alt Hacker NewsI've got some personal litmus tests:
1) Syntax/semantic split. Can the person accept that a function called "multiplyBy5(a,b) { return a+b }" doesn't actually multiply by five, but adds the numbers? 2) PR speak: Does the person recognize that public relation speak is usually intentionally misleading, as in "the Russian Ministry of Defense said that a fire [onboard the Moskva] had caused ammunition to explode" (obviously caused by an Ukrainian missile and not an accidental fire, even though that's what's implied.) [0] 3) They're, their, there: There easy to tell apart, since they're meaning is so different. /s 4) Viewpoints: Can this person understand and articulate viewpoints that they consider "wrong" or simply don't hold themselves? 5) (new) LLM introspection: Does the person understand that LLMs have no secret understanding of themselves? An LLM like "Grok" doesn't actually understand "Grok" better than Gemini understands "Grok" - apart from minor differences in model strength maybe.
Replies
>LLMs have no secret understanding of themselves
What do you mean by "themselves" here? Grok is RL'd to behave like a Grok, so it trivially knows the qualities that define Grok better than Gemini does, which can only go by second hand sources.
I've heard that a non-Mensan asked a Mensan what it's like at a Mensa event. They replied, "If I have to explain something during a conversation, I only have to explain it _once_."
Not bad litmus tests. And yes a lot of idiots seem to fail at steel manning. I mean if you can't steel man your opponent what are you even doing?
Are there people that would legitimately argue point 1?
If you are only looking at the call site, sure that could be confusing, but if you are looking at the definition as provided in your post, surely anyone that is able understand the concept of a function can see the problem?
I'm not arguing they don't exist, sure they do, but I'm confused as to how you came up with it as a litmus test? Is it that common?
Surely we can agree in a real scenario renaming it (or fixing implementation to match name) is likely appropriate, but to completely miss the error?
Hope this comes across as curiousity, because I am curious about this one from your list in particular.