r18novel Chapter 78 has nothing to do with materials or AI.
Chapter 78 has nothing to do with materials or AI.
Wang Lei's fingers hovered over the keyboard for two seconds before he continued typing.
He wasn't sure he fully understood. But he was certain of one thing—every time Lin Yu uttered those philosophical-sounding words, they eventually turned into a running program.
On the first day after the adversarial self-testing module went online, the output of Tianyan plummeted.
Judging from the more than 200 messages per hour, the number has dropped to less than 50.
The reason is simple. Every time it reaches a conclusion, it has to spend a lot of computing power refuting itself. Most of its time is spent in internal debate.
Wang Lei is on the verge of a mental breakdown.
"Boss, Director Zhao called again. He said the higher-ups want to see the first phase evaluation report and asked when we can submit it."
Tell him, two weeks.
"You said two weeks last time too."
"This time it really is two weeks."
Wang Lei glanced at Xia Zijing, hoping to find some comfort in her expression.
Xia Zijing was looking at the data and didn't even look up.
"Don't look at me. If he says two weeks, then it's two weeks."
Wang Lei sighed and went back to writing code.
On the third day, things changed.
The output of the Sky Eye has begun to rebound. Not because it has stopped its internal debate, but because it has accelerated its debate.
On the first day, a self-check takes four to six seconds.
On the third day, it was shortened to 1.2 seconds.
It is adapting. It is evolving.
Xia Zijing's analytical model shows that the length of FAST's reasoning chain is increasing, but redundant nodes are decreasing. It is beginning to learn to filter—which conclusions are worth in-depth self-examination and which can be quickly approved.
"It's establishing its own judgment priorities," Xia Zijin said.
"Translate it into human language?" Wang Lei asked.
"It knows when to take things seriously and when to let things go."
The afternoon of the fifth day.
When processing a military satellite image, FAST (Five-hundred-meter Aperture Spherical Telescope) produced a completely different output than before.
Image content: A reconnaissance photograph of a certain sea area. Three civilian cargo ships are on the sea.
Standard task: Identify vessel type, course, and load status.
The output of the Heavenly Eye:
[Vehicle A: Bulk carrier, heading southwest, fully loaded. Vessel B: Container ship, heading due south, half-loaded. Vessel C: —]
The output stopped here.
It didn't get stuck. The Sky Eye stopped on its own.
Five seconds later, new text appeared.
[Vessel C's external features are consistent with those of a civilian oil tanker. However, the following anomalies need to be noted: 1. The waterline depth does not match the tanker's empty-load condition; the actual draft is between full load and half load, contradicting the declared "empty-load" status. 2. There are two abnormal protruding structures covered by tarpaulins on the deck. Their dimensions do not match standard deck equipment. 3. The vessel's course has been corrected twice in the past four hours without any apparent reason, each correction angle being 12-15 degrees. This is inconsistent with normal commercial route planning.]
[Comprehensive assessment: Vessel C is likely not a civilian vessel. Confidence level: 71%. It is recommended to expand the monitoring area and cross-verify the AIS signals.]
[Self-check result: The above judgment is based on indirect evidence reasoning, which carries a risk of misjudgment. However, if it is not marked, the potential risk outweighs the cost of misjudgment. Therefore, marking is chosen.]
Zhou Weiguo was the first non-technical person to read this output.
He watched it three times.
Then he turned to look at Lin Yu.
"Did it discover this itself?"
"Nobody taught it to read the waterline."
Lin Yu's voice was calm, but his speech was a beat slower than usual. "No one taught it that the disguised ship can adjust its course to avoid certain monitoring areas."
"Where did it learn that?"
"Among the dialogue training samples we fed it before, there were liars."
Xia Zijing continued, "It has transferred the human behavior pattern of concealing true intentions to the analysis of ship behavior."
Zhou Weiguo remained silent for ten seconds.
"It believes the ship is lying."
To be precise,
Lin Yu leaned back in his chair. "It believes that the ship's apparent state is inconsistent with its actual state. It has seen this inconsistency in human conversations before."
Zhou Weiguo walked out of the main control center.
He returned twenty minutes later, carrying a document.
"I just confirmed with the Office of Naval Intelligence."
He placed the documents on the table. "There was indeed a routine patrol in that area last week. The patrol report mentioned a suspicious oil tanker."
He looked at Lin Yu.
"The AIS signal indicated no load. However, the radar echo characteristics were abnormal. Later, the ship entered international waters and was not tracked further."
The control center fell silent.
Wang Lei walked in carrying a bowl of instant noodles, saw everyone's expressions, and silently withdrew.
"Zhou Chu," Lin Yu said.
"Um."
"Help me arrange a meeting with Director Zhao."
"When?"
"Now."
Zhou Weiguo picked up the phone.
Director Zhao arrived at the base twenty minutes later. After reviewing the output records of the Sky Eye, he remained silent for a long time.
"Are you sure this isn't your pre-set template?"
"It has no template."
Lin Yu said, "The template is given by people. This is something that grows on its own."
Director Zhao glanced at the timestamp output on the screen.
3:14 PM.
An AI that is still in the learning stage made a judgment in less than two seconds that would take front-line analysts several hours to make.
"It's about the window of opportunity."
Director Zhao closed the document. "I won't rush you anymore."
He stood up and paused at the door.
"Lin Yu."
"Um."
"You said before that you wanted to teach it to think."
"Um."
"It's thinking about more things than you're thinking now. Are you ready?"
Lin Yu did not answer.
Director Zhao has left.
That evening, Lin Yu sat alone at the control panel.
The phone lit up. Shen Yue.
[Urgent. Kevin's new materials consortium suffered a complete failure in its experiments last week. The samples they produced based on the new data showed increased power consumption instead of decreased. Kevin held an emergency internal meeting today, and is said to have been furious.]
Lin Yu finished reading it without showing any expression.
The fake data worked. Kevin wasted at least a month of the three-month window going in the wrong direction.
Just as he was about to put his phone down, a second message came in.
But one thing is strange. Kevin's lab hired a new person yesterday.
[A pure mathematics PhD student specializing in number theory. Has nothing to do with materials science or AI.]
Lin Yu's finger stopped on the screen.
Number theory.
Pure mathematics.
It has nothing to do with materials. It has nothing to do with AI either.
Why did Kevin hire him?
unless--
Lin Yu narrowed his eyes.
Unless Kevin isn't pursuing his skills.
He was pursuing his mathematics.
The underlying technology of the Tianyan instruction set is not computer science.
It's mathematics.
He tapped his fingers three times on the table.
"He's digging the foundation."
Lin Yu did not reply to Shen Yue immediately.
He opened his notebook and flipped to the design draft of the Tianyan instruction set. The dense formulas filled the entire page, but at the very bottom, there was only one thing.
A mathematical structure he began building in high school.
Dimensionality reduction projection of a gauge field on a high-dimensional topological manifold.