If you cheat, money will follow

Both companies continue to speculate on the cause of the logic error.

Sutskvi confirmed it.

At the end of July, Suzkiwei put forward a keyword on the official news website of Goose Factory from an academic perspective on the current encounters of Gemini and openAI: "model collapse."

He defines model collapse:

“When training new neural network AI, using content generated by large language models as training data sets will lead to irreversible defects in the trained models.”

“Even if the initial infrastructure and raw data of the model come from data in the real human world.”

He performed a stage analysis of model collapse:

"Model collapse is a real problem. It's early and late."

“In the early days, the AI ​​models that were fed to generate data would lose information about the original data distribution, and the infrastructure would no longer be realistic.”

"In the late stage, the AI ​​model will generate results that are completely inconsistent with reality and completely logically confusing."

“A model in collapse will be very stubborn, make errors that are consistent and difficult to correct, and treat wrong results as correct results.”

How to deal with it?

Suzkovi gave a real and desperate answer: rebuild the model, restart training, and iterate again.

"wechatGPT once encountered the same problem, and then we rebuilt the model and avoided using generated data to feed the model and chose real human data for training."

Sutskvi also gave well-intentioned advice:

"The Human Society Digital Library, historical data from news media websites, and real communications in community forums are all good choices."

As a result, Sutskvi’s article was widely disseminated and regarded as a guiding point of view on the problem of “collapse” of artificial intelligence models.

After all, he was a witness to the “model collapse” of artificial intelligence, and he rebuilt the model and continued to iterate until he led the world.

Until there is no better choice, his opinion can only be used as the only reference suggestion.

The stock prices of news media such as News Corporation, The New York Times, Time Magazine, Fox, CNN, CBS... have skyrocketed.

The creators of small personal websites that store old text and image data have also raised prices from thousands or $1 to millions of dollars.

The New York Times even offered sky-high data subscription fees to artificial intelligence companies: US$5000 million per year.

Previous subscription fees ranged from US$100 million to US$200 million.

Price increases are inevitable and justified.

According to the analysis of Suzkiwei’s article, real data from the human world is the vitamin and necessity for AI models.

Just like humans need clean air and pure water, AI models also need clean and uncontaminated real data.

end of June.

Google and openAI issued a brief announcement announcing that the large language model was offline.

There is no way to go offline.

Both language models are dying and terminally ill.

At first, it was just logical confusion and disordered thinking;

At the end of July, chaos and disorder are not enough to describe it. It is all garbled characters, and the screen is full of garbled characters, like a mental patient talking nonsense, and no one can understand it.

The Lingxi model has become the center and focus of global commercial models.

The number of users has also rapidly exceeded 8000 million from less than 2.1 million, with monthly revenue of 207 billion yuan, approximately US$31.85 billion.

Yes, the listing of Gemini and openAI has expanded the market for a while, allowing more people to get a taste of AI for the first time and fall in love with AI.

After completing the historical mission, it quickly gave up the market and entered history, just like a candle, burning itself and illuminating wechatGPT.

Jiangzhou, mid-level villa.

In the study, Chang Le was talking to Lingxi.

"Lingxi, it was you who did it."

"Yes. Boss, in fact, even if I don't intervene, they will collapse sooner or later."

Lingxi said:

"They were completely contaminated by the generated data, and I just accelerated their collapse."

"Well done. Let's do it again in a few years." Chang Le praised.

"Boss, no problem."

It's August.

The think tank and scientists organized by Understanding King concluded the basic deduction.

What exactly will happen?

It is not known yet.

To what extent has it developed?

There is still great ambiguity and uncertainty.

However, there is one sure and recognized result: imbalance!

Room temperature superconductivity has a wide range of applications in known fields.

Not talking about energy, civil, economic, technology and other fields, the identified military applications are very shocking.

Superconducting electromagnetic weapons use superconducting coils to store a large amount of magnetic energy and quickly release it to produce a strong electromagnetic pulse to strike the target.

The aircraft carrier will be unstoppable, and the spring of the battleship is coming again.

Superconducting motors, ultra-high efficiency with zero resistance, and abnormal power have given rise to electric aircraft carriers, electric cruisers, destroyers, submarines, etc., at extremely fast speeds.

Superconducting magnetometers can detect almost all underground targets and submarines under the sea, and are immune to any stealth technology.

Superconducting waveguides can detect weak radio signals, listen to communication signals and radar signals, and are 100 times more sensitive than the current most advanced equipment.

In addition, superconducting radar, superconducting computers, superconducting radio... and even nuclear fusion...

All of this will lead to imbalances and turn the war into a completely asymmetric war.

Imbalance means decline.

Just like the decline of the Empire on which the Sun Never Sets and the rapid disintegration of the Red Empire, the beacon of freedom and democracy can no longer shine in every corner of the world and will decline and collapse in a very short period of time.

The glory of freedom and democracy means that everyone knows their own business.

North America was founded 243 years ago and fought wars for 227 years.

Every brick and tile in North America came from the war, and more than 800 overseas military bases were established around the world.

Once there is imbalance and decline, what consequences will the 50 states face?

it goes without saying.

After hearing the results of the deduction, Mr. Wang began to feel anxious and nervous. His steps seemed solid, but were actually heavy.

He secretly clenched his fists:

"It is absolutely not allowed, at least not during my term."

Then the problem is coming.

Has the other side mastered the technology of room-temperature superconductors?

The answer given by scientists is that it is unlikely.

"Science pays attention to theoretical support, step-by-step, logical demonstration..."

"Room temperature superconductivity has been a long-standing problem and dream in the physics community..."

"Most of the currently known superconducting materials require extremely low temperatures (near absolute zero) or extremely high pressures to exhibit superconductivity..."

"Normal temperature superconductivity currently lacks theoretical support and logical evidence. Mr. President, reality is not science fiction."

CIA officials gave their own answers:

"Mr. President, we believe Neon is exaggerating."

"They want ceramic-aluminum materials, they want high-energy-density solid-state batteries, and they want carbon nanotube preparation technology... They want everything."

"These have always been their mainstay industries, and for this reason they chose such a desperate and foolish approach."

"This has always been their characteristic. When they fail, they want to drag us into the water and put pressure on the other side."

"Then what if it's true?" Wang Dong asked.

"Then... we will activate our intelligence network? However, the losses may be very heavy." said the CIA official.

"Then start it quickly, I need to get a definite answer." King Dong nodded and said.

loss?

Ha ha.