After Fable 5, the AI competition window is closing

2026/06/17 00:24
🌐en

MODELS BEGIN TO TRAIN MODELS. CALCULATOR AND SOVEREIGNTY AI WILL DETERMINE THE NEXT STAGE OF ENTRY ELIGIBILITY

After Fable 5, the AI competition window is closing
Original title: The Window Has Closed
It's the first time I've seen it
Photo by Peggy Block Beats

The editor pressed: After Fable was briefly opened and suddenly disappeared, some heavy users began to realize that changes in the AI model were no longer easily explained by running points, parameters or tools alone. The paper draws on the experience of the use of Fable to discuss a larger judgment: Frontline AI competition is moving from "Who Trains Stronger Models" to a self-acceleration phase of "Live Models Help Train Next Generation Lead Models."。

In the author ' s view, Mythos of Anthropic and its derivative models may indicate that this threshold has been crossed. The real important changes are not just the advances in product patterns such as Claude Code, Codex, but the models themselves are beginning to become part of the R & D system, promoting internal development efficiency and continuing to speed up the next round of models. Within this framework, 2023 to 2026 may be the key window for countries to enter the front line of the AI competition, and the difficulty of catching up later will increase dramatically as computing, chip exports, modelling capabilities and talent become more concentrated。

THE SHARPEST PART OF THE ARTICLE IS THE REDEFINITION OF "SOVEREIGN AI": WHETHER A COUNTRY HAS REAL AI CAPACITY IN THE FUTURE DEPENDS NOT ON WHETHER IT HAS A NATIONAL LABELED APPLICATION OR MODEL SHELL, BUT WHETHER IT HAS CONTROLLABLE ALGORITHMS, TALENT, TRAINING SYSTEMS AND FRONT-LEVEL MODELS. IF CRITICAL SYSTEMS ARE SIMPLY A LAYER PACKAGED ON TOP OF HEAD LABORATORY MODELS IN THE UNITED STATES OR CHINA, ONCE MODEL ACCESS IS CUT OFF, THE IMPACT MAY NO LONGER BE A PRODUCT THAT GOES DOWN, BUT A SYSTEMIC PARALYSIS AT THE ECONOMIC, EDUCATIONAL, MEDICAL, MILITARY AND INFRASTRUCTURE LEVELS。

This is also the central warning of this paper: AI is not a mere codification tool, nor a competition for copyright, artistic or mathematical abilities, but a common technology that is reshaping national capacity, industrial order and the infrastructure of civilization. Fable's short-lived appearance and disappearance may be a small event, but it allows more people to see in advance the true shape of the front-line AI competition: windows are closing, calculations are becoming strategic resources, and those who are not involved in building can only be observers for the next phase。

The following is the original text:

If you've used Fable during its useful life, you'll know that its specialness is not reflected in the benchmark test. I often publish benchmarking tests because many people care about them, but for a long time they have meant little to me. The only thing I care about is a benchmark that can't be measured in numbers: the shape of a model mind. How deeply it senses users, deduces intentions; how much information it can think about and evolve; how much "living" it feels. Fable is different and special at this point. It gives me a feeling back to 2023. From my timeline, I'm not the only one who has that experience. As soon as it was banned, many people reacted as if their wings had been broken。

Since last November, we've been accelerating in a steeper orbit. For some of us, it is as clear as a bell. This feeling has been increasing over the past five months. After using Fable, I am more convinced than ever that the transformation that we feel is not only the result of advances in tools such as Claude Code and Codex, but also the ability that Mythos released after completing training in early February. Mythos will give birth to many "general generations" and the increased efficiency of internal development it brings has changed the shape of the competition。

I'm not saying it's over. Nothing. Other large laboratories will be able to develop equally powerful models — some of which I believe have already been done. In the end, they'll figure out how Anthropic injects Mythos' magic and replicates it. But for some, the competition is over. The front line has now become an acceleration system: leading models will help produce next-generation lead models. Many have predicted over the years that we would eventually cross that threshold. Now that threshold has been crossed。

In 2023, Anthropic leaked a C Round-Treave book accurately predicted what was happening today. The document states: "We believe that the company that trained the strongest model in 2025/2026 will lead the follow-up cycle to the point where others cannot catch up. In many ways, they have amazing foresight, and I think we are now standing at the moment they described. If you want your country on the front line, I think the window period is only three years: from February 2023 to February 2026. That's the window. And now it's closed。

