OpenAI co-founder and CEO Sam Altman joined the editor in a wide-ranging interview late last week, answering questions about some of his most ambitious personal investments and the future of OpenAI. rice field.
There was much to discuss. The outfit, now 8, is dominating the national conversation two months after she released her ChatGPT, a chatbot that answers questions like a person. OpenAI’s products did more than just wow users. The company is reportedly in talks to oversee the sale of existing shares to new investors at a valuation of $29 billion, despite the relatively meager earnings.
Altman declined to discuss the company’s current business dealings, and issued a few warnings when asked a related question during our roundtable.
However, he did shed some light on the company’s plans for the future. For one thing, in addition to ChatGPT and DALL-E, a popular digital art generator for costumes, Altman confirmed that video models are also coming. It may take some time.”
Altman said OpenAI’s evolving partnership with Microsoft — which first invested in OpenA in 2019 and today confirmed plans to build AI tools like ChatGPT into all of its products — isn’t an exclusive deal. clarified.
Additionally, Altman confirmed that OpenAI can build its own software products and services, in addition to licensing its technology to other companies. This is notable for industry watchers who wondered if the day would come when OpenAI would compete directly with Google via its own search engine. (When asked about this scenario, Altman said: “Whenever someone says technology is the end of other giants, it’s usually wrong. We forget that we can teach.They are pretty smart and pretty competent.”)
As for OpenAI’s plans to release a fourth version of GPT, the sophisticated language model that ChatGPT is based on, Altman said the hotly anticipated product will only “be released when we are confident it can be done.” I won’t say [release] Be safe and responsible. ” He also In contrast to OpenAI’s current deep learning models, we tried to soften expectations about GPT-4. Artificial general intelligence, or technology that has its own emergency intelligence, versus OpenAI’s current deep learning models that solve problems and identify patterns through trials, saying, “We don’t have real AGI.” means and error.
“In my opinion [AGI] GPT-4 will “disappoint” people with its expectations, he said.
Meanwhile, I was asked the following: when Altman expects artificial general intelligence to emerge, and believes it’s closer than expected, but also says the shift to “AGI” won’t be as drastic as some might expect. . “The closer [to AGI]I think it’s going to be a much more vague and gradual transition than people think, which makes it difficult to answer,” he said.
Naturally, before wrapping up, we spent some time talking about safety, including whether society has enough guardrails against the technology OpenAI has already released to the world. Many critics believe otherwise, including concerned educators who are increasingly blocking access to ChatGPT for fear that students will use it to cheat. is reluctant to release its own AI chatbot, LaMDA, reportedly concerned about its “reputational risks.”)
Here, Altman said, OpenAI “has an internal process to try to break things and study the impact. We use external auditors. We have external red teams. We work with other labs. We have asked the safety organization to investigate.”
At the same time, the technology is emerging from OpenAI and others, and people need to start thinking about how to live with it, he suggested. “There are societal changes that ChatGPT is or is about to bring about. . [product releases] now [makes sense]if the stakes are still relatively low rather than simply announcing what the industry as a whole will have in a few years without society having time to update.
In fact, educators, and perhaps parents too, need to understand that the genie should never be put back in the bottle. Altman said OpenAI and other AI organizations are “experimenting” with watermarking and other verification techniques to help assess whether students are trying to disguise AI-generated copies as their own. I did, but I also hinted at focusing too much on this particular scenario. Useless.
“There may be ways to allow teachers to detect the output of systems like GPT a bit more, but honestly, determined people are trying to get around them, and I don’t think that’s something society can do. or must be long-term dependent.”
It’s not the first time people have adapted well to big changes, he added. Altman observes that calculators have “changed what we test in math classes,” and that Google places far less emphasis on the need to memorize facts, and deep learning models are a key part of both developments. said it represented a “more extreme version”. Understandably, I’ve heard from teachers who are very nervous about the impact this will have on their homework. He also hears a lot from teachers like, “Wow, this is an incredible tutor for each child.”
For a full conversation about OpenAI and Altman’s evolving views on AI commodification, regulation, and why AI is going “in the opposite direction” many imagined 5-7 years ago. , worth checking out the clip below.
You can also hear Altman address best-case and worst-case scenarios when it comes to the possibilities and dangers of AI. short version? “The good case is so incredibly good that it sounds like a really crazy person to start talking about it,” he said. But it’s like we all turn off the lights.”
