Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I'd rather read the prompt

claytonwramsey.com

Post reply on HN
Print view View original

681–690 of 880 posts

Re: I'd rather read the prompt

#681

Post by indymike »

jameshart wrote: post #329
sn9 wrote: post #221

> I should hope that the purpose of a class writing exercise is not to create an artifact of text but force the student to think; a language model produces the former, not the latter. It's been incredibly blackpilling seeing how many intelligent professionals and academics don't understand this, especially in education and academia. They see work as the mere production of output, without ever thinking about how that…

I’ve come across this analogy that I think works well: Using an LLM to do schoolwork is like taking a forklift to the gym. If all we were interested in was moving the weights around, you’d be right to use a tool to help you. But we’re doing this work for the effect it will have on you . The reason a teacher asks you a question is not because they don’t know the answer.

> Using an LLM to do schoolwork is like taking a forklift to the gym.

I'm sure the time has come for college students to master using LLMs. It's just as important as grammar or basic math now. The software I build (and the entire tech industry) automates huge swaths of business processes with AI. Students need to be able to understand, work with, and manage swarms of AI agents doing work.

To stick to the analogy:

I need skilled forklift drivers, not big buff workers like I used to.

Re: I'd rather read the prompt

#682

Post by ebiester »

bumby wrote: post #303
ebiester wrote: post #276

Earlier quoted context omitted.

This seems to be the general feeling of students right now. Academia put itself as a gateway and barrier to the middle class. Why would we be surprised when people with no interest in anything but the goal are not enthralled by the process?

Academia clearly lost their monopoly on information. Since their last moat is a monopoly on credentials, I expect them to defend it intensely. We could make it less meaningful if employers weren’t so keen on using credentials as their own gateway. That may have more of a chance of happening if the OPs perspective becomes more prevalent and the credential becomes an increasingly worse signal for meaningful skills.

They didn't have a monopoly on information anytime in the 20th century. If you wanted to have all the knowledge, it could be expensive to acquire a personal library, but nothing was stopping you from acquiring the same textbooks. Heck, until the information age it was pretty easy to forge a credential.

But it wasn't about credentials even. It was about inculcating a culture. You knew that someone had the knowledge and ability to reach university, and you knew that they had a shared common culture with you. Shared common culture and norms increases trust. Credentials mattered for doctors, but universities, in the end, were selling something far more intangible: culture.

Re: I'd rather read the prompt

#683

Post by indymike »

sn9 wrote: post #221

> I should hope that the purpose of a class writing exercise is not to create an artifact of text but force the student to think; a language model produces the former, not the latter. It's been incredibly blackpilling seeing how many intelligent professionals and academics don't understand this, especially in education and academia. They see work as the mere production of output, without ever thinking about how that…

> Students who know least of all and don't understand the purpose of writing or problem solving or the limitations of LLMs are currently wasting years of their lives letting LLMs pull them along as they cheat themselves out of an education

My high-school age daughter told me how her small private school solved this problem:

They brought back oral exams.

There aren't a lot of other good options. Written take-home work and online tests have always been fertile ground for cheating. Another benefit of oral exams: you learn to communicate under stress.

Re: I'd rather read the prompt

#684

Post by bobdosherman »

Writing sharpens your thoughts. Working math problems sharpens your ability to do math. In the context of the education system though, where grades are a signal of future ability, there's a strong incentive to engage in rent-seeking by either searching for a solutions manual or "refining" an LLM's output. The little I've done when I've taught math-based econ is to make it clear that in-class tests have very high weight on your final grade, and out-of-class problem sets have very low weight. I can only mouth words as to why it's vital for students to struggle independently on the problem sets as a tool for learning.

Re: I'd rather read the prompt

#685

Post by phamilton »

walleeee wrote: post #655
phamilton wrote: post #458

Earlier quoted context omitted.

If we can detach content and presentation, then the reader can choose tone and length. At some point we will stop making decisions about what future readers want. We will just capture the concrete inputs and the reader's LLM will explain it.

I don't think form and function can be separated so cleanly in natural language. However you encode what's between your ears into text, you've made (re)presentational choices. A piece of text does not have a single inherently correct interpretation. Its meaning is a relation constructed at run- (i.e. read-)time between the reader, the writer, and (possibly) the things the text refers to, that is if both sides are wel…

My LLM workflow involves a back-and-forth clarification (verbose input -> LLM asks questions) that results in a rich context representing my intent. Generating comments/docs from this feels lossy.

What if we could persist this final LLM context state? Think of it like a queryable snapshot of the 'why' behind code or docs. Instead of just reading a comment, you could load the associated context and ask an LLM questions based on the original author's reasoning state.

Yes, context is model-specific, a major hurdle. And there are many other challenges. But ignoring the technical implementation for a moment, is this concept – capturing intent via persistent, queryable LLM context – a valuable problem to solve? I feel it would be.

