The startup behind ChatGPT unveiled a brand new AI model with a long-coveted feature: human-like reasoning.
What happened: The new model, called o1, is trained to think through problems on its own and fact-check the answers. Unlike current AI models — which only mimic patterns from the data they are trained on — the model can use ‘critical thinking’ to work through questions.
- In the short term, the model can code faster and answer complex math questions, but in the long term, these capabilities can make AI tools less prone to hallucinations.
Why it matters: Reasoning capabilities are critical to building a more human-like AI. OpenAI’s vision is to eventually have models with strong enough reasoning capabilities to power autonomous systems, or agents, that can make decisions or take actions for us.
- For now, the o1 model is still three to four times as expensive to run compared to the most recent GPT model, it can’t browse the internet, and it takes longer to respond.
Big picture: OpenAI is rushing to drum up more money as its rivals Google and Anthropic push to develop their own advanced reasoning models. The startup is reportedly looking to raise US$5 billion in its next funding round at a staggering $150 billion valuation.—LA