In a recent statement, Sam Altman suggested something extraordinary:

“Superintelligence probably by the end of 2028.”

He further claimed that:

“More of the world’s intellectual capacity could reside inside of data centers than outside of them.”

And added a provocative comparison:

a superintelligence could outperform CEOs, researchers, and top executives across domains.

This is not just a technical prediction.

It is a framing of a near-future world order.

But before accepting or rejecting it, we need to clarify something most public discussions blur intentionally:

What is AGI, what is ASI, and what is actually meant by “superintelligence”?

1. AGI, ASI, and Superintelligence are NOT the same thing

These terms are often used interchangeably in public discourse.

Technically, they describe very different levels of capability.

AGI (Artificial General Intelligence)

AGI refers to a system that can:

• perform most economically useful tasks at human level
• transfer knowledge across domains
• generalize beyond narrow training distributions

But AGI does NOT require:

• superhuman performance
• autonomous goal formation
• recursive self-improvement

AGI is essentially:

human-level general cognition in machine form

ASI (Artificial Superintelligence)

ASI goes beyond human level.

It implies systems that are:

• consistently better than humans in almost all cognitive tasks
• capable of scientific discovery at accelerated rates
• strategically superior in planning and reasoning

ASI is not just general intelligence.

It is:

dominance over human cognitive capability

Superintelligence (Broader philosophical term)

Nick Bostrom’s definition is the most cited:

“Any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest.”

This is even stronger than ASI in emphasis on scale and dominance.

Key distinction:

• AGI = parity
• ASI = superiority
• Superintelligence = systemic cognitive dominance

So when Altman says “superintelligence by 2028,” he is not describing AGI.

He is describing:

a full structural inversion of human intellectual primacy

2. Where are we actually today?

Despite the rhetoric, the current technical state is much more grounded.

Modern frontier models:

• are still statistical next-token systems
• rely heavily on scaling laws (Kaplan / Chinchilla paradigm)
• show strong capabilities in language, coding, and reasoning benchmarks
• but still struggle with persistent agency, long-horizon planning, and grounded verification

Even advanced “agentic systems” today are:

brittle orchestration layers on top of probabilistic models

Not autonomous intelligence.

Not self-directed systems.

Not recursive superintelligence.

We are still in what could be called:

early tool-augmented general intelligence systems

Not AGI in the strict philosophical sense.

Let alone ASI.

3. The Altman narrative: acceleration as inevitability framing

Sam Altman’s public positioning has been consistent over time:

• AI will be transformative
• governance is necessary
• deployment must be controlled

He even testified before the U.S. Congress advocating for AI regulation.

But there is a structural tension:

the same narrative that calls for regulation also accelerates global AI competition

This creates a feedback loop:

• regulatory urgency increases attention
• attention increases funding
• funding accelerates model scaling
• scaling increases perceived inevitability

This is not unique to Altman.

It is a classic pattern:

existential framing → capital acceleration → system-wide hype amplification

In that sense, Altman is not just predicting superintelligence.

He is participating in the construction of its perceived inevitability.

4. Dario Amodei: safety, open models, and controlled intelligence

Dario Amodei (Anthropic) has been one of the most vocal advocates of:

• frontier model safety
• controlled release of model capabilities
• restrictions on open-weight distribution

In 2023, Anthropic publicly emphasized risks of open-source frontier models being misused.

This aligns with a broader argument:

once models reach a certain capability threshold, unrestricted release becomes a systemic risk

Recently, reports have indicated tighter regulatory scrutiny on advanced model distribution and export controls.

At the same time, companies like Anthropic have adjusted deployment policies of frontier models such as internal safety versions (e.g., experimental models with restricted release pathways).

Whether this is voluntary alignment or regulatory pressure is less important than the pattern:

advanced models are increasingly treated as controlled infrastructure, not open software

This marks a shift:

from

software distribution

to

capability governance

5. Ilya Sutskever and the silence of SSI

Ilya Sutskever represents a different intellectual trajectory.

As OpenAI’s former chief scientist, he was one of the key figures behind early scaling breakthroughs.

After leaving OpenAI, he founded Safe Superintelligence Inc. (SSI).

And SSI has been notable for one thing:

almost complete absence of public product output or model releases

This silence is often misinterpreted as inactivity.

But structurally, it suggests a different hypothesis:

intelligence itself is the product boundary, not applications built on top of it

If SSI’s thesis is taken seriously, then:

• releasing intermediate systems is not the goal
• incremental deployment may be viewed as unsafe
• intelligence should not be “productized” prematurely

This leads to a radical implication:

the endpoint of AI development may not be distribution—but containment

6. The hidden convergence: regulation is already here

A key misconception in public discourse is that:

AI regulation is something that will happen in the future

In reality, it is already happening:

• model export controls
• restricted deployment of frontier systems
• internal safety gating
• government involvement in evaluation frameworks
• closed release of high-capability models

Even the distinction between:

• open models
• restricted models
• internal models

is becoming a governance layer.

As of now, frontier AI is no longer just an engineering problem.

It is:

a controlled release ecosystem under emerging state oversight

7. The hype cycle problem: everyone is both right and wrong

This is where the narrative becomes uncomfortable.

• Altman pushes inevitability of superintelligence
• Dario pushes risk and containment
• Ilya builds in silence, implying caution at the deepest level

All three narratives are internally coherent.

But they also reinforce the same global dynamic:

AI progress is being framed as both imminent and existential

This dual framing produces:

• investment acceleration
• regulatory expansion
• geopolitical competition
• public expectation inflation

This is not just technological evolution.

It is a coordination system for global belief about intelligence.

8. So is superintelligence by 2028 realistic?

From a strict technical standpoint:

There is currently no evidence of:

• autonomous recursive self-improvement systems
• stable long-horizon agency at scale
• fully reliable world-model grounded reasoning
• general-purpose autonomous scientific discovery loops

What we do have:

• rapidly improving foundation models
• emerging agent frameworks
• increasing tool integration
• early multi-step reasoning systems

This is powerful.

But it is not yet:

an intelligence explosion system

So 2028 predictions should be read less as forecasts and more as:

strategic narrative positioning inside a high-capital, high-regulation race

Final synthesis

The real story is not whether superintelligence arrives in 2028.

The real story is:

AI is becoming a regulated, narrativized, and politically structured technology long before it becomes autonomous intelligence.

We are not watching the arrival of superintelligence.

We are watching the formation of:

the governance system that decides what intelligence is allowed to become.