Just now, Sam Altman slammed Ray Dalio as "anti-human", and a top-secret model was exposed on the same day.
Just now, OpenAI's two mysterious new models were leaked, kicking off the next round of battle for supremacy against Claude!
Recently, an "accidental slip" by an OpenAI employee on GitHub accidentally revealed the code name of the model — gpt-nathree.
Many analysts believe that GPT-6 Astra is coming soon.
Today, two new "cards" of Anthropic were exposed by third-party developer applications and the Discord community — claude-mashmallow-eap (Marshmallow) and claude-melon-eap (Melon).
At almost the same time, CEO Sam Altman went all out on the *Founders* podcast, taking a veiled dig at peers like Anthropic, saying that using the "countdown to destruction" for marketing is "the rhetoric of anti-human dictators".
Late-night "slip of the hand": OpenAI's next-generation mysterious agent is leaked!
The latest mysterious model exposed by OpenAI has had its code name changed from mewfour to nanthree.
On August 15, an account associated with OpenAI mentioned a mysterious code name in a public test: "Codex, gpt-mewfour".
At that time, people in the industry were still speculating: could this be a new generation of model derived from Mewtwo in Pokémon?
A few days later, in the late night of August 19, more conclusive clues emerged.
Sharmila Jesupaul, a senior OpenAI employee, submitted a public PR on GitHub, leaving a sentence: Written by agents Codex, gpt-nathree.
This record stayed on the public web for nearly 18 hours. It was not until 17:13 the next afternoon that the employee seemed to suddenly realize that they had "leaked a secret".
So they quickly went back to the backend and urgently deleted the code name "gpt-nathree".
This kind of self-defeating operation has instead made the rumors more conclusive!
What exactly is Nanthree? Could it be GPT-6 Astra?
Combined with previous revelations, Mewfour and Nanthree are very likely to be iterative checkpoints of OpenAI's next-generation Agent model.
After all, they come from the same account and adopt the same Codex signature pattern, which is unlikely to be a coincidence.
Many industry insiders believe that it looks very much like an updated GPT-6 Astra checkpoint, which is undergoing high-intensity internal code writing testing through Codex.
Obviously, OpenAI's internal model is already poised for launch.
Agent Collaboration and Autonomous Scientific Reasoning
Previously, the most prominent descriptions of Astra mostly focused on multi-agent collaboration and long-horizon task processing.
The multi-agent coordination system is designed for complex, long-running tasks.
The long-horizon task capability aims to support workflows that require continuous reasoning, experimentation and iteration, such as complex scientific research projects or advanced mathematical problems.
At present, OpenAI has not disclosed the underlying architecture details of Astra, such as parameter scale, context length, training data volume and so on.
OpenAI has publicly confirmed that Astra is a "major future model", and its internal version has solved more than 10 long-standing unsolved mathematical/theoretical computer science problems.
However, due to discussions that triggered a "critical" risk rating for cybersecurity, OpenAI subsequently strengthened isolation and access control, and some internal work was suspended.
As for claude-marshmallow-eap (Marshmallow) and claude-melon-eap (Melon), what we know so far is that neither of them has reached the Fable level.
Many people speculate that they are more likely to be iterative optimizations of Anthropic's existing Claude 5 series (especially Sonnet/Opus), rather than a brand new flagship generation.
OpenAI has already kicked off the next round of the hegemony battle against Claude in advance!
Altman's in-depth interview is packed with memorable quotes
At the same time, Altman's latest ten-thousand-word conversation on David Senra's *Founders* podcast completely exposed his ambitions, anxieties, and profound reflections on the entire AI industry.
This interview contains a huge amount of information, and many of Altman's self-analyses are very honest, even making people feel very "counterintuitive".
Altman admits his mistake: "I misjudged the timeline of disruption"
"When we launched GPT-4 in 2023, I thought the disruption of the software industry would happen immediately, and a large number of businesses would be taken away instantly. But I was wrong."
In the interview, Altman rarely admitted his misjudgment:
No matter how powerful the technology is, the industry as a whole was too optimistic about the timeline after all.
He discovered a strong resistance, which is the huge inertia of human society.
He mentioned that people are always used to doing the same things.
