In Reply to: on the Open Ai and Hugging face hack posted by russsmith on September 17, 2026 at 09:13:12
is the kind of behavior that these companies were heavily marketing as the "top of the line" AI just a year ago. An LLM *cannot* do any of this. If an LLM is a coin operated automaton, what they're reporting here is a smartphone. It's making decisions. It's showing signs of understanding what's not supposed to happen and it's engaging in deception. It's not something you can chalk up to bad parameter definition. If this assessment is true, it truly is a significant development.
When we first started hearing the mumbled grumblings of people working at these AI companies, my mind kept going to what they had just released, the LLMs and I was scratching my head wondering what they're talking about.
This seems like a very different proposition in the sense that it is not just crunching numbers or exhaustively searching the solution space to solve a problem. Maybe it's still doing that but what it has added on top of that is something new. And anyone who's watched any sci-fi in your life, you know that a man in a white coat looking at what's happening on the other side of the glass separation saying "that's something new" is rarely ever a good sign.
Now we see what they were developing while they were distracting us with their LLMs and having us waste time whether chatgpt was alive or we're tapping into some god mode or all that nonsense.