Mar 08, 2024 06:00:00

Report that Claude-3, who appeals to exceed GPT-4, achieved IQ over 100 for the first time in AI

Chatbot AI based on large-scale language models (LLM) that can have natural conversations like humans, such as ChatGPT, Claude, and Gemini, have appeared. Maxim Lott, a reporter and television producer, measured the estimated IQ of various chatbot AIs by having them solve human intelligence quotient (IQ) tests, and found that

Claude 3 , an LLM at Anthropic, had an estimated IQ that was higher than that of a human. It was reported that it exceeded the standard value of 100.

AIs ranked by IQ; AI passes 100 IQ for first time, with release of Claude-3
https://www.maximumtruth.org/p/ais-ranked-by-iq-ai-passes-100-iq

Mr. Lott had various LLMs take the Norwegian Mensa IQ test twice and estimated their IQ from the average number of correct answers. The image below summarizes the results. The person with the best IQ test score was Claude 3, released by Anthropic in March 2024, with an estimated IQ of 101. Generally, the standard value for IQ is 100, so Claude 3 is about the same as the average human. Also, Anthropic appealed that ``Claude 3 outperforms GPT-4'', but at least according to the IQ test results conducted by Mr. Lott, Claude 3 outperforms ChatGPT-4.

Claude 3's predecessors, Claude 1 and Claude 2, had estimated IQs of 64 and 82, respectively. Since Claude 1 was released

in March 2023 , Claude 2 was released in July 2023 , and Claude 3 was released in March 2024, Mr. Lott said, ``Within a few years, an estimated IQ of around 140. I predict that this will lead to the birth of new AI.

Looking at the other rankings, ChatGPT-4's estimated IQ is 85, ranking it in second place. On the other hand, ChatGPT-3.5 was 64. In addition, Microsoft's Bing Copilot has an estimated IQ of 79 points, and Lott said, ``Microsoft uses technology from OpenAI, which developed GPT-4, so it may not be surprising that the score is close to ChatGPT-4. No,” he commented.

Mr. Lott also said that he was surprised that Bing Copilot used ``ASCII art'' to present visual answers to the questions he entered.

Gemini, Google's multimodal AI, had an estimated IQ of 77.5, and its higher-end version, Gemini Advanced, had an estimated IQ of 76. The higher version has a lower estimated IQ, but the reason is unknown.

The estimated IQ of Grok , a chatbot AI developed by xAI, an AI development company founded by Elon Musk, is 68.5, but Grok Fun (fun mode), where you can expect more exciting answers, was slightly lower at 64. . And Meta's open source LLM, Llama-2, has an estimated IQ of 67.

'AI may have some intelligence, not just a large database,' Lott said. 'Furthermore, IQ tests can be used to gauge the rate at which AI is progressing. If things continue like this, the world will be very different in a few years, but whether we need to worry about AI ``taking over the world'' is probably not a realistic concern.'' I spoke.

Please note that this estimated IQ is only derived by Mr. Lott from the results of the IQ test, and just because Claude 3's estimated IQ exceeds 100 does not mean that it exceeds human intelligence. Caution is required.