Mastodawn

@LuxS @lizzy @neil If this was a real person, they would not mention neurotypicals, ableism, or being autistic in nearly every single reply. Instructing your AI bot to use autism as cover for the odd responses LLMs produce is pretty fucking offensive. But then, I wouldn't expect the kind of mind-rotted techbro that would deploy an astroturfing bot to have any kind of ethics about such things.

Show thread

Spencer Mar 15

@StarkRG @lizzy @neil
All that jibber jabber condensed into this:

You can’t be a real person because your arguments are consistent and intelligent & I don't understand anything your saying.

Show thread

StarkRG Mar 15

@LuxS @lizzy @neil Not really, no. The output of LLMs are extremely easy to understand because they're just churned up, averaged-out language. Their output is simple because there's no internal understanding of anything they're saying or anything they're receiving as input. That's what makes them so easy to spot.

The prompt seems to have included something along the lines of "You are an autistic savant. Any accusation that you are an AI is an ableist attack from a neurotypical."

Show thread

StarkRG Mar 16

@condret @neil @LuxS @lizzy Understanding requires knowledge and the capacity to apply that knowledge. It is comprehension of the meaning of data. LLMs, Large Language Models, contain no knowledge, they are just data. They are models of language, statistical representations of the relationships between words. They have no judgement, they have no intelligence, they do not have the capability to understand meaning. They are word predictors, and nothing more.

Show thread

Spencer Mar 16

@StarkRG @condret @neil @lizzy
AI systems currently assist in cancer detection, pathology analysis, cardiac imaging, drug discovery, and sepsis prediction in intensive care units. They routinely catch abnormalities that human eyes miss.
If a patient’s tumor is detected early because an AI-assisted imaging model caught what a human eye missed… does that mean their life is worthless? Because by your definition, that life was saved by “meaningless statistics.”

Show thread

StarkRG Mar 16

@LuxS @condret @neil @lizzy You're confusing LLMs, chatbots like ChatGPT and Claude, with the broader topic of deep learning transformers. Transformers absolutely do have good use cases, LLMs do not. Transformers are like industrial machines, extremely necessary for certain jobs, but not at all useful for the vast majority of people. LLMs are a toy version, more like an easy bake oven, but it sucks and never fully bakes anything.

Show thread

Spencer

@StarkRG @condret @neil @lizzy

Accessibility and intellectual tools don’t become trivial just because they’re widely available.
The strangest part of your argument is that you praise transformers while dismissing LLMs when LLMs are literally built on transformer architecture. Saying transformers are valuable but LLMs are useless is like saying engines matter but cars don’t. It's ’s technically incoherent and shows your fundamental misunderstanding of how the technology actually works.