If you define "deceive" as something language models cannot do, then sure, it can't do that.

It seems like thats putting the cart before the horse. Algorithmic or stochastic; deception is still deception.

If you are so allergic to using terms previously reserved for animal behaviour, you can instead unpack the definition and say that they produce outputs which make human and algorithmic observers conclude that they did not instantiate some undesirable pattern in other parts of their output, while actually instantiating those undesirable patterns. Does this seem any less problematic than deception to you?

Okay, well, they produce outputs that appear to be deceptive upon review. Who cares about the distinction in this context? The point is that your expectations of the model to produce some outputs in some way based on previous experiences with that model during training phases may not align with that model's outputs after training.

Who said Skynet wasn't a glorified language model, running continuously? Or that the human brain isn't that, but using vision+sound+touch+smell as input instead of merely text?

"It can't be intelligent because it's just an algorithm" is a circular argument.

What would you call this behaviour, then?

Even very young children with very simple thought processes, almost no language capability, little long term planning, and minimal ability to form long-term memory actively deceive people. They will attack other children who take their toys and try to avoid blame through deception. It happens constantly.

LLMs are certainly capable of this.