Hacker News Reader

Comment by pclmulqdq | Hacker News Reader

Having seen others use GPT-4 for a bunch of business documents, I disagree. Its advantage is that it produces average-- (40-50th percentile) copy very quickly, but that quality degrades as you ask for documents that are more unique (ie not in the training set).

Incidentally, this makes a fine tuned version perfect for the government, which produces a shit load of low-effort documents, but today employs a bunch of morons to write them.

If you want a lot of text that you don't want to think about, LLMs are great. If you want text that is pithy or persuasive, it doesn't help much, even when you use prompting tricks (eg "you are a CEO/professor..."). By the way, this has convinced me that local LLAMA-scale models are the future, not massive remote GPTs.

View on HN

Implying all government employees are morons is pretty harsh. I know a lot of intelligent, driven people working in government or for government contractors. I know we think we're gods here for writing websites but a little humility goes a long way.

I know a lot of government employees who are smart and driven, too. They are not the people who write things like IRS audit letters, military after-action reports, and the like. Many government employees are barely literate, and told to spend all day writing stuff.

I'll have to agree to disagree. The government certainly employs a solid amount of stupid people, as does every institution, but I would quibble with the idea that writing reports as your main job function makes you more likely to be a barely literate moron.

I appreciate the rest of your post though and I think you make a great point about AI being very effective for certain types of work

For the record, I didn't say that writing reports all day makes you a barely literate moron. I said that a lot of barely literate morons are told to write reports all day - very different - and categorized the kind of reports that these folks often write. Lots of smart people also write reports all day at the government, and probably would also benefit from a tool that makes their writing a lot faster but worse, since the quality basically doesn't matter.

Do you know this for certain? As in, have you done the analysis to prove this? Collected all the IQ and other aptitude tests of these people to conclude this for certain?

Some of them could be in the beginning part of their government career attending school. Did you run a survey to ask these people what they do with their lives?

Yeah, I'm sure it depends on the training and the document type. I haven't had direct experience with LLM-generated business documents, but I do think that, given a small time window, GPT-4 is more effective at communicating ideas than the average human even if those ideas are wrong. As far as my use goes, GPT-4 now serves the role that Wikipedia used to, which is to give me a baseline of understanding that I use to refine my research, given that most internet content is either verbose, hogwash, or both. Ironically, I've found that Wikipedia has made itself nearly obsolete in this regard given how seemingly every article has become so academically-written that it's an undertaking to just get a rough idea of something. Neither source I trust, but at least GPT-4 will introduce me to a topic in a way that is more effective and concise than most human writing today. But that's just my experience.

> If you want a lot of text that you don't want to think about, LLMs are great. If you want text that is pithy or persuasive, it doesn't help much, even when you use prompting tricks (eg "you are a CEO/professor..."). By the way, this has convinced me that local LLAMA-scale models are the future, not massive remote GPTs.

Granted, we are judging LLMs based on highly generalized training sets. What if a GPT was fine-tuned on all of the writing and speeches of figures whom are considered the most persuasive?

A bleak Byzantine future, of endless bureaucracy.

Government reports continue to balloon in size, because of technological advancements (especially copy/paste).

In the age of typewriters, anything not important, or duplicate, was cut. With word processors, document sizes unnecessarily grew.

Reply on news.ycombinator.com