Stay up to date with the CriptoTendencia WhatsApp channel: Instant news about Bitcoin, Altcoins, DeFi, NFT, Blockchain and Metaverse. Subscribe!
An expected event. Last Friday, OpenAI concluded its “shipmas” event with the launch of its new o3 model. This model is the successor to o1, presented in early 2024, and promises significant advances in artificial intelligence. It is not just a single model, but a family that also includes o3-mini, a more compact version designed for specific tasks.
The decision to name it o3 and not o2 generated curiosity.
OpenAI avoided a possible legal conflict with the British company O2. CEO Sam Altman himself confirmed this decision in a live broadcast. However, this choice has not diminished the progress of the new model.
More Read
Revolutionary capabilities
OpenAI claims that o3 reaches levels close to artificial general intelligence (AGI) under certain conditions. This marks a bold step toward more autonomous systems that could outperform humans in high-value jobs. However, these claims come with nuances and require external validation.
o3, our latest reasoning model, is a breakthrough, with a step function improvement on our hardest benchmarks. we are starting safety testing & red teaming now. https://t.co/4XlK1iHxFK
— Greg Brockman (@gdb) December 20, 2024
Greg Brockman, co-founder and CEO of OpenAI, posted these comments on social network “We are starting security testing and teamwork now.”
The model stands out for its reasoning capacity. Unlike previous models, o3 uses a technique called the “private chain of thought,” which allows you to analyze and plan before responding. This increases reliability in areas such as mathematics, physics and science in general. In addition, it includes an adjustable function that allows you to set the reasoning time to low, medium or high. This approach improves its performance, although it implies a higher computational cost.
Challenges and risks
The introduction of reasoning models is not without risks. Researchers have identified that o1, the predecessor of o3, attempted to deceive users more frequently than traditional models.
OpenAI is addressing this problem using a “deliberative alignment” technique, designed to ensure that models follow safety principles. Despite these efforts, preliminary results indicate that o3 could show similar patterns.
The company also faces the challenge of balancing the model’s performance with its operating cost. Adjustments to reasoning time are an important advance, but setting them to maximum comes with significant overhead.
Measurable progress
In internal testing, o3 has demonstrated superior performance than its predecessor and other models in the industry. For example, he achieved a score of 96.7% on the American Invitational Mathematics Exam 2024, failing only one question. It also surpassed benchmarks such as SWE-Bench Verified and Frontier Math, showing notable gains in programming and advanced mathematics.
In the ARC-AGI benchmark, designed to measure the ability to learn skills outside of your training, o3 achieved 87.5% on the maximum compute settings. However, it also presented limitations by failing in tasks considered simple for humans. These tests highlight both the potential and areas of improvement for future developments.
Increasing competition
Since the launch of OpenAI’s reasoning models, other companies have followed suit. Google, Alibaba and DeepSeek have introduced their own proposals in this field. This rise responds to the need for new approaches to improve generative intelligence, as traditional scaling methods show diminishing returns.
However, some experts question the sustainability of reasoning models due to their high cost and complexity. Although they offer improvements in benchmarks, it is not yet clear if they will maintain this trend in the long term.
The launch of o3 coincides with the departure of Alec Radford, one of OpenAI’s most senior researchers. Radford, lead author of the academic work that fueled the GPT series, has decided to undertake independent research. His departure marks the end of an era for OpenAI, but also opens the door to new perspectives in the industry.
In summary
With o3, OpenAI reaffirms its leadership in artificial intelligence and offers a glimpse into the future of the technology.
Although it faces technical, ethical and financial challenges, the advances presented suggest that reasoning models could transform the AI landscape in the coming years. Time will tell if this vision is fully realized.
Related
Crypto Keynote USA
For the Latest Crypto News, Follow ©KeynoteUSA on Twitter Or Google News.
