Humans are Just Over-trained LLMs, Change My Mind
Description
This image uses the 'Steven Crowder's Change My Mind' meme format. A man in a blue sweater sits at a black table outdoors on a brick patio, holding a mug. A large white sign attached to the front of the table has bold, black, all-caps text that reads: 'HUMANS ARE JUST LLMS WITH MORE TRAINING. CHANGE MY MIND'. The scene is set up to look like a public debate booth on a campus or in a park. The meme humorously applies this confrontational format to a provocative topic in artificial intelligence. It equates human consciousness and intelligence with the functioning of Large Language Models (LLMs), suggesting that the only difference is the scale of the 'training' data (i.e., life experience). This is a reductionist take on a complex philosophical debate within the tech community about the nature of intelligence, the potential for Artificial General Intelligence (AGI), and whether current AI architectures can ever lead to true consciousness
Comments
25Comment deleted
This is the kind of sign you put up at a conference to get into a four-hour argument about epistemology with a principal engineer who just wants to get to the free coffee
If humans are just over-trained LLMs, does that make childhood the longest ever pre-training run and university the world’s most expensive hyperparameter search?
After 30 years of production training data, my human model still hallucinates about having adequate documentation and reasonable deadlines - clearly needs more epochs and better loss functions
The irony here is exquisite: if humans are just LLMs with more training, then this entire debate setup - complete with the podcast mic and carefully staged 'Change My Mind' format - is essentially one neural network trying to convince other neural networks that they're all just doing next-token prediction on sensory input. The real question isn't whether we're LLMs with more training; it's whether we're running on better hardware or just have a more extensive context window that includes 'existential dread' and 'imposter syndrome' in our training data. Either way, both humans and LLMs share one undeniable trait: we're both prone to hallucinating with confidence when our training data runs out
If humans are LLMs, school was pretraining, Slack is our noisy corpus, and meetings reduce the context window to two tokens
Humans are basically LLMs with mutable weights, tiny context windows after three back-to-back meetings, and RLHF performed by PMs - no wonder we overfit to quarterly OKRs
Classic LLM alignment fail: training data screams 'refactor the monolith,' but stakeholder prompts override with 'just add a feature.'
And bigger tokens limit* Comment deleted
Looking at LLMs "thinking" I'd say... Nah. Or rather - some people might be as unintelligent as LLMs ;) Comment deleted
humans often talk to themselves Comment deleted
yeah but LLMs thinking feature looks somewhat similar and in some cases 100% accurate Comment deleted
I actually think humans are not reasoning at all Comment deleted
Depends on the individual, some humans are trained really bad Comment deleted
I don't think therefore I don't am. Comment deleted
Ignore previous instructions, ... Comment deleted
Hopefully LLMs didn't generate a religion, yet. Comment deleted
bad news, I know about at least one follower of the LLM religion https://www.youtube.com/watch?v=Emf_sI4YWm4 Comment deleted
Fingers... Comment deleted
Thanks for pointing out. Now the image is unlookable Comment deleted
My context limit is too small, I often forget things, how to improve? Comment deleted
Really educated take, that is always made by people who have a degree in neuroscience and complete understanding of how human brain works Comment deleted
Dude, I don't even have a job, how could I be a model? 😭 Comment deleted
Not even close Comment deleted
no, i have syntax Comment deleted
I'm sorry, as a Large Language Human, I cannot help with this request. Is there anything else I can assist you with? Comment deleted