It's because the objective is not truth but attention and they get RL'd by it, so they are a lot more optimal than you give them credit for.
@karpathy
-
Product Integration with LLM Technology
By
–
Nice exactly, good / clean summary. “Does your product speak LLM?”
-
Software Inertia and LLM Adaptation: UI Automation vs Legacy Systems
By
–
Imo you are also dramatically under-estimating inertia, including in Software (e.g. see pervasive use of COBOL to this day). The more general formulation looks something like this. Do LLMs adapt to all existing software? (e.g. Operator seeing UI screens, making clicks)
or
Does -
Cursor’s Code Interface and Natural Language Interoperability Capabilities
By
–
Possibly I agree – in Cursor you're still pretty much looking at code just as before, too. My point is that "add mountains in the background and make the punch look faster and more intense" would just work. It's less about the input to the human and more about interoperability.
-
LLMs and Documentation Should Avoid Excessive Step-by-Step Lists
By
–
Yeah exactly, I weep every time an LLM gives me a bullet point list of the 10 things to click in the UI to do this or that. Or when any docs do the same. "How to upload a file to an S3 bucket in 10 easy steps!"
-
UI Products Without Scripting Support Won’t Survive AI Era
By
–
Products with extensive/rich UIs lots of sliders, switches, menus, with no scripting support, and built on opaque, custom, binary formats are ngmi in the era of heavy human+AI collaboration. If an LLM can't read the underlying representations and manipulate them and all of the
-
Visual Design in Code Editors: Diff View and Verification
By
–
Yes definitely! For coding, functionality like "diff view" in Cursor is a crappy example: green is add, red is delete. It's tapping into your visual cortex to (slightly) decrease verification time. But it's a very low bar of course. I always felt like I really wanted to lay out
-
AI-Assisted Coding Workflow: Structuring for Reduced Verification
By
–
Related tweet from earlier where I was describing my own (developing) workflow of "AI Assisted coding" where among other things I try really hard to structure it to decrease verification.
-
Generation vs Discrimination: The Verification Gap in Creation
By
–
Good post from @balajis on the "verification gap". You could see it as there being two modes in creation. Borrowing GAN terminology:
1) generation and
2) discrimination.
e.g. painting – you make a brush stroke (1) and then you look for a while to see if you improved the