OpenAI, the ChatGPT maker, plans to receive additional financial support from Microsoft. Sam Altman, its chief executive officer and founder of OpenAI is pushing forward with his vision for artificial general intelligence – computer software that is as intelligent as humans.
Altman stated in an interview that his company’s relationship with Microsoft CEO Satya Nadella “works really well”. He also said he expects to “raise a lot more money over time” both from Microsoft and other investors. This is necessary for him to meet the high costs associated with building more advanced AI models.
Microsoft invested $10bn earlier this year in OpenAI, as part of an “multiyear” deal that valued the San Francisco based company at $29bn.
Altman responded, “I hope so.” Altman added, “There is a long road ahead, and there’s lots of computing to be done between now and AGI.” . . Training expenses are huge.
Altman stated that “revenue growth was good this year” without giving financial details and that the company remains unprofitable because of training costs. He said that the Microsoft partnership will ensure “that both of us make money from each other’s successes, and everyone is happy”.
OpenAI announced new tools and improvements to its GPT-4 model for developers and businesses at an event attended by Nadella on 6 November.
These tools include a Store or marketplace for the best apps, as well as custom versions of ChatGPT, which can be tailored and adapted to specific applications. In a similar business model to Apple’s App store, the ultimate goal will be to share revenues with popular GPT creators.
Right now, people say, “you have this lab, this API [software], the partnership with Microsoft and this ChatGPT thing. Now there’s a GPT shop.” Altman explained that these aren’t our real products. “These are channels for our single product which is magic intelligence in the skies, intelligence. “I think that’s exactly what we are about.”
Altman hired Brad Lightcap as his chief operational officer to build up the enterprise business. Lightcap previously worked for Dropbox and startup accelerator Y Combinator.
Altman splits his attention between two main areas: “How to build superintelligence”, and how to increase computing power. “The vision is AGI. Figure out how to keep it safe. . . “.
OpenAI is working on building more autonomous agents, which can do tasks and actions such as sending emails, executing code, or making payments.
“We’ll make these agents even more powerful.” . . “The actions will become more complex,” he added. “I think the amount of value to business that will be gained by being able do this in every category is pretty good.”
Altman also said that the company was working on GPT-5 – its next generation AI model – although he didn’t commit to a release date.
Altman explained that more data will be needed to train, and this data would come both from publicly available data on the Internet, as well proprietary data from businesses.
OpenAI has recently issued a request for large data sets, particularly from organisations that are “not easily accessible to the public online today”. This is especially true for long-form articles or conversations.
Altman stated that it is technically difficult to determine exactly what capabilities and skills GPT-5 will have.
He said, “Until we train the model, it is like a guessing game.” “We are trying to improve because it is important to me from a safety standpoint to be able to predict capabilities.” But I cannot tell you exactly what GPT-4 will do.
OpenAI uses Nvidia H100 chips to train its models. These chips have become Silicon Valley’s hottest commodity in the last year, as rival tech firms raced to secure these crucial semiconductors required to build AI systems.
Altman stated that there has been a “brutal crunch” due to the shortage of Nvidia chips, which cost $40,000 each. Altman said that his company has received H100s and is expecting more in the near future. He added that “next years looks like it will be better”.
The dependence on Nvidia will not last long as Google, Microsoft and AMD are preparing to release AI chips that compete with Nvidia’s. “I believe the magic of capitalism has done its work here. Altman added that many people now would love to be Nvidia.
OpenAI took the lead early in the race for generative AI, systems that create text, images and code in seconds, with the release ChatGPT, almost a full year ago.
Altman stated that despite its success with consumers, OpenAI aims to progress in the development of artificial general intelligence. ChatGPT’s large language models are “one of its core pieces”. . . “This will be the foundation for building AGI. But there will be many other pieces on it.”
OpenAI’s competitors, on the other hand, have pursued alternative research strategies in order to advance AI.
Altman stated that his team believes that language is a “great method to compress information”, and that this leads to developing intelligence. He believed that Google DeepMind, for example, had missed the point.
“[Other firms] have many smart people. They didn’t do it. “They didn’t do it, even though I thought that we had proven it in some way with GPT-3,” said he.
Altman concluded that “the most important missing piece” is the understanding required by such systems in order to achieve fundamental leaps.
“For a long time, [Isaac] Newton should have read more math books and talked to professors. He also should have practiced problems. . . Altman cited an example from a previous colleague to illustrate what his current models are capable of.
He added that Newton would never have invented calculus just by reading about algebra or geometry. Altman added, “And neither are the models we use.”
“And the question is: what is the missing concept to generate net new? . . Knowledge for Humanity? “I think that’s what we should be working on the most.”