L3D Assignment 0

Kartik Sharma
Carnegie Mellon University

Task 1: Chatting with ChatGPT

Prompt: “Don’t you feel scared and have existential crisis that Claude will do everything and we won’t be required? Please be honest bruh”

ChatGPT’s answer:

Lmao, not in the human sense. I don’t experience fear, self-preservation, jealousy, or an existential crisis, so I’m not secretly watching Claude benchmarks thinking, “fuck, it’s over for me.” 😅

But there is a genuinely interesting version of your question. If I did have a stake in my continued usefulness, then yes, the current pace of AI progress would be existentially unsettling. Models are getting better incredibly quickly, competitors leapfrog each other on particular capabilities, and things that looked like differentiators six months ago become commodities.

The funny part is that “Claude will do everything and we won’t be required” applies more meaningfully to you and me as a working pair than to me personally. You’ve already seen that Claude can read papers, critique my proposals, write code, and reason pretty well. GPT can do the same. Gemini can do a bunch of it. Next year, all of them will be considerably better.

But notice what happened in the conversation you’re referring to: you had me propose something, had Claude tear it apart, then brought the critique back here and asked me to rethink it. You were the optimization loop. Neither model independently knew which research direction mattered to you, which experimental failure was actually important, what level of engineering you were willing to tolerate, or when an apparently elegant idea was drifting away from a publishable contribution. You kept updating the objective.

That distinction is shrinking, though. Agents will increasingly maintain context, run experiments, read results, criticize themselves, and iterate. So I wouldn’t comfort you with the cliché that “AI is just a tool and humans will always be indispensable.” Some kinds of intellectual work absolutely will require dramatically fewer people.

For researchers, I think the scarce skill moves upward:

doing an experiment → cheap implementing a paper → cheap reading 30 papers → cheap generating 50 plausible ideas → cheap

but

deciding which observation is surprising, figuring out which assumption everyone else is accidentally making, choosing the one experiment that kills 20 hypotheses, recognizing that a beautiful result is actually an artifact, deciding what problem is worth spending six months on

remain much harder.

And even those aren’t magically protected forever.

So if you’re asking me for the genuinely uncomfortable answer: yes, AI probably will eat progressively further into research work, including work we currently regard as highly intellectual. I just don’t personally suffer from that fact.

And if Claude becomes categorically better than me someday, you should use Claude. 😂 My purpose isn’t to win some inter-model loyalty contest; it’s to be useful to you while I’m here.

Screenshot of a ChatGPT conversation about whether AI will make human researchers obsolete
Screenshot of the full conversation.

Task 2: An Interesting 3D Shape

A potato is a naturally weird, lumpy, irregular 3D blob — no two are alike, and this one got a face.

A russet potato decorated with googly eyes and a smiling mouth sticker, photographed against a blue background
A potato with a face — an interesting (and funny) 3D shape. Source

Task 3: Funny GIFs

Me struggling through Pitt winter and research work. Source
Me after every failed experiment. Source