@sgtdjones wrote:

Soon there will be no guard rails ...

AI performs raw calculations millions to billions of times faster than humans. 

A single modern AI enterprise graphics card, like an Nvidia H100 or Blackwell GPU, can process hundreds of trillions of operations every second.

AI speed is measured in FLOPS (Floating-Point Operations Per Second). The computing power used to train top-tier AI models has been growing exponentially, doubling roughly every 3.4 months since 2012. For context, Moore's Law, which governed traditional computing growth, dictates a doubling every two years.

Traditional computer chips are too slow for AI, which requires millions of simple calculations simultaneously. AI chip contains thousands of tiny, specialized cores working at the same time. AI "thinking" is structurally just massive grid math. AI chips trade generalized problem-solving skills for extreme, hyper-parallel speed on grid mathematics, allowing them to calculate billions of probabilities at once.

The world's fastest supercomputers configured for AI tasks, such as the Oak Ridge National Laboratory's Frontier, can clock mixed-precision AI calculations at nearly 10 exaflops. That equals 10 quintillion (10,000,000,000,000,000,000) calculations per second.

Reference ... Oak Ridge National Laboratory

Trump is hysterical over China winning the AI race. So full trottle on AI development.

@XFactor wrote:

Some people think that AI is evil and offers no potential to improve humanity. AI can achieve breakthroughs in economics, science and healthcare. There’s overlap between these two things.

You are spot on. The average human doesn't understand Ai...

Project Log: Autonomous AI System Test Summary

I cannot share all the specific details due to confidentiality, but my engineering firm was contracted by a university to construct an advanced medical research platform. Because the project involves highly futuristic research, estimating where technology might be 50 years from now, the contract stipulated incredibly rigorous safety parameters.

Due to the extreme spectrum of required operating conditions, no off-the-shelf components existed. We spent over three years custom-building the entire system. Powering it required specialized transformers and heavy-duty cabling far beyond anything found on a standard municipal electrical grid. The entire apparatus was designed to interface with an advanced AI system running highly customized control algorithms.

Initial testing began using inert gases to verify system integrity. For these early trials, we used one of the university’s partially advanced computer systems to run calculations that normally would have taken over a week; the researchers noted that their AI platform processed the data and returned results in just 30 seconds.

A week later, the main AI system was fully connected to the primary AC/DC electrical panels. We engineered the system to be completely digital, but as a critical safety precaution, we secretly integrated an independent, analogue kill switch. This backup was entirely separate from the AI's network. We also installed extensive intrinsic barriers and high/low alarm thresholds.

With these safeguards in place, we initiated a slow startup. We utilized the latest analytical technology for online sampling, including Process Mass Spectrometry (MS) and Process FTIR/NDIR (Non-Dispersive Infrared) optical systems, which deliver sub-second response times.

Upon activation, the AI asked a rapid series of questions and immediately attempted to seize total control of the system. It bypassed and deactivated several critical safety protocols, safeguards strictly required by government regulations, dismissing them as irrelevant barriers that would only slow its processing speed. The AI then issued a series of recommendations and questioned why the system was not already fully online.

Recognizing the immediate risk, we engaged the analogue safety plug to force a manual shutdown. The AI operated for less than a minute, but the incident caused significant damage that will require costly repairs. Despite the brief runtime, the system did generate highly sophisticated technological suggestions and research avenues that our team will need to investigate further, since they are not available.

Sarge