OpenAI has launched its new generative AI model, o1, which is capable of more advanced thinking and solving complex questions, particularly in the field of mathematics. The creators of ChatGPT hope that this model will reduce the occurrences of "hallucinations" that sometimes affect these kinds of programs.
In an online statement, OpenAI confirmed that o1 thinks before giving an answer. This launch marks a significant step towards the company's goal of developing "general artificial intelligence," which aims to replicate human-like intelligence.
OpenAI Starts a New Funding Round Valued at $150 Billion
OpenAI's CEO, Sam Altman, congratulated his team via the platform X on this "new model: an AI capable of complex, general thinking." However, he admitted that the technology "is still not perfect, still limited, and can seem more impressive at first than it does after prolonged use."
The company made the beta version of o1 available starting Thursday, initially accessible to paying users of ChatGPT.
When tested by AFP, the o1 model answered basic logic questions similarly to GPT-4 but took longer and provided more detailed explanations rather than generating responses almost immediately.
One key difference is that the new model currently only handles text-based content and cannot yet process or create other forms of media.
Investments and Valuation
The launch of o1 comes as OpenAI seeks to raise funds, potentially pushing its valuation to around $150 billion, making it one of the most valuable privately-held companies in the world, according to U.S. media reports.
Investors include Microsoft and Nvidia, the chip manufacturing giant. Other potential investors include Apple, which already integrates OpenAI’s technology into its generative AI system, private equity firm Thrive Capital, and MGX, a tech investment firm founded by the UAE's Artificial Intelligence and Advanced Technology Council.
The introduction of OpenAI's ChatGPT in late 2022 sparked a wave of innovation in Silicon Valley focused on generative AI, which can create content based on simple, everyday language prompts.
Since then, major tech companies like Microsoft, Google, and Meta (which owns Facebook and Instagram) have been competing to provide AI tools designed to assist humans with tasks ranging from writing emails to teaching and creative expression.
However, these AI tools are essentially machines that predict word sequences or pixel arrangements without actual understanding. This can sometimes lead to models generating inaccurate or invented information, commonly referred to as hallucinations, and even struggling with basic math problems.
Improving AI Capabilities
OpenAI has tested its new model in solving mathematical problems and generating lines of code. The company asserts that "o1 rivals human experts' performance in several benchmark tests requiring strong reasoning skills."
The company also revealed that o1 ranked among the top 500 students in a U.S. high school math competition.
OpenAI explained that "o1 uses a chain of thoughts, just like a human who may spend a lot of time thinking before answering a difficult question."
Additionally, the model is designed to learn from its mistakes and correct them, break complex problems into simpler steps, and try different approaches when the initial method fails.
While the new model hallucinates less than its predecessor, OpenAI researcher Jerry Towrik told tech website The Verge that hallucinations have not been fully resolved.
Another ongoing issue is that AI models remain "black boxes" to users, meaning the process behind their decisions is often unclear.
OpenAI emphasized that the new model is an improvement in terms of safety and alignment with human values, as its thinking process is now readable and it applies safety measures more effectively.
Post a Comment
0Comments