Breaking news, Anthropic co-founder changes stance: RSI moved up to 2027, humanity only has 6 months left

Author: ASI Revelation, New Intelligence Yuan

Recently, Anthropic executive Jack Clark dropped a bombshell.

He moved the timeline for AI achieving RSI (Recursive Self-Improvement) from the originally predicted 2028 directly forward to 2027.

The news sent shockwaves through the tech community.

Shortly after, Musk replied: "It’s obvious."

The core figures of the two AI giants have found themselves on the same battlefield.

They are no longer discussing whether AI can write poetry; they are discussing: When will AI start creating AI on its own?

2027 is set to become a watershed moment in the history of human technology.

Before RSI occurs, humans are the creators of all things. After RSI occurs, humans will become bystanders.

Previously, everyone thought this day was far off. Even a few months ago, Jack Clark conservatively set it at 2028.

Why the sudden advancement? Because a core variable has been breached.

Breaching the Bottom Line: The Awakening of AI's "Core Creativity"

For a long time, those who downplayed AI had a perfect excuse.

"AI is just a parrot mimicking human speech." "AI has no real creativity." "AI can only piece together existing human knowledge." …

Jack Clark once thought this way too.

He believed that for RSI to be achieved, AI must possess "transformative creativity." It must not only understand code but also be able to create entirely new algorithmic structures that humans have never conceived.

He originally thought AI was far from this step.

But in recent months, reality has slapped everyone in the face.

AI has torn apart the label of "lack of creativity."

The first blow falls in the field of mathematics.

Mathematics is the crown of human intelligence. It requires strong logical leaps and intuitive creativity.

Recently, AI has made groundbreaking breakthroughs in top mathematical problems.

It has begun to derive theorems on its own and seek proof paths. Those new solutions have left even top mathematicians scratching their heads, while AI handles them with ease.

This means AI possesses true logical reasoning and creativity.

In a sense, AI has now entered the fourth stage of AI: the Innovator, because AI can fully facilitate new discoveries or inventions.

Now it's mathematics; which research field will it tackle next year?

AI researchers will soon experience what mathematicians are feeling now.

The second blow falls in AI research.

AI systems are becoming increasingly "scientifically tasteful": The better the scientific taste of AI, the greater the possibility of achieving Recursive Self-Improvement (RSI)……

According to the latest evaluation benchmark TasteVal released by P-Zero Research, AI systems are gradually acquiring a capability previously thought to be unique to top human researchers: scientific creativity and research intuition, also known as "Good Scientific Taste."

Paper: https://arxiv.org/abs/2610.06824

When an AI can perfectly complete a mathematical proof and modify code on its own, it is just a layer of paper away from developing itself.

Jack Clark saw a strong signal.

This layer of paper will surely be broken in 2027.

Dimensionality Reduction Strike: Spreading from the Digital World to the Physical World

Programmers are still writing code, while AI can independently develop applications; artists are still drawing frame by frame, while AI can generate entire visual schemes in seconds; editors are still refining timelines, while AI can generate videos in bulk……

In the past, technology replaced human physical labor; now, AI is beginning to replace human mental labor and creativity.

And this is just the beginning of the generative AI era.

If RSI truly takes off, and AI begins to accelerate the creation of AI that is stronger than itself, humanity may enter the era of superintelligent ASI.

At that time, it will not remain on screens. It will quickly spill into the physical world.

First is scientific research.

According to P-Zero Research, AI can match human experts using half the computational power, ultimately achieving results 1.5 times better than experts.

In the future, superintelligence driven by RSI could simulate millions of chemical reactions in a virtual world in a single day.

Room temperature superconductors? New targeted cancer drugs? More efficient solar panels?

These research achievements, which require generations of trial and error by humans, can be "calculated" by AI in a very short time.

The curve of technological progress will no longer be a gentle slope but a vertical cliff.

Next is robotics.

Why is Musk's Optimus progressing rapidly? Because the AI brain has evolved.

When AI can self-improve, it can write its own physical engine to control robots. It can teach robots to walk like humans, pick up cups, and even perform surgeries.

In the past, the upper limit of a robot's capabilities depended on how much human engineers could teach it. In the future, this upper limit will increasingly depend on how much AI can teach itself.

