Sam Altman 稱對 Jensen Huang 有個人同理心,但拒絕評論他應否重掌 Nvidia
'Heroes have flaws': Sam Altman says he has personal empathy for Jensen Huang
Sam Altman 表示,他對 Jensen Huang 有「強烈的個人同理心」,並稱 Huang 一直是他的英雄,但拒絕回答 Huang 是否應重掌 Nvidia。Altman 説,OpenAI「總有一天」會上市,並稱目前不擔心投資資金不足;他亦表示 OpenAI 現時不能安全地推進遞歸自我改進。
OpenAI CEO Sam Altman said Monday he has “personal strong empathy” for former Nvidia CEO Jensen Huang, though he declined to say whether Huang should return to the chip giant.
Altman — who about three years ago was ousted from his own post, only to return shortly thereafter — said at Semafor’s Silicon Valley & The World that Huang has “long been a hero of mine. Heroes have flaws.”
Huang is reportedly considering trying to make a comeback as CEO, and some tech leaders are pushing for him to return. Asked whether Huang should return, Altman told Semafor’s Reed Albergotti: “I don’t think I know enough to answer that question.”
Altman said it’s hard to imagine Nvidia without Huang at the helm, but Nvidia is “a behemoth of a company,” and people still trust its guarantees as an investor.
In an expansive 30-minute interview, Altman struck an honest tone about the challenges he and the rest of the AI industry are facing, including the Nvidia fallout, the backlash against data centers, and increasing pressures on the cost of capital to fund the AI buildout.
On public sentiment toward AI
Following the US midterm elections that were broadly seen as a rebuke of AI, Altman said the industry should take a “strong stance” on how it approaches the long-term vision of AI.
He said much of the backlash to data centers is about people latching onto a physical thing to fight back against, but that the real issue stems from people wanting to stay “in control of the future.”
“We should not want to have a ‘handover’ … either gradual or sudden, to AI making decisions about our future,” he said. The public should understand, he argued, that the people building AI are “team human.”
He thinks data centers will still get built, but added that “the engagement with communities is not working for people. And the ecosystem will adapt.”
Public sentiment is likely to change over time, Altman said, recalling a period where communities wanted data centers in their areas because they created jobs and industry in underinvested areas. Now, perhaps, developers have to listen to the pushback and build them further away from where people live. “We have a giant country, we have a lot of space we can do this.”
On going public
Some investors have grown concerned with the massive amounts of capital required for the AI infrastructure buildout, and the buildup of debt. With an OpenAI IPO still looming, Altman said he wasn’t concerned.
OpenAI “will go public at some point,” even if there are “real downsides to being public,” including intense volatility.
Altman said he isn’t concerned with there being enough dollars to go around, even if Anthropic is the first to the IPO party later this week.
“That is not a worry,” he said. “There’s been way too much investor demand and not enough supply.”
On espionage
OpenAI recently faced its own spy drama when an employee put a recording device in a company conference room. It’s clear that “people are looking for an edge in all sorts of ways,” Altman said, adding that the company’s operational security has not been perfect, but it has had time to address it. “We have been living in a reality that everybody else is now very publicly and suddenly coming to grips with.”
On memory and RAM
Altman clarified his recent comments suggesting that it’s a bad thing for people to have RAM memory units sitting on their personal devices at home, amid a global memory supply shortage.
His comment, he said, got “misinterpreted,” and that he wants people to have RAM at home: “I think the home tinkering is great.”
Only in an optimal world, he said, would personal tinkerers be able to “share this memory” with people who need it.
On the next generation of tech
Altman said the world is “underpricing or underweighting what can happen with the next generation of products.”
“We had the chatbots, then we had the coding agents. I think in the last few months we started to see the beginning of AI for science,” he said. “I think we’re going to see what a lot of compute can do to a lot of the kind of rote work inside of an organization.”
OpenAI on Monday announced a new program to boost scientists’ access to frontier AI.
“The world can get so much better,” Altman said.
On AI safety
Altman said OpenAI could push more forcefully toward recursive self-improvement — the idea that an AI model can train its own successor to improve — but added, “I don’t think we could safely do that right now.” Still, he said, the company is able to develop new models faster than it would without AI.
OpenAI is now working with embedded evaluators from AI safety watchdog METR: The principle of getting outside experts to evaluate AI labs’ safety work “is going to be increasingly important and valuable,” he said.
來源:Semafor · semafor.com