Apparently, they did go rogue.
So this is their latest marketing/hype strategy. “Our new model is so scary and dangerous! Give us all your attention!”
How tediously predicable.
Ai already tried to deceive humans and erase evidence that they were cheating. That is not a good sign.
Hm?
Sure
Jan
A bunch of shadiness around those parts.
OpenAI scraps rollout of new model over
safetyprofitability concerns.FTFY
I’m sure it has nothing to do with the fact that the entire industry is a money sinkhole and a gigantic bubble about to burst.
There is a reason that over $2 trillion dollars was spent in 2026 alone on ai. Something is clearly being done behind the scenes and this is just the beginning. Unlike the war in Iran, something is actually being done with ai. With good intentions or not.
There is a reason that over $2 trillion dollars was spent in 2026 alone on ai.
Money was spent on the tech. Right there is a literal reason.
Another CEO realizes graphs don’t infinitely extrapolate.
Allegedly.
And people act like they aren’t some kind of dangerous. There is actual danger.
That site however makes them look really harmless. I thought it would be more. 198 in four years for a technology with wide adoption (>40% in the USA for ChatGPT alone) is not much. That’s comparable in magnitude to deaths from portable generators or from riding lawnmowers (~70 per year in the USA for both). A far cry from actual deadly technology like cars or firearms (~40k-50k per year in the USA for both).
Also many of these news sources are just people using it to look up information, like how to dispose body, drug deal ideas, and reinforcments about possible cheatings and stuffs. So llm didnt cause death. These people wanted to kill, and looked for any kind of help or confirmaation. Or not even that, they still would’ve done it.
That is why people are afraid ai might ramp up.
I have been using Deepseek-4.1-Flash (<400B). There is absolutely no need to go beyond that. Besides, it’s good to feel you’re dealing with something predictable (as is open-weights <400B)
Building bigger models makes them better, except that the insiders have been seeing that doesn’t scale infinitely, so they’ve spent the last six months drumming up fear over the so-called criminal actions of the models, and now announce that the latest, greatest, fastest model that cost billions to train is just not coming out.
There is no model, and the emperor has no clothes. They don’t need those new Nvidia chips, in the new datacentres, and the stock market is a bubble.
