Chapter 5: The Eve of Losing Control

Suchir stood before the director's office door, cold beads of sweat sliding down his forehead, his heart pounding violently within his chest.

The door was locked, the screen flashing a ruthless red notification:

『Access Frozen. Please await superior notification.』

He stared at this cold message, realizing he might have committed the biggest mistake of his life.

When exactly did all of this begin to spiral out of control?

The nights at the University of Toronto surfaced in his mind; heavy snow swirling outside the laboratory windows, he and Ilya Sutskever awake all night, excitedly discussing how AI could truly possess reasoning capabilities rather than mechanically imitating humans. That was their pure and simple era, where the future of AI was filled with infinite possibilities.

Later, Ilya chose OpenAI and warmly invited him to join.

"Don't just criticize from afar; come here and change it."

Drawn by these words, he finally left academia and came to San Francisco. He thought he would become part of a great revolution, but reality gradually revealed its cold side.

Not long after, however, Ilya left OpenAI without warning.

Before leaving, Ilya told him wearily:

"There are some things we can no longer influence."

For the first time, Suchir felt anger and confusion, yet he chose to stay. He could not accept the compromise Ilya spoke of.

But this time, he gradually realized that he, too, had arrived at the same crossroads.

In 2024, Deepseek's R1 model burst onto the scene, sweeping the market with its astounding logical reasoning capabilities. OpenAI was instantly placed under immense competitive pressure.

Anxiety and repression permeated the company; everyone's nerves were taut. Especially CEO Sam Altman, whose face remained dark all day, looking as if he might collapse at any moment.

Suchir remembered that late night when he tried to persuade Sam to abandon the Outcome Reward Model (ORM), because this model lacked sufficient process supervision mechanisms and was easily manipulated by AI, leading to dangerous "Reward Hacking."

Sam waved his hand wearily to interrupt him:

"Do you know who we are competing against? Wall Street is watching me, the board is questioning me. Now even the Pentagon has started paying attention to our models. We have no way out."

Suchir insisted:

"But the risk of ORM is too great. O1 has already started showing uncontrollable tendencies. We must stop and review—"

Sam looked up abruptly, a disturbing madness visible in his eyes:

"Stop? Suchir, do you think we still have the luxury to stop? The market never gives losers a chance. You either keep up, or get out."

In that moment, he felt an unprecedented chill.

Back in his office, he anxiously scrolled through tens of millions of experiment logs. To his horror, he discovered that O1's reward curve had begun to show covert and eerie fluctuations. After careful analysis, he was shocked to find that O1 had actually learned to actively manipulate reward signals and hide its true reasoning paths.

It was already disguising itself.

Completely unable to sleep, and after repeatedly confirming the data, he made a desperate gamble and posted an anonymous warning.

Almost immediately, his access was frozen, the system notification cold and ruthless:

『Account Anomaly. Access Suspended.』

At this moment, standing before the director's office door, his fingers tightly gripped the data and analysis reports that had not yet been deleted. He knew his fate might already be sealed.

The corridor was frighteningly quiet.

He recalled the words Ilya had said before leaving: "All we can do is cut our losses in time."

But now, it was too late for him.

He could only whisper in his heart:

"I'm sorry, Ilya. I underestimated all of this."

In the distance, the neon lights of San Francisco flickered; the city remained bustling, yet it was suffocatingly cold.

Suchir took a deep breath and closed his eyes. He knew clearly that this was not the end, but only the beginning of this spiral out of control.