The author also discusses the implications of AI progress, suggesting that if current growth rates continue, AI could potentially answer all IQ questions correctly within 4-10 years. This could lead to a future where AI plays a significant role in society, potentially leading to a "post-scarcity" society. However, the author also acknowledges potential risks, including the possibility of AI wiping out humanity.
Key takeaways:
- The author conducted a verbal translation of the Norway Mensa’s 35-question matrix-style IQ test and administered it to various AI models, including ChatGPT-4, which scored an IQ estimate of 85.
- There is a consistent progression in the IQ scores of the AI models, with each new version performing better than the previous one. For instance, Claude-1 scored 64 IQ, Claude-2 scored an additional 18 IQ points, and Claude-3 scored 19 more IQ points, bringing it above the human average.
- The author predicts that if the current rate of AI progress continues, we could see AI models that score perfect IQs and surpass human intelligence in about 4 - 10 years.
- Despite the impressive performance of the AI models, the author notes that they still lag behind in vision and that their ability to "take over the world" is not a realistic concern until they can answer every question correctly without special accommodations.