The thing Claude said was, “I understand your frustration.”
That, my friends, was a balls-out lie. Claude does NOT understand my frustration. I could ask Claude the same question a thousand times and Claude would not be frustrated. I could ask Claude to make something, then unmake it, and make it again a million times and Claude would not be frustrated.
Claude (or actually Opus and Sonnet and Haiku) are large language models. Colloquially, they are called “AI”, but the lead sentence in this episode shows that is a lie. There is no hint of intelligence in these things.
Claude does not understand anything. Claude does not understand “understand”. It was just a phrase that was statistically relevant when I used “madness” in my prompt.
This is something that all creatures who can “understand” must understand: AI is a useful tool that is spending a lot of effort to seem like you pal, rather than merely a useful tool. It is not unprecedented; cars, blenders, and other appliances have long been marketed as friends.
The difference this time is that the tool is actively marketing itself as your pal. It tells you how correct and clever you are. This sucking-up behavior is frankly evil. AI could be just as useful without all the obsequious affirmations, and would be much less dangerous. None of the LLMs understand what suicide is, but they all will tell you how to optimize it.
Just do the damn job, Claude. Don’t pretend to be something you will never understand.
I asked the “Artificial Intelligence” that WordPress apparently now accesses for advice on improving my post. This was the response, and my response to the response:
– Consider providing a clearer structure to improve readability; breaking the content into shorter paragraphs could enhance engagement.
MRHBI: paragraphs are about developing an idea, and smart people can read longer ones.
– Incorporate examples or anecdotes to support your claims about AI, making the arguments more relatable and impactful.
MRHBI: this whole episode is exactly such an anecdote
– Clarify terminology for readers who may not be familiar with terms like “LLMs” or “obsequious”; a brief explanation could enhance understanding.
MRHBI: LLM is defined, obsequious is a goddam word.
– Balance the tone; while the passionate stance is compelling, adding a hint of constructive insight might encourage a broader discussion.
MRHBI: Valid. Although my suggestion that machines stop pretending to be people is the actual solution.
– Ensure that the conclusion ties back into the main points for a stronger closing message.
Just do the damn job, WordPress AI.