Re: I'd rather read the prompt

#686

Post by hk__2 »

altilunium wrote: post #674
bost-ty wrote: post #4

I like the author's take: it isn't a value judgement on the individual using ChatGPT (or Gemini or whichever LLM you like this week), it's that the thought that went into making the prompt is, inevitably, more interesting/original/human than the output the LLM generates afterwards. In my experiments with LLMs for writing code, I find that the code is objectively garbage if my prompt is garbage. If I don't know what I…

Sometimes, good writing is like an NP-complete problem, hard to create, but easy to verify. If you have enough skill to distinguish good output from garbage, you can produce reasonably good results.

> Sometimes, good writing is like an NP-complete problem, hard to create, but easy to verify.

Doesn’t this match pretty much all human creation? It’s easier to judge a book that to write it, it’s easier to watch a rocket going up in the space than to build it, it’s easier to appreciate some Renaissance painting or sculpture than to actually make it.

Re: I'd rather read the prompt

#687

Post by AppleBananaPie »

gnatolf wrote: post #279
egglemonsoup wrote: post #203

Earlier quoted context omitted.

FWIW: Your original comment, in the first message you sent ChatGPT, was way better than the one you posted. Simple, authentic, to the point

I couldn't agree more, this 'polished' style the finished comment comes in is super boring to read. It's hard to put the finger on it, but overall flow is just too... Samesame? I guess it's perfectly _expected_ to be predictable to read ;)

And then one of the iterations was asking additional ways llms could be used and then adding some of those as content which seems odd but plausibly helpful brainstorming. Just the phrasing of the original question makes it sound like things the user isn't actually doing but wants in their comment if that makes sense.

Thanks for the example chat it was a valuable learning for me!

Re: I'd rather read the prompt

#688

Post by dennis_jeeves2 »

chongli wrote: post #353
palata wrote: post #275

Earlier quoted context omitted.

> I'm there for the degree. If I wanted to learn and engage with material, I could save $60,000 I would argue that if it costs $60,000, both your education system and the recruitment in those companies that require this degree are broken. It's not the case in all countries though. Not that it is your fault, just stating the obvious.

It is broken. For every coveted job there are thousands of applicants. Employers will accept any signal that reliably predicts a modicum of intelligence, conscientiousness, and agreeability. University degrees cover all three. But that's just the job market. The other elephants in the room are inflation and the housing market. People who don't have top-notch jobs (that require degrees) can't afford to buy a house. Th…

>The other elephants in the room are inflation and the housing market.

I'm not even sure if people are aware of inflation/housing as a completely solvable issue by the govt. I guess it's because people most people are clueless on how it is to be solved.

>We are a society of ladder-pullers.

It's by design, to serve the rulers. It's an assembly line of slaves who are given some freedoms and are put through various stages of school, university, work and retirement. When most people retire they are left with little to nothing.

Re: I'd rather read the prompt

#689

Post by necovek »

_AzMoo wrote: post #302
derefr wrote: post #187

Earlier quoted context omitted.

There is one other very useful form of "expansion" that LLMs do. If you aren't aware: (high-parameter-count) LLMs can be used pretty reliably to teach yourself things. LLM base models "know things" to about the same degree that the Internet itself "knows" those things. For well-understood topics — i.e. subjects where the Internet contains all sorts of open-source textbooks and treatments of the subject — LLMs really…

I use LLMs a huge amount in my work as a senior software engineer to flesh out the background information required to make my actual contributions understandable to those without the same background as me. eg, if I want to write a proposal on using SLO's and error budgets to make data driven decisions about which errors need addressing and which don't, inside a hybrid kubernetes and serverless environment, I could do…

For a teacher, they want to see you spend those 3h to see what you come up with, and if there's something they should direct your attention to, or something they should change in their instruction.

But ultimately, getting the concise summary for a complex topic (like SLIs and SLOs are) is brilliant, but would be even better if it was full of back-links to deeper dives around the Internet and the SRE book.

Re: I'd rather read the prompt

#690

Post by chipsrafferty »

animal_spirits wrote: post #424
rohansood15 wrote: post #400

Earlier quoted context omitted.

Students want the diploma because it has value. It has value because a student can only get it by learning and problem-solving. If students find a way to get a diploma without doing the work, it will soon be worth less than the paper on which it is printed.

Further, if a student can get a diploma without work, then the diploma does not have value anymore. If diplomas are no longer valuable, the signal they provide in the labor market will turn into noise. If employers no longer look for the diploma-signal in an employee, what will be the reason an employer will hire an employee? I think this story will become true, and society will radically shift into one where critica…

Why wouldn't the shift to be to raise the standards of the schools?
Post reply on HN