AI Researcher Nostalgic for the Pre-Superintelligence Era
Description
A screenshot of a tweet from user Stephen McAleer (@McaleerStephen) on a black background. The tweet's text reads, 'I kinda miss doing AI research back when we didn't know how to create superintelligence.' The timestamp below indicates it was posted on '10:16 AM · 04 Jan 25'. This meme uses dry, satirical humor to comment on the rapid acceleration of AI development. The joke lies in the audacious and hyperbolic implication that researchers *now* know how to create superintelligence, a milestone still considered theoretical and distant. For a senior technical audience, it's a witty take on the current AI hype cycle, the dizzying pace of progress with models like LLMs, and the mix of excitement and existential unease that accompanies the pursuit of Artificial General Intelligence (AGI)
Comments
7Comment deleted
The good old days, when 'aligning the model' meant centering a div and not preventing a rogue AGI from turning the internet into a paperclip factory
Remember when our biggest worry was vanishing gradients, not vanishing control of the species?
Remember when our biggest AI concern was overfitting on MNIST? Now we're writing alignment papers that read like hostage negotiation strategies for our own code
Ah yes, the good old days of AI research - when 'gradient descent' was just our career trajectory, not the path to AGI. Now we've gone from 'can we make it learn XOR?' to 'should we be worried it's reading our Slack messages?' Turns out the real superintelligence was the existential dread we accumulated along the way. At least back then, our biggest fear was overfitting the training data, not the AI overfitting humanity's entire decision-making apparatus
Product now asks for an ETA on “alignment” like it’s a feature flag - back when risk meant “overfit,” not p(doom)
Pre-superintelligence days: vanishing gradients kept us up at night, not vanishing jobs
Superintelligence went from thought experiment to a Jira epic; the remaining blockers are compute, alignment, and convincing Legal that “emergent capabilities” isn’t a compliance category