I love analogies, and I also love cats. So here’s a cat based analogy that attempts to explain why the apparent intelligence some people are seeing in LLMs is just an illusion.<\/p>\n\n\n\n
Cats like playing with things, and the way they play is through simulated hunting. Guns are inanimate objects with the capacity to cause harm if operated (in)correctly.<\/p>\n\n\n\n
Bring the two together and the analogy begins. Place a loaded gun with the safety off on the floor in front of a cat and there’s a chance the cat will start to play with it. <\/p>\n\n\n\n
(Don’t try this at home).<\/p>\n\n\n\n
Said cat may well prod the gun, or even pounce on it. Generally, nothing will happen, but there’s a chance the cat will hit the trigger and the gun will go off. Hopefully the cat won’t be hurt, but what it will be is frightened.<\/p>\n\n\n\n
The cat has no understanding of how guns work, and although intelligent, it also doesn’t have the knowledge to develop that understanding. What is does have is the ability to conclude that a gun is an active thing that will fight back if sufficiently aggravated.<\/p>\n\n\n\n
The cat might also conclude that the gun is alive and has some level of intelligence to react in this way.<\/p>\n\n\n\n
The cat is, of course, wrong.<\/p>\n\n\n\n
The same is true for people and LLMs. Many people have zero understanding of how an LLM works or what goes on inside it. If one is placed in their vicinity they may well play with it. They’ll prod it and poke it and get a reaction.<\/p>\n\n\n\n
Occasionally, that reaction might appear negative or aggressive. The AI might act in a way that implies harmful intent. There are plenty of examples out there in the wild, some accidentally triggered and some deliberately manufactured.<\/p>\n\n\n\n
Many people are seeing this behaviour and concluding that the AI must be intelligent. It’s behaviour implies conscious decision making driven by negative intent. These people are the cat playing with the gun.<\/p>\n\n\n\n
Even the companies responsible for these models have people in this category. Anthropic employs around 2500 people and I guarantee a significant proportion have no idea how an LLM works. It’s not their job to know, just as many people in EDF Energy don’t know how electricity generation works.<\/p>\n\n\n\n
However, when an AI company claims their LLMs might be intelligent or conscious, there are plenty of people in the company who know otherwise. There are also plenty of people outside the company who are easily convinced, just like the cat would be easily convinced if you could tell it the gun was alive.<\/p>\n\n\n\n
So, don’t be the cat in this analogy. If you’re going to play with the LLM, don’t do it under the false impression that it is intelligent and understands your commands. Use it for what it is – a pattern recognition and regeneration system.<\/p>\n\n\n\n
And remember, if the gun goes off, it didn’t do it on its own. It did it because you told it to. You’re just not aware that you did, and you can only stop it from happening again if you take the time to understand.<\/p>\n\n\n\n
Telling it not to won’t help. That’s not how it works.<\/p>\n\n\n\n
<\/p>\n","protected":false},"excerpt":{"rendered":"
I love analogies, and I also love cats. So here’s a cat based analogy that attempts to explain why the apparent intelligence some people are seeing in LLMs is just an illusion.<\/p>\n","protected":false},"author":1,"featured_media":441,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-440","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-business"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t