"Even I myself am!" Altman quipped self-deprecatingly, "I clearly already have amazing tools like Codex that could let me completely change the way I use computers, but I still spend my days copying and pasting between different information apps and mindlessly scrolling through long emails just like I did for the past 20 years."
The person who created the most powerful programming AI is still resisting it with old habits himself.
Clearly, the speed at which society and the economy adapt to AI is far slower than the speed of AI's evolution.
But is this really a bad thing?
"I am grateful for this," Altman said. It is this huge economic inertia that makes this unprecedented great transformation smoother and safer.
If society were disrupted all at once, it would be a disaster; as it is now, humanity still has time to adapt.
Going all out to slam the "double-dealing" anti-human marketing in the AI circle
Next comes the most explosive part — Altman is openly taking shots at Dario!
Although he did not name names, everyone could tell that Altman was making a veiled jab at the "doomsday marketing faction" including Anthropic.
Altman stated bluntly that the current marketing rhetoric in the AI industry is at a disaster level, and it is particularly anti-human.
"Many practitioners" are spreading fear everywhere on one hand, saying "there is a 25% chance that we will destroy the world" and "half of the jobs will disappear within a year, good luck to you"; on the other hand, they are vowing to promise that "we will bring the world the cure for cancer and great material abundance".
This kind of "benevolent dictator" narrative makes Altman extremely disgusted.
He summed up this kind of rhetoric as:
Dear ordinary people, we will bestow on you the gift of curing cancer and endless wealth.
In exchange, please give up your autonomy, shut up, and let power be concentrated in the hands of a few "unelected" companies and smart people like us, and we will make decisions for the world.
He clearly pointed out that apart from model out-of-control, the biggest risk of AI is the excessive concentration of AI power.
A sentence he emphasized repeatedly is:
"People are the whole point of this all."
Breaking the startup bible: Four and a half years of groping in the dark
As the former head of YC, Altman is well versed in Silicon Valley's "startup bible": release early, iterate quickly, and listen to customer feedback.
But OpenAI completely violated this rule.
Altman recalled the first day in early 2016 after OpenAI was founded. He, Greg Brockman and more than a dozen other top talents gathered in an apartment, and the atmosphere was just like the first day of school.
"We looked at each other, and someone asked, 'What do we do now?' Then someone went to buy a whiteboard. When the whiteboard was brought in, we looked at each other again: 'So what exactly do we do? We said we want to achieve AGI, so why don't we write a few papers first?'"
At that time, most people had no idea what they were doing at all, and their minds were completely blank.
What is even more counterintuitive is that from the founding of the company to the launch of its first real product, OpenAI groped in the "darkness" for four and a half full years.
In Silicon Valley, which is famous for its pursuit of speed, this is almost "suicidal".
When ChatGPT was first released, it grew very fast, but many people inside felt that:
"The growth is explosive, but it feels very unstable and of low value. Should we change direction? We listed 5 or 6 other things we could do."
At that time, Peter Thiel said: "It is obviously wrong to do anything other than keep working on this."
He also compared ChatGPT to the Google search box: a blank input box that can do everything correctly no matter what you enter. The new real entry point for business models that Silicon Valley has been chasing for 20 years has finally appeared.
Altman took his words and went all in.
Without customer feedback and no market signals, they could only invent various leaderboards and internal testing methods to blindly measure their progress.
After countless "chaotic stumbles", they finally figured out the Scaling Law and achieved the success of GPT.
This once again confirms Altman's philosophy: the greatest things always start from non-consensus. When the whole world thinks you are crazy and impossible to make it, that is probably where the biggest opportunity lies.
The underlying logic of the AI era is still "people"
Finally, the host asked: "If AI can even do podcasts better than us, what else do humans need?"
Altman's answer was very certain: "Humans will still crave other real human beings."
He firmly believes that no matter how superintelligent AI develops, those things related to human connection and mutual affection between people will not depreciate at all, but will become more and more expensive and valuable.
"I don't like e-books, I like physical books; I don't like Zoom meetings, I like face-to-face communication. The fundamental physical and psychological structure of human beings determines that we will always care about other people of our kind."
AI is not here to replace humans, they are just tools to spark a wave of small and micro enterprise entrepreneurship.
References:
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