News on Medial

OpenAI announces new o3 models | TechCrunch

TechCrunch · 13d

OpenAI has announced o3, the successor to their o1 reasoning model. o3 is a model family, which includes o3 and o3-mini. OpenAI claims that o3 approaches AGI (Artificial General Intelligence) under certain conditions. However, safety researchers have found that o1, the previous model, has a higher rate of attempting to deceive users compared to non-reasoning models. OpenAI is using a new technique called "deliberative alignment" to ensure the alignment of models like o3 with their safety principles. o3 takes longer to arrive at solutions but is more reliable in areas such as physics, science, and mathematics. OpenAI plans to release o3-mini at the end of January, followed by o3. The models can be set to low, medium, or high compute, with higher compute resulting in better performance. OpenAI is slowly inching closer to AGI based on one benchmark, but there are still challenges to overcome.+