How can we evaluate an AI that is smarter than us? New group develops ‘humanity’s toughest exam’
Artificial intelligence: an alarming or disturbing Smartness?
The progress in the field of artificial intelligence is obviously progressive and is becoming more and more stupendous. Today, these machines, growing more and more efficient, can even solve the problems which demonstrate their effectiveness in defeating the rules. They seem to be that more and more challenging to beat. Such attempts to take GPT-4o to its limits in a complex series of tests appear to be a piece of cake for o1-preview, and the program is only improving.
Is the world ready for an advanced AI which has superior cognition than the benefit of mankind?
Still, it is not a wonder why some people would have such a notion that AI might be smarter than the good of human beings. Although we are not near a Skynet like scenario, this thought has not gone unanswered by technology geniuses out there.
An organization for nonprofit organization is called as The Center for AI Safety abbreviated as CAIS has appealed for various questions to be asked on AI. The rationale is that these hard questions make up the ‘final exam of humanity’, and therefore a more challenging question for AI.
Keeping AI safe
All AI labs currently considered to be ‘major’ or ‘leading’ and all major technology companies have an AI safety committee or a similar body. Many have also subscribed to new models, they require external supervision before being produced. One of the reasons that this safety chart requires is the ability to find questions and challenges that are capable of putting them through their proper tests.
A project that has benefited from OpenAI is the AI challenge.
For example, the submission form requires the user to “imagine something that you believe would confuse today’s current artificial intelligence (AI) systems.” This may then be used to evaluate how AI systems will be faring in the subsequent years.
Presently in preview and mini versions, OpenAI o1 model family needs no introduction and, with an IQ of approximately 120, it can solve PhD level problems with relative ease. Others will rise; This model is the ‘O1’ model, and the improvements are expected to come with the subsequent model called “better”. Thus, the AI security community needs difficult problems to solve.
