OpenAI’s new o1 models push AI to PhD-level intelligence

[bjcheung77]

Wow, o1 seems so much better than all the rest... Most of the ones already widely available get the job done already...

[LevelUP]

Wow, o1 seems so much better than all the rest... Most of the ones already widely available get the job done already...

I try to remain agnostic about which LLM is best.

I believe Claude 3.5 Sonnet is the best overall for coding, mainly due to the strict limitations on o1-preview, which only allows 50 messages per week. Additionally, you can't access the API unless you're a level 5 user who spends at least $1,000 monthly on APIs.

That said, I agree that most LLMs are sufficient for basic tasks.

ChatGPT is getting most of the attention, likely because they are the ones raising significant capital. According to their last funding round, the company is now valued at $157 billion.

[NotJoeBiden]

How can an IQ test be valid if the AI model was likely trained on IQ tests questions?

You bring up a good point.

In the article, they did talk a bit about this.
"He noted that the Mensa IQ test used for this assessment was designed specifically for offline use and is not included in any AI training data."

If you really want to dig deep into the data, you can read this article from the person who ran the test on AI, which was linked in the other article above.
https://www.maximumtruth.org/p/massive-b...telligence

Also, you can view the questions from the Mensa Norway test
https://www.maximumtruth.org/p/massive-b...telligence

I'm not saying the test was perfect, but it's just one way to validate that there were some significant improvements in intelligence from the previous models.

Mensa sells study guides for their tests. IQ tests are not meant to be studied.