions with acceptable accuracy will become increasingly important.
The ability to modify how AI solves a problem may be transformative in conjunction with advances in the field of semiconductors, which will allow for AI models that are many times larger (and therefore more capable) than they are today.
Subtly charming pop culture geek. Amateur analyst. Freelance tv buff. Coffee lover
from WordPress https://ift.tt/3EXBzLD
via IFTTT
0 Comments