
El informático británico Simon Willison preguntó: "¿OpenAI ha compartido algún ejemplo concreto de mensajes que fallan en o1 normal pero que tienen éxito en o1-pro?" Al contrario, ya se han documentado fallos. El Sudoku es un desafío para el modo profesional o1 y se ve obstaculizado por una comedia de ilusión óptica que es fácilmente evidente para cualquier ser humano. Un ingeniero de inteligencia artificial, Tibor Blaho, escribió en X : "O1 y o1-pro fallaron aquí, probablemente debido a las limitaciones de visión (lo mismo con los Sudoku)". Tibor Blaho también señaló que tanto o1 como o1-pro fallaron en términos de capacidades visuales. o1 y o1-pro fallaron aquí, probablemente debido a las limitaciones de visión (lo mismo con los Sudoku) https://t.co/mAVK7WxBrq pic.twitter.com/O9boSv7ZGt
o1-pro code failed to understand/read the puzzle properly.
According to OpenAI’s internal benchmarks, the standard o1 performs only marginally better than the o1 pro mode on coding and math problems.
In order to demonstrate the consistency of the o1 pro mode, OpenAI conducted a more stringent evaluation of the same benchmarks. The model was only deemed to have solved a question if it correctly identified the answer four out of four times. However, the enhancements were not substantial in these assessments.
On Thursday, OpenAI CEO Sam Altman was compelled to reiterate that ChatGPT Pro is not suitable for the majority of individuals, despite his previous assertion that OpenAI was progressing toward intelligence that was “too cheap to meter.”
He said, “Most users will be very happy with the o1 in the [ChatGPT] Plus tier!”
Bob McGrew said that improved reliability may be the most important unlock for o1 pro, especially for the medical field.
Ameet Talwalkar, an associate professor of machine learning at Carnegie Mellon and a venture partner at Amplify Partners, said “It seems like a big risk to me to raise the price tenfold,” he added, “I think we’ll have a much better sense in just a few weeks as to the appetite for this functionality.”
Guy Van den Broeck, a computer scientist at UCLA, said, “I don’t know if the price point makes sense and if pricey reasoning models will be the norm.”
Prospective consumers are not provided with a significant amount of information when the o1 pro mode is described as the most effective at resolving “the hardest problems.”
It is difficult to rationalize paying more in the absence of specific examples of the model’s purportedly enhanced capabilities, which include the ability to “think longer” and exhibit “intelligence.”
From Zero to Web3 Pro: Your 90-Day Career Launch Plan