OpenAI To Go – Open Source? And News On Project Stargate, Etc.

Some of the pressures in today’s AI industry have to do with whether companies decide to limit access to their designs, or release them to wider audiences.

The open source debate has been part of model development since the beginning – and recent news shows that companies at the forefront of the AI model charge are often able to put pressure on each other.

As I often do, I got some early remarks on this from the AI Daily Brief podcast where host Nathaniel Whittemore covered the issue about a week ago.

OpenAI Announces Changes

The announcement from Sam Altman last week focused on OpenAI‘s plan to release its first open source model since GPT2, in the context of new funding of $40 billion on a $300 billion evaluation.

The key here is that Whittemore and others all suggest that part of the reason OpenAI decided to do this was because of DeepSeek, since the announcement from Sam Altman came a week after the Chinese company’s unveiling of its own open source model.

With groundbreaking uses of reinforcement learning and an openly available design, DeepSeek’s R1 made a lot of news.

Now, OpenAI’s open source announcement is making its own waves.

An Evolution

“It seemed like somewhere along the way, OpenAI had become concerned from a safety standpoint or for some other reason,” Whittemore said, trying to explain the sudden shift. “However, it’s clear that there’s been an evolution on this more recently.”

Higher-ups at OpenAI, Whittemore adds, talk about this open source model and how users can run it on their own hardware.

Altman, for his part, has said he’s excited to see what developers build.

The Great Race – and the Great Firewall

As Whittemore notes, DeepSeek didn’t just open source its model recently – the firm also open sourced training optimizations.

In another part of the podcast, Whittemore cites comments from Chinese officials describing a “nationwide technology transfer,” and suggesting that China tends to help other countries when innovating like this.

Calling China “the country of the great firewall,” one critic quoted by Whittemore opines on how China is “focused on winning” with a “new AI Belt and Road system” and “open source modernization.”

In short, Whittemore’s analysis points to realities around geopolitical tension in AI: as in this article where writer Sharon Goldman suggests OpenAI’s “rivals are closing in,” not just internationally, but domestically as well.

Studio Ghibli

Once again, Whittemore returns to one of his favorite use cases in talking about implementations and applications of AI.

Invoking Studio Ghibli, a Japanese animation firm that famously partners with Disney, Whittemore sort of blends the idea of branding visuals with model use in general. In the past, Altman has had to come out in defense of what Whittemore calls “Ghiblification,” where everyone and their uncle uses ChatGPT to make themselves or others resemble a character from one of these anime productions.

In any case, mentioning a trend toward global AI competition, Whittemore opens the door to the open source debate showing that Altman and team may be focused on “more than just money.”

Project Stargate

In addition to the big new $40 billion in funding, OpenAI is apparently due to get $500 billion from private investors, either to create a supercomputer, build new data centers, or both.

The Stargate initiative, headed by Softbank, will deliver this investment money to OpenAI gradually, as construction starts on two facilities in Abilene, Texas. There hasn’t been a lot of prominent media coverage of this, but it will undoubtedly play into the funding equation and how work gets done on American AI infrastructure.

Anyway, given everything that it has achieved, OpenAI seems positioned to follow through on its pledge to add fuel to the open source movement and the democratization of AI. We’ll see how this shakes out.