Elon grabbed this window and, through near-Hracles-style efforts, approached the front in just over two years — 26 months, to be exact. Few others with sufficient capital have tried seriously, including many nation States. European countries do worse than do nothing: They spend a great deal of time setting barriers to those who might have tried. Now, as leaders finally begin to see the contours of the future, some of these obstacles are being quickly removed. But it's too late。

Seeing the future too late, sometimes worse than not seeing it at all, because you have to experience the illusion of what could have happened. European leaders could have built a front-line model of true sovereignty. They have time and resources, and they could have recruited. But because they did not do so, they plunged themselves into a systematic dependence on others: These systems don't belong to them, they don't control them, they can't be completely censored, and they're not always allowed to use them. – These systems may even be evacuated at once because of a random decision. They have completely failed their people. In my judgement, they have been doing so for decades, but this failure may have been the most catastrophic。

Fable and Mythos' sudden disappearance woke up a lot of people. Today, there are voices calling for a sovereign AI project. But again, the boat has gone. A lot of boats have gone. The question is not just whether the Mythos level model can be obtained to start training on the next generation model. Hardware constraints will also increase. We are entering a new phase: the calculus required to reach that level of capacity will itself become a strategic resource。

In the next phase of the competition, chip exports will be rationed, narrow, politicized and increasingly seen as a tool of national security. The Biden government had previously outlined this with the "proliferation rules" that are now almost forgotten. Those rules have even capped the export of chips to allies. The nuclear weapons analogy has been misused, but soon the state-of-the-art calculus will be treated like uranium: monitored, licensed, tracked, guarded and retained in the United States. Those who control these calculations will use all necessary means to maintain the situation and to prevent other countries from building their own complete technological depots. At first, it would be political pressure, it would be preferential agreements using its model, it would be guarantees and commitments, it would be treaties, it would assure countries that they would never be cut off suddenly -- – It provides just enough to allow others to say to themselves and to their citizens, “We are part of it, and we are making a contribution.” If none of this works, they will use tougher methods. Of course, they may not have to do so at all, because the premise of these scenes is that other countries are seriously trying to join the race, and they are showing no such sign at all。

The introduction of this technology is still very early. It won't ever be like that. Imagine a future country: its economy, institutions, schools, army, hospitals and infrastructure are deeply intertwined with a proud "national" AI model. In fact, however, the model is only a thin shell placed on a lead laboratory model in the United States or China. And imagine that this model, like the Fable this week, was blocked in an instant. It'll look like an air strike. The result would be catastrophic paralysis. This is the power entrusted to others by the vast majority of States through inaction。

There will be no second chance unless there is espionage or deliberate disclosure. Open sources alleviate the problem to some extent and are significant in some respects, but open sources alone cannot save countries that lack calculus, talent and own front-level models. By 2030, it is likely that there will also be three quantitative leaps in training calculus, which will be increasingly concentrated in the United States and China. The ability to train beyond a certain threshold will only belong to the leaders; the models required for the next generation will also be in their hands. Worse still, as open source models approach Mythos-level capabilities, even beyond them -- This is likely to start later this year — they will face regulatory and global pressures. I'm not happy to say that, but it's inevitable. The capacity will reach a critical point, and the public will demand regulation even if there are no "dangerous" incidents. But it is not even necessary to wait until the public gets there, because the threshold of tolerance for those in power is much lower than for the public. They will coordinate the promotion of a global regulatory pressure and use the cutting edge, increasingly powerful closed-source laboratory model to help them do this。

Whatever you were supposed to do with AI's development schedule, even using Fable for one day, should have brought you up to date. This trajectory will become clearer throughout the year. But for most people, what it reveals is the depth of the missed opportunity and the magnitude of what they squander. Global leaders should have seen what many have seen in 2022 — what we have been talking about and writing since — the most important technology in human history. It's not about writing codes, it's not about copyright, it's not about art or mathematics. It is about the transformation of civilization, society and human beings themselves. It's a miracle. And if you have not been involved in building it, from now on you will be only one bystander。

[ Chuckles ]Original Link]

QQlink

Tidak ada "backdoor" kripto, tidak ada kompromi. Platform sosial dan keuangan terdesentralisasi berdasarkan teknologi blockchain, mengembalikan privasi dan kebebasan kepada pengguna.

© 2024 Tim R&D QQlink. Hak Cipta Dilindungi Undang-Undang.