OpenAI has been on the ‘wrong side of history’ concerning open source

[Pelican Press 0]

This is the hidden content, please

• Sign In
• or
• Sign Up

OpenAI has been on the ‘wrong side of history’ concerning open source

To cap off a

This is the hidden content, please

• Sign In
• or
• Sign Up

, OpenAI researchers, engineers, and executives, including OpenAI CEO Sam Altman, answered questions in a

This is the hidden content, please

• Sign In
• or
• Sign Up

on Friday.

OpenAI the company finds itself in a bit of a precarious position. It’s battling the perception that it’s ceding ground in the AI race to

This is the hidden content, please

• Sign In
• or
• Sign Up

, which OpenAI alleges might’ve stolen its IP. The ChatGPT maker has been trying to

This is the hidden content, please

• Sign In
• or
• Sign Up

and simultaneously pursue an

This is the hidden content, please

• Sign In
• or
• Sign Up

, while reportedly

This is the hidden content, please

• Sign In
• or
• Sign Up

for one of the largest financing rounds in history.

Altman admitted that DeepSeek has lessened OpenAI’s lead in AI, and he also said he believes OpenAI has been “on the wrong side of history” when it comes to open-sourcing its technologies. While OpenAI has open-sourced models in the past, the company has generally favored a proprietary, closed-source development approach.

“[I personally think we need to] figure out a different open source strategy,” Altman said. “Not everyone at OpenAI shares this view, and it’s also not our current highest priority […] We will produce better models [going forward], but we will maintain less of a lead than we did in previous years.”

In a follow-up reply, Kevin Weil, OpenAI’s chief product officer, said that OpenAI is considering open-sourcing older models that aren’t state-of-the-art anymore. “We’ll definitely think about doing more of this,” he said, without going into greater detail.

Beyond prompting OpenAI to reconsider its release philosophy, Altman said that DeepSeek has pushed the company to potentially reveal more about how its so-called reasoning models, like the o3-mini model released today, show their “thought process.” Currently, OpenAI’s models conceal their reasoning, a strategy intended to prevent competitors from scraping training data for their own models. In contrast, DeepSeek’s reasoning model, R1, shows its full chain of thought.

“We’re working on showing a bunch more than we show today — [showing the model thought process] will be very very soon,” Weil added. “TBD on all — showing all chain of thought leads to competitive distillation, but we also know people (at least power users) want it, so we’ll find the right way to balance it.”

Altman and Weil attempted to dispel rumors that ChatGPT, the chatbot app through which OpenAI launches many of its models, would increase in price. Altman said that he’d like to make ChatGPT “cheaper” over time, if feasible.

Story Continues

Altman previously said that OpenAI was losing money on its priciest ChatGPT plan, ChatGPT Pro, which costs $200 per month.

In a somewhat related thread, Weil said that OpenAI continues to see evidence that more compute power leads to “better” and more performant models. That’s in large part what’s necessitating projects such as Stargate, OpenAI’s

This is the hidden content, please

• Sign In
• or
• Sign Up

massive data center project, Weil said. Serving a growing user base is fueling compute demand within OpenAI, as well, he continued.

Asked about recursive self-improvement that might be enabled by these powerful models, Altman said he thinks a “fast takeoff” is more plausible than he once believed. Recursive self-improvement is a process where an AI system could improve its own intelligence and capabilities without human input.

Of course, it’s worth noting that Altman is notorious for overpromising. It wasn’t long ago that he

This is the hidden content, please

• Sign In
• or
• Sign Up

.

One

This is the hidden content, please

• Sign In
• or
• Sign Up

user asked whether OpenAI’s models, self-improving or not, would be used to develop destructive weapons — specifically nuclear weapons. This week, OpenAI announced a partnership with the U.S. government to give its models to the U.S. National Laboratories in part for nuclear defense research.

Weil said he trusted the U.S. government.

“I’ve gotten to know these scientists and they are AI experts in addition to world class researchers,” he said. “They understand the power and the limits of the models, and I don’t think there’s any chance they just YOLO some model output into a nuclear calculation. They’re smart and evidence-based and they do a lot of experimentation and data work to validate all their work.”

The OpenAI team was asked several questions of a more technical nature, like when OpenAI’s next reasoning model, o3, will be released (“more than a few weeks, less than a few months,” Altman said), when the company’s next flagship “non-reasoning” model, GPT-5, might land (“don’t have a timeline yet,” said Altman), and when OpenAI might unveil a successor to DALL-E 3, the company’s image-generating model.DALL-E 3, which was released around two years ago, has gotten rather long in the tooth. Image generation tech has improved by leaps and bounds since DALL-E 3’s debut, and the model is

This is the hidden content, please

• Sign In
• or
• Sign Up

.

“Yes! We’re working on it,” Weil said of a DALL-E 3 follow-up. “And I think it’s going to be worth the wait.”

This is the hidden content, please

• Sign In
• or
• Sign Up

#OpenAI #wrong #side #history #open #source

This is the hidden content, please

• Sign In
• or
• Sign Up

This is the hidden content, please

• Sign In
• or
• Sign Up