OpenAI CEO Sam Altman criticizes Anthropic's 'anti-human' marketing amid model leaks

Just now, OpenAI’s mysterious double strike was leaked, kicking off the next battle against Claude!

Recently, an OpenAI employee accidentally exposed the model’s codename—gpt-nathree—through a slip-up on GitHub.

Many analysts believe that GPT-6 Astra is coming soon.

Today, Anthropic has unveiled two new "cards," exposed by third-party developers and the Discord community—claude-mashmallow-eap (Marshmallow) and claude-melon-eap (Melon).

At nearly the same time, CEO Altman went on the podcast *Founders* and launched a full-scale critique, mocking competitors like Anthropic for using “countdown to destruction” as marketing—calling it “the rhetoric of anti-human dictators.”

OpenAI's latest revealed mysterious model has changed its codename from mewfour to nanthree.

On August 15, OpenAI-related accounts mentioned a mysterious code name during public testing: "Codex, gpt-mewfour".

At the time, the community was speculating whether this was a new-generation model inspired by Mewtwo from Pokémon.

On the night of August 19, a more concrete lead emerged a few days later.

Senior OpenAI employee Sharmila Jesupaul submitted a public PR on GitHub, leaving this note: Written by agents Codex and gpt-nathree.

This record remained publicly visible for nearly 18 hours. Until 5:13 p.m. the following day, the employee suddenly seemed to realize that they had "leaked" it.

Immediately returned to the backend and urgently deleted the identifier "gpt-nathree".

This attempt to cover it up only makes the rumors seem more credible!

What exactly is Nanthree? Could it be GPT-6 Astra?

Combined with previous leaks, Mewfour and Nanthree are highly likely to be checkpoint versions of OpenAI's next-generation agent model.

After all, they come from the same account and use the same Codex signature pattern—it doesn’t seem like a coincidence.

Many industry insiders believe that 0 appears to be an updated GPT-6 Astra checkpoint undergoing intensive internal code-writing tests via Codex.

Clearly, OpenAI’s internal model is already on the verge of launch.

Previously, the most prominent descriptions of Astra focused on multi-agent collaboration and long-term task processing.

Multi-Agent coordination system, designed specifically for complex, long-running tasks.

Long-term task capability, designed to support workflows requiring sustained reasoning, experimentation, and iteration, such as complex scientific research projects or advanced mathematical problems.

Currently, OpenAI has not disclosed details about Astra’s underlying architecture, such as parameter count, context length, or training data size.

OpenAI has publicly confirmed Astra as the "next major model," and its internal version has solved over ten long-standing problems in mathematics and theoretical computer science.

However, due to discussions surrounding a "critical" cybersecurity risk rating, OpenAI subsequently strengthened isolation and access controls, leading to a temporary suspension of some internal work.

Regarding Claude-Marshmallow-EAP (Marshmallow) and Claude-Melon-EAP (Melon), we know that neither has reached the Fable tier.

Many speculate that they are more likely iterative improvements to Anthropic’s existing Claude 5 series, particularly Sonnet and Opus, rather than an entirely new generation of flagship models.

OpenAI has already launched the next round of competition against Claude!

Meanwhile, Otomo’s latest ten-thousand-word conversation on David Senra’s podcast “Founders” fully reveals his ambitions and anxieties, as well as his profound reflections on the entire AI industry.

This interview is packed with insights; Altman’s many self-reflections are remarkably honest, even feeling counterintuitive at times.

Ultraman admits fault: "I misjudged the altered timeline."

When we launched GPT-4 in 2023, I thought software industry disruption would happen immediately, with vast amounts of business being instantly taken away. But I was wrong.

In the interview, Altman rarely admitted to a misjudgment:

Even with the most incredible technology, the industry as a whole remains overly optimistic on its timeline.

He encountered a powerful resistance: the immense inertia of human society.

He mentioned that people are always accustomed to doing the same things.

“Even I am!” Ultraman joked. “I already had amazing tools like Codex, which could have completely transformed how I use my computer, yet I still spend my days copying and pasting between apps and mindlessly scrolling through long emails, just like I have for the past 20 years.”

The person who created the strongest programming AI still resists it with old habits.

