Radio/Music News Tues-Thurs-Fri
OpenAI Introduces ‘o1 Model’ Aimed At Complex Problem
Solving
OPENAI recently released its new AI reasoning model, o1, on 9/12. The model aims to solve complex problems and includes a smaller, more affordable version called o1-mini. ChatGPT Plus users can access both models now, while ENTERPRISE and EDU users will receive access soon.
The o1 model, codenamed “STRAWBERRY,” represents a step toward human-like artificial intelligence. It uses a “chain of thought” method that focuses on reasoning rather than pattern recognition. According to OPENAI, o1 performs better in coding and complex mathematical problems than previous models like GPT-4o.
OPENAI Chief Research Officer BOB MCGREW said, “In tests, o1 scored 83 percent in a qualifying exam for the International Mathematics Olympiad, compared to GPT-4o’s 13 percent. The model excels in science, coding, and other challenging benchmark tasks.”
OPENAI Research Lead JERRY TWOREK added, “o1 has been trained using a new optimization algorithm and a tailored dataset. The model employs reinforcement learning to solve problems, which involves a system of rewards and penalties. While it is more accurate and hallucinates less, we can’t say we solved hallucinations. The issue still persists.”
In simple terms, “hallucinations” in AI means that AI made something up. It can give answers that sound right but are wrong or made up because it doesn’t really understand what it’s talking about. It’s just guessing based on patterns it learned from the data it was trained on.
OPENAI also introduced o1-mini as a faster and more cost-effective option for tasks that require reasoning but not extensive world knowledge. At launch, ChatGPT Plus users can select both o1-preview and o1-mini through the model picker. Weekly message limits are set at 30 for o1-preview and 50 for o1-mini.