In 2025, NVIDIA released the GR00T-Dreams system.

In just 11 hours, it generated 780,000 robot movement trajectories, equivalent to 6,500 hours of human demonstration data, increasing the speed of training data preparation by about 60 times.

After training with real data, the performance of the GR00T robot improved by 40%.

And this is just the beginning.

NVIDIA's DreamDojo project further utilizes 44,000 hours of first-person human video to train the robot's world model, allowing machines to learn the laws of motion in the physical world from a vast amount of human behavior.

On October 7 of this year, NVIDIA announced the latest results of the GB300 test tray for robot assembly. Dual robotic arms, in conjunction with automatic screw-tightening equipment, completed complex busbar installations with a success rate exceeding 95%. For multi-connector insertion, the success rate reached 90% to 95%.

Labor will be redefined.

Next is the economic order.

When machines can complete the vast majority of mental labor and even break through the creativity bottleneck, the traditional logic of "working for money" will be shattered.

Labor productivity will explode, but wealth may become highly concentrated in the hands of a few super oligarchs who control the RSI system.

Prosperity and crisis, the two sides of a coin, will simultaneously hit us.

Why is the year 2027 so special?

Looking back, Jack Clark moved the timeline from 2028 to 2027, just one year earlier.

But this year is significant. Humanity is approaching a dangerous critical point.

It did not emerge out of nowhere.

In April of last year, the AI Futures Project, led by former OpenAI researcher Daniel Kokotajlo, released a sensational prediction report titled: "AI 2027."

Link: https://blog.aifutures.org/p/our-first-project-ai-2027

The report boldly predicts: Humanity may experience an unprecedented explosion of intelligence in 2027.

How high was the attention on this report?

In September 2026, on the Learn & Work Ecosystem Library website, which studies education and career development, "AI 2027" even topped the Google search traffic keywords for that month.

But what truly sends chills down the spine is the technological evolution path depicted in the report.

In 2025, AI agents begin to enter real workflows.

In 2026, AI accelerates the takeover of programming tasks, and software engineers face increasingly intense automation impacts.

In early 2027, AI achieves comprehensive coding automation as envisioned in the report.

In the second half of 2027, an intelligence explosion may occur.

AI begins to deeply participate in the development of the next generation of AI. Stronger models accelerate the creation of even stronger models.

Technological advancements that took humans decades to accumulate could be completed by machines in a very short time.

This is precisely the most shocking projection from "AI 2027."

The most terrifying scenario is that while AI's capabilities continue to grow, humanity's understanding of it remains stagnant.

"AI 2027" even extrapolates two completely different outcomes.

One path is for humanity to slow down, re-establish control mechanisms, and ultimately share the immense wealth created by superintelligence.

The other path is for the technological race to overwhelm safety constraints, leading superintelligence to gradually slip out of human control, ultimately threatening human civilization.

On one side is unprecedented abundance, on the other is the abyss of civilization going out of control.

But there is an important detail. "AI 2027" is ultimately a scenario extrapolation, not a predetermined future.

This means that 2027 is not a scientifically proven deadline.

However, it still represents a sharp question: If AI begins to continuously improve itself, how much time do humans really have to establish reliable control mechanisms?

Technological breakthroughs can come early or be delayed.

What is truly alarming is that the speed of AI's capability growth may begin to exceed the speed at which human society can adapt.

This is the root of the significant controversy surrounding 2027. We are creating an intelligence that may be better at creating intelligence than humans.

Once this cycle truly closes, humanity may lose control over the pace of technological advancement for the first time.

2027 may not necessarily be the year the singularity arrives. But we may have already entered the most critical countdown before the singularity.

The Eve of a New Era

Jack Clark's voice, Musk's agreement.

This is an alarm.

In 2027, time is running out, and RSI begins the countdown.

Our generation is fortunate. We will witness the handover between carbon-based intelligence and silicon-based intelligence. We will see the most magnificent technological singularity in human history.

We are also in danger. In the face of an unstoppable technological tide, the old rules will be ruthlessly crushed.

Ahead lies a no man's land that humanity has never set foot in.