OpenAI’s newest AI venture GPT-5 (code name: Orion), is running into big hurdles. Its rising price tag and slow progress are making people wonder if it’s doable. People are also worried that there might not be enough good data to get the super-smart AI the project wants. Orion aims to be a big step up from the tech that powers ChatGPT now.
The team has been working on it for over a year and a half. Microsoft, which has put the most money into OpenAI and works with them thought it would come out around the middle of 2024, say people who know about it. So far, OpenAI has run at least two big training sessions for Orion. Each one took months to process huge amounts of data to boost its smarts. But people on the inside say these tries have hit snags over and over.
The software just can’t reach the high marks set by the research team. Right now, Orion’s upgrades over OpenAI’s current tech are small at best. What’s more, these tiny steps forward don’t seem worth the sky-high costs. Some think that one six-month training round could eat up more than $500 million just for computer power.
People in the know hint that Orion beats OpenAI’s current tech. The key issue? These small gains don’t match the huge running costs. Guesses using public and private info on various training parts, show that one six-month training session can cost about half a billion bucks just for computer power. Experts think big tech firms might pour a whopping $1 trillion into AI projects in the years to come. This big number depends on OpenAI, the firm that kicked off the AI boom.
