OpenAI has introduced a new AI model called “Strawberry” or “o1” that boasts advanced reasoning abilities. The model uses a technique called “chain of thought” to solve complex problems step by step.
While anticipating GPT-5, OpenAI has released an advanced language model that thinks critically before responding, demonstrating intelligence similar to a PhD student. This improved model is a big step… https://t.co/n100odDTqY
— KDnuggets (@kdnuggets) September 16, 2024
O1 has shown impressive results in tests, ranking in the 89th percentile on competitive coding questions and among the top 500 high school students in the USA Math Olympiad.
It also performs well on PhD-level questions in various scientific fields. However, OpenAI is keeping the inner workings of o1 hidden from users.
@OpenAI's O1 series of AI models can enhance performance in complex reasoning tasks with problem-solving capabilities. Here is what you must know about it.https://t.co/keZ7dUMeRg#OpenAI #O1Series #AI #MachineLearning #ProblemSolving
— People Matters (@PeopleMatters2) September 16, 2024
When users ask the model a question, they only see a filtered interpretation of its thought process, not the raw chain of thought.
This has led to a race among hackers and researchers to uncover o1’s raw reasoning.
.@OpenAI's newest model – OpenAI o1- improves its accuracy by reasoning first before responding to a problem or prompt. While it could expand the number of legal use cases #genAI can tackle, certain barriers may make it less ubiquitous in #legaltech:https://t.co/GBtIjzhbQt
— Rhys Dipshan (@R_Dipshan) September 16, 2024
OpenAI is cracking down on these attempts, sending warning emails and threats of bans to users who try to probe the model’s inner workings. OpenAI says it wants to keep the raw chain of thought private so it can monitor the model’s reasoning for signs of manipulation.
AI is in very nascent stage like browser or search engine was in 90s incredible trillion$ opportunity.
How to prompt on OpenAI’s new o1 models https://t.co/yxEkyR6NX9— Siddarth Pai (@siddarthpaim) September 16, 2024
o1 model’s reasoning secrecy
The company also wants to maintain a competitive advantage by not exposing the model’s training data to competitors. Some experts are frustrated by this lack of transparency.
AI researcher Simon Willison says interpretability is crucial for developers working with language models. He believes OpenAI’s decision is a step backward. Others caution against comparing o1’s reasoning to human-level skills, as the way AI models and humans approach problem-solving can be fundamentally different.
Despite these concerns, o1 has the potential to become a valuable tool for researchers in various scientific fields. However, access to the model is expensive, which could limit its use to well-resourced institutions. As AI technology continues to advance, models like o1 are pushing the boundaries of what’s possible.
But striking the right balance between innovation and accessibility will be key to realizing their full potential.