Clearly, society and the economy are adapting to AI much more slowly than AI is evolving.

But is this really a bad thing?

“I am deeply grateful,” Ultraman said, noting that this immense economic inertia has made this unprecedented transformation smoother and safer.

If society were suddenly overturned, it would be a disaster; but now, humanity still has time to adapt.

Full steam ahead, calling out the anti-human marketing tactics of the AI community for their hypocrisy

Next comes the most explosive part—Ultraman launches an attack on Dario!

Although he didn’t name anyone directly, everyone could tell that Ultraman was subtly mocking the “doomsday marketing” crowd, including Anthropic.

Ultraman bluntly said that the current marketing jargon in the AI industry is catastrophic and profoundly anti-human.

Many practitioners spread fear, claiming there’s a 25% chance we’ll destroy the world, that half of all jobs will vanish within a year, and wishing you good luck—while simultaneously making firm promises that we’ll deliver a cure for cancer and unprecedented material abundance to the world.

This "benevolent dictator" narrative deeply annoys Ultraman.

He summarized this phrasing as:

Dear ordinary people, we will grant you the gift of curing cancer and endless wealth.

In exchange, you must surrender your autonomy, stay silent, and concentrate power in the hands of a select few un-elected corporations and elites who will make decisions for the world.

He explicitly pointed out that, beyond model runaway, the greatest risk of AI is its excessive centralization.

The phrase he repeatedly emphasized is:

People are the whole point of this all.

Breaking the Startup Bible: Four and a Half Years in the Dark

As a former head of Y Combinator, Altman understands Silicon Valley’s “entrepreneurial bible”: launch early, iterate quickly, and listen to customer feedback.

But OpenAI completely violates this principle.

Otterman recalled the first day after OpenAI's announcement in early 2016, when he and a dozen other brilliant individuals, including Greg Brockman, gathered in an apartment, the atmosphere like the first day of school.

Everyone looked at each other, and someone asked, “What should we do now?” Then someone went to buy a whiteboard. When the whiteboard arrived, everyone looked at each other again: “So, what exactly are we supposed to do? We said we want to achieve AGI—how about we start by writing a few papers?”

At the time, most people had no idea what they were doing; their minds were a mess.

More counterintuitively, OpenAI spent a full four and a half years in the dark before releasing its first real product.

This is practically "suicide" in Silicon Valley, where speed is paramount.

When ChatGPT was first released, growth was rapid, but many inside the company felt:

Growth is strong, but it feels very unstable and has low value. Should we consider shifting direction? We’ve listed five or six other potential initiatives.

At this point, Peter Thiel said, “Doing anything other than continuing this is clearly wrong.”

He also compared ChatGPT to Google’s search box: a blank input field where you can type anything, and it gets it right. After 20 years of pursuit by Silicon Valley, a truly new entry point has finally emerged.

Ultraman listened and gave it his all.

With no customer feedback or market signals, they could only invent various rankings and internal testing methods to blindly gauge progress.

After countless chaotic setbacks, the Scaling Law was finally perfected, paving the way for GPT’s success.

This again confirms Altman’s philosophy: the greatest things always begin as non-consensus. When the whole world thinks you’re crazy and it’s impossible, that’s often when the biggest opportunity lies.

The underlying logic of the AI era is still "people".

Finally, the host asked: "If AI can do podcasts better than us, what do humans still need?"

Ultraman replied with certainty: "Humans will still long for real humans."

He firmly believes that no matter how superintelligent AI becomes, things related to human connection and mutual affection between people will not only retain their value but will become increasingly expensive and valuable.

I don’t like reading e-books; I prefer physical books. I don’t like Zoom meetings; I prefer face-to-face interaction. The fundamental physiological and psychological nature of humans ensures that we will always care about our fellow beings.

AI is not here to replace humans; it is simply a tool that sparks a wave of entrepreneurship among small and micro businesses.

Reference materials:

https://x.com/davidsenra/status/2091514583420686743

https://x.com/firesidealpha/status/2091506987137896560

https://x.com/davidsenra/status/2091575832036724852

https://x.com/firesidealpha/status/2091506987137896560

https://x.com/davidsenra/status/2091538100623151216

This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation, edited by Aeneas David.