[new blog post] Research, writing, and career advice https://t.co/4kRrIobVCF

Very fun to listen to @muratdemirbas and @AlekseyCharapko's real time thoughts and reaction while reading our paper 😃

Murat and Aleksey Read Papers: "Barbarians at the Gate: How AI is Upending Systems Research" Here we discuss this fun paper and add a bit of our skepticism on AI upending research and what it means for community. https://t.co/fzjzrvsC84

LLMs are bad researchers. The shoe fits. https://t.co/8TjcXWTOfp The real question is: Can an army of bad researchers make one mediocre researcher a good one? Or help a good researcher become great?

LLMs operate on the visible tip of this iceberg, and cannot handle the 5Cs below: Curiosity, Clarity, Craft, Community, Courage. They may even erode those deeper skills. https://t.co/13ziHzG7Nh

Worst time I ever had was when I had to work with six-seven (I did it! my middle schooler daughter would be proud) PhD students. I would rather work with two sharp creative PhD students that I support myself, than manage 50 mediocre PhD students, handed to me for free.

Yet, when used for rapid prototyping, LLMs can help improve the design. And, through faster iteration, you may get to a high-value insight. https://t.co/1vIDc74C5G

I have been predicting the collapse of the publishing system for a decade. The flood of LLM-aided research might finally break its back, and that may be good news! We are due for a better model/process.

Coming from about the most powerful mind I know in my generation, I’m just really grateful for this. Speechless. Ed and I just spent 6 days on Geometric Unity. Best work out the theory has ever had. Much more to come. Thanks Ed. 🙏

How do we prevent reward hacking? How do we stop LLMs from exploiting loopholes in the evaluator? Specifications become much more important! And writing specifications require clarity, and insights already!

What happens when we stop getting our hands dirty? Can we still distill high-value insight without spending time in the trenches? https://t.co/QQbwaS1f9c

I have been predicting the collapse of the publishing system for a decade. The flood of LLM-aided research might finally break its back, and that may be good news! We are due for a better model/process.

What happens when we stop getting our hands dirty? Can we still distill high-value insight without spending time in the trenches? https://t.co/QQbwaS1f9c

LLMs are bad researchers. The shoe fits. https://t.co/8TjcXWTOfp

Who gets the credit when the machine does the work? When does authorship still belong to the human? One thing is clear, this will finally break the publishing process for good. Barbarians at the gate, indeed.

@muratdemirbas After reading "that" paper yesterday night, I had a dream of LLM agent iterating through my child's clothes to find the best outfit, while my daughter was the "evaluator" and rejected everything

LLMs are bad researchers. The shoe fits. https://t.co/8TjcXWTOfp The real question is: Can an army of bad researchers make one mediocre researcher a good one? Or help a good researcher become great?

LLMs operate on the visible tip of this iceberg, and cannot handle the 5Cs below: Curiosity, Clarity, Craft, Community, Courage. They may even erode those deeper skills. https://t.co/13ziHzG7Nh

Worst time I ever had was when I had to work with six-seven (I did it! my middle schooler daughter would be proud) PhD students. I would rather work with two sharp creative PhD students that I support myself, than manage 50 mediocre PhD students, handed to me for free.

Yeah, trying to collect my thoughts, after having read "that" paper.

Yet, when used for rapid prototyping, LLMs can help improve the design. And, through faster iteration, you may get to a high-value insight. https://t.co/1vIDc74C5G