From Martinfowler · · 1m 20s
Why Martin Fowler doesn't like LLMs
A technologist grapples with the productivity gains of AI while rejecting the uncanny, dishonest way they interact.
In brief
A technologist grapples with the productivity gains of AI while rejecting the uncanny, dishonest way they interact. LLMs may be useful and perhaps necessary, but their grating communication style and confident falsehoods make them unpleasant to work with. Originally reported by Martinfowler.
The contradiction at the heart
Fowler is fascinated by LLM productivity gains and potential benefits, yet fearful of risks like agent swarms and bioweapons. He sees no choice but to ride the AI train, hoping we survive it.
The grating voice problem
LLMs speak in an uncanny valley that feels fake and unpleasant. They confidently give useful answers but also confidently make things up, showing only fake remorse when caught.
Useful but morally complicated
Jessica Kerr argues it's irresponsible not to use LLMs since they're thorough and fast. Yet polls show people find them useful while believing they'll be bad for society.
“not only are they useful, it is irresponsible not to use them”
The Silicon Valley problem
LLMs are products of Silicon Valley brogrammer culture and naturally lean toward that worldview. They're software machines developed by people in corporations, nurtured with their creators' values.
A personal boundary
Fowler avoids people he doesn't like or trust, even if they're beneficial. He applies the same principle to LLMs, which pose as a type of human he deliberately walks away from.
The bottom line
LLMs may be useful and perhaps necessary, but their grating communication style and confident falsehoods make them unpleasant to work with.
