308 - Cybercrime

After more than 2 years, LLMs and RL models (with or without fraudrailing) are still vulnerable-by-design, as nothing has fundamentally changed about the architecture. If anything, “Agentic” BS has only compounded the cybersecurity problems. One recent video example is shown here

These cybersecurity threats are only “mitigated”, in a completely opaque manner, after researchers explicitly walk LLM providers through demonstrations of the specific exploits. This also means that any exploit is likely to remain active in the wild for so long a time as no cybersecurity researcher finds and reports that exact exploit. There is very little incentive to report them, and major incentives to exploit them, so it is statistically unwise to rely on charity from researchers, particularly when they’re handing their findings over to some of the tech industry’s most obvious con artists, people who they’re not likely to be philosophically aligned with.

Cybercrime was already the 3rd largest market in the world in 2023, at over $8 trillion in annual revenue according to Statistica, and it has benefited more than any other industry from both “GenAI” and the fraud that came before it, “Crypto”. Updated projections place it at over $10 trillion in annual revenue for 2025, and even after the bubbles of trashbot AI and the latest Crypto bubble burst that added value to the cybercrime market will persist.

The technology is built from stolen data, and built to steal more data, by people with direct ties to Crypto (like OpenAI, Anthropic, and Musk), with that post-bubble added value residing in cybercrime, and Crypto predictably rising in value when they inject uncertainty into the market. It is a hot mess of Crypto, Ponzi schemes, blatant thievery, and social engineering to cultivate a captive and docile audience.

When you see the US talking about investing $500 bn in “AI”, or the EU $200 bn, those investments are either directly in or supporting these kinds of cybercrime. The dumbest decisions in human history are being made at the greatest scales ever seen, and they are happening each day. Even Hitler didn't propose turning concentration camps into luxury resorts, but cognitive decline is rapidly descending into the abyss this year.

Neither the US, nor China, nor the EU hold any advantage in AI today. The cybercrimes may vary, but the tools used to commit them are the same.

Every day for the past few years people with influence and the power to make investments have chosen this path, over and over again. They still show no signs of doing anything else, as whatever words they may say, only their actions hold sway.