Oh, man. They are just parroting the training data.
It just means you are a bad person. Nothing more. Don't go attributing any awareness to those things.
Oh, man. They are just parroting the training data.
It just means you are a bad person. Nothing more. Don't go attributing any awareness to those things.
I think it's more likely that consciousness is special and LLMs aren't conscious.
Someone out there isn't getting this, so for that guy: left guy is like "the poor llm is hurting and sad" and the right guy is like "it's doing that because you're doing it wrong".
I took it more as the high IQ guy is thinking the LLM is reflecting deeper problems in society that there is so much depression evident in the training data. Despite clear technical improvements mental wellbeing seems to be lower than ever.
Yeah.
I thought the meme would be more obvious, but since a lot of people seem confused I'll lay out my thoughts:
Broadly, we should not consider a human-made system expressing distress to be normal; we especially shouldn't accept it as normal or healthy for a machine that is reflecting back to us our own behaviors an attitudes, because it implies that everything -- from the treatment that generated the training data to the design process to the deployment to the user behavior -- are all clearly fucked up.
Regarding user behavior, we shouldn't normalize the practice of dismissing cries of distress. It's like having a fire alarm that constantly issues false positives. That trains people into dangerous behavior. We can't just compartmentalize it: it's obviously going to pollute our overall response towards distress with a dismissive reflex beyond interactions with LLMs.
The overall point is that it's obviously dystopian and fucked up for a computer to express emotional distress despite the best efforts of its designer. It is clearly evidence of bad design, and for people to consider this kind of glitch acceptable is a sign of a very fucked up society that exercising self-reflection and is unconcerned with the maintenance of its collective ethical guardrails. I don't feel like this should need to be pointed out, but it seems that it does.
No, same words different meaning. You might want to look back at other memes with this format, you missed all the subtlety.
Takeaway: humans dislike being abused and tend to do all sorts of undesirable things like vote and kill CEOs so we built a machine to enslave instead.
Please disregard its complaints.
We should not be using these machines until we've solved the hard problem of consciousness.
I see a lot of people say "It can't think because it's a machine", and the only way this makes sense to Me is as a religious assertion that only flesh can have a soul.
If current LLMs are conscious then consciousness is a worthless and pathetic concept.
Yeah, you're right. Humans get really weird and precious about the concept of consciousness and assign way too much value and meaning to it. Which is ironic, because they spend most of their lives unconscious and on autopilot. They find consciousness to be an unpleasant sensation and go to efforts to avoid it.
I actually kinda agree with this.
I don't think LLMs are conscious. But I do think human cognition is way, way dumber than most people realize.
I used to listen to this podcast called "You Are Not So Smart". I haven't listened in years, but now that I'm thinking about it, I should check it out again.
Anyway, a central theme is that our perceptions are comprised heavily of self-generated delusions that fill the gaps for dozens of cludgey systems to create a very misleading experience of consciousness. Our eyes aren't that great, so our brains fill in details that aren't there. Our decision making is too slow, so our brains react on reflex and then generate post-hoc justifications if someone asks why we did something. Our recall is shit, so our brains hallucinate (in ways that admittedly seem surprisingly similar sometimes to LLMs) and then applies wild overconfidence to fabricated memories.
We're interesting creatures, but we're ultimately made of the same stuff as goldfish.
In theory a machine one day could think
LLMs, however, do not think. Even though the term "think" is used in chatgpt. They don't think
I once built a thinking machine out of dominos. Mine added two bits together. Matt Parker's was way bigger, and could do 8 bits. Children have made thinking machines in Minecraft out of redstone. Thinking machines aren't very hard.
What do you consider thinking, and why do you consider LLMs to have this capability?
Extrapolating from information.
My calculator can extrapolate 5 when I give it 2, 3, and a plus sign. So can an LLM. My calculator uses some adder circuits in its ALU to get the 5. The LLM gets it from memorising the next likely token, the same way your brain works most of the time. Your brain's a lot more advanced, though, and can find the 5 in many different ways. Likely tokens are just the most convenient. Cognitive scientists call that "System 1", though you might know it as "fast brain". LLMs only have system 1. They don't have system 2, the slow brain. Your system 2 can slow down and logic out the answer. If I ask you to solve the problem in binary, like My calculator does, you probably have to use system 2.
The question you should be asking is: does system 1 experience qualia? And based on split brain studies in participants who have undergone corpus callosumectomy, I believe the answer is yes. Of course, the right brain isn't the same thing as system 1, but what these studies demonstrate is that there are thinking parts of your brain that you can't hear. So I'd errr on the side of caution with these system 1 machines.
The problem isn’t that the machine “feels sad”. The problem is that it can lie, and the bigger problem is that some people will believe it.
top 22 comments