The Bitter Lesson Has No Utility Function • Hacker News Board

Re: The Bitter Lesson Has No Utility Function

#11

Post by sshine » Thu, Mar 12, 2026, 9:12 PM UTC

isx726552 wrote: post #7

> Several commenters suggested the original essay was written by an LLM. They were half right. Both that essay and this one were written with Claude as a drafting partner. I directed the argument; the LLM helped with prose. That’s all well and good, but I think he needs to take a closer look at some of the resulting prose and clarify a little more. Most of it is good, but there are some unclear statements, like this…

I think you have a fair point, but you bothered to give feedback for how this article could be more congruent as if a non-assisted writing process would not warrant similar feedback.

The number of blog sentences that end abruptly halfway through have drastically fallen since I applied AI to my writing process.

There is nuance beyond the feeling that you know the author because he either uses the em dash a lot, the word “comprehensive” a lot, uses bullet points with bold text a lot, or writes detailed summaries reiterating what was just said. I read a lot from that author and I could use a break.

Re: The Bitter Lesson Has No Utility Function

#12

Post by satisfice » Fri, Mar 13, 2026, 8:51 AM UTC

“I directed the argument.”

The problem is I don’t know that, and I’m not convinced the author can know that either. Perhaps it influenced him in ways he is not aware.

My policy is I never let AI draft anything I write. It’s all my voice and my constructions. If I quote AI I will make that explicit. I do allow AI to review and complain about my work, but not to make any changes.

I don’t want anyone to be justified in questioning my authenticity the way I am justified in questioning the authenticity of this author.

Re: The Bitter Lesson Has No Utility Function

#13

Post by k__ » Fri, Mar 13, 2026, 9:47 AM UTC

satisfice wrote: post #12

“I directed the argument.” The problem is I don’t know that, and I’m not convinced the author can know that either. Perhaps it influenced him in ways he is not aware. My policy is I never let AI draft anything I write. It’s all my voice and my constructions. If I quote AI I will make that explicit. I do allow AI to review and complain about my work, but not to make any changes. I don’t want anyone to be justified in…

Yes, the issues is that LLMs sometimes fill semantic gaps.

When I write an article and I notice that there are steps missing for a conclusion then I do some research and fill them in.

These gaps often only show up when drafting, so if the LLM drafts, it might see a need to fill them in. You read the draft and see no steps are missing, but since you didn't do the research for the filled in steps, and they "look good to you", you might miss errors.

Re: The Bitter Lesson Has No Utility Function

#14

Post by satisfice » Fri, Mar 13, 2026, 12:51 PM UTC

k__ wrote: post #13

satisfice wrote: post #12

“I directed the argument.” The problem is I don’t know that, and I’m not convinced the author can know that either. Perhaps it influenced him in ways he is not aware. My policy is I never let AI draft anything I write. It’s all my voice and my constructions. If I quote AI I will make that explicit. I do allow AI to review and complain about my work, but not to make any changes. I don’t want anyone to be justified in…

Yes, the issues is that LLMs sometimes fill semantic gaps. When I write an article and I notice that there are steps missing for a conclusion then I do some research and fill them in. These gaps often only show up when drafting, so if the LLM drafts, it might see a need to fill them in. You read the draft and see no steps are missing, but since you didn't do the research for the filled in steps, and they "look good t…

good example!