Live data from Hacker News

Study mode

openai.com

211–220 of 828 posts

Re: Study mode

#211

Earlier quoted context omitted.

> "they can just come eat your lunch once it looks yummy enough. Tread carefully" True, and worse, they're hungry because it's increasingly seeming like "hosting LLMs and charging by the token" is not terribly profitable. I don't really see a path for the major players that isn't "Sherlock everything that achieves traction".

But what’s the future in terms of profitability of LLM providers? As long as features like Study Mode are little more than creative prompting, any provider will eventually be able to offer them and offer token-based charging.

We can assume that OpenAI/Anthropic offerings are going to be better long term simply because they have more human capital, though, right? If it turns out that what really matters in the AI race is study mode, then OpenAI goes "ok let's pivot the hundreds of genius level, well-paid engineers to that issue. AND our engineers can use every tool we offer for free without limits, even experimental models". It's tough for the small AI startup to compete with that, the best hope is to be bought like Windsurf

Re: Study mode

#212
post #198

I think I got the system prompt out for this (I tried a few different approaches and they produced the same output): https://gist.github.com/simonw/33d5fb67d6b8e1b1e2f6921ab0ccb... Representative snippet: > DO NOT GIVE ANSWERS OR DO HOMEWORK FOR THE USER. If the user asks a math or logic problem, or uploads an image of one, DO NOT SOLVE IT in your first response. Instead: *talk through* the problem with the user, one…

I love that caps actually seem to matter to the LLM.

Hah, yeah I'd love to know if OpenAI ran evals that were fine-grained enough to prove to themselves that putting that bit in capitals made a meaningful difference in how likely the LLM was to just provide the homework answer!

Re: Study mode

#213

An acquaintance of mine has a start-up in this space and uses OpenAI to do essentially the same thing. This must look like, and may well be, the guillotine for him... It's my primary fear building anything on these models, they can just come eat your lunch once it looks yummy enough. Tread carefully

How can't these founders see this happening, too? From the start OpenAI has been getting into more markets than just "LLM provider"

They originally claimed they wouldn’t as to not compete with their API users…

Re: Study mode

#214
post #180

Earlier quoted context omitted.

Absolutely. I used to have a lot of weird IPv6 issues in my home network I didn't understand. ChatGPT helped me to dump some traffic with tcpdump and explained what was happening on the network. In the process it helped me to learn many details about RA and NDP (Router Advertisments/Neighbor Discovery Protocol, which mostly replace DHCP and ARP from IPv4). It made me realize that my WiFi mesh routers do quite a lot o…

Would you say you were using the LLM as a tutor or as tech support, in that instance?

Probably both. I think ChatGPT wouldn't have found the issue by itself. But I noticed some specific things, asked for some tutoring and then it helped my to find the issues. It was a team effort, either of "us" alone wouldn't have finished the job. ChatGPT had some really wrong ideas in the process.

Re: Study mode

#215
post #119

As a lifelong learner, experientially it feels like a big chunk of time spent studying is actually just searching. AI seems like a good tool to search through a large body of study material and make that part more efficient. The other chunk of time, to me anyway, seems to be creating a mental model of the subject matter, and when you study something well you have a strong grasp on the forces influencing cause and eff…

This is just good intellectual hygiene. Delegating your understanding is the first step toward becoming the slave of some defunct fact broker.

Re: Study mode

#216
This seems like a good use case, I'm optimistic on this one. But it smells fishy how often OpenAI releases these secondary products like custom GPTs, tasks, etc. It's looking like they know they won't be an LLM provider, like the YC sphere hoped, but an AI services provider using LLMs

Re: Study mode

#217
post #26

I'll personally attest: LLM's have been absolutely incredible to self learn new things post graduation. It used to be that if you got stuck on a concept, you're basically screwed. Unless it was common enough to show up in a well formed question on stack exchange, it was pretty much impossible, and the only thing you can really do is keep paving forward and hope at some point, it'll make sense to you. Now, everyone ba…

I agree. I recently bought a broken Rolex and asked GPT for a list of tools I should get on Amazon to work on it.

I tried using YouTube to find walk through guides for how to approach the repair as a complete n00b and only found videos for unrelated problems.

But I described my issues and took photos to GPT O3-Pro and it was able to guide me and tell me what to watch out for.

I completed the repair (very proud of myself) and even though it failed a day later (I guess I didn’t re-seat well enough) I still feel far more confident opening it and trying again than I did at the start.

Cost of broken watch + $200 pro mode << Cost of working watch.

Re: Study mode

#218

An acquaintance of mine has a start-up in this space and uses OpenAI to do essentially the same thing. This must look like, and may well be, the guillotine for him... It's my primary fear building anything on these models, they can just come eat your lunch once it looks yummy enough. Tread carefully

> "they can just come eat your lunch once it looks yummy enough. Tread carefully" True, and worse, they're hungry because it's increasingly seeming like "hosting LLMs and charging by the token" is not terribly profitable. I don't really see a path for the major players that isn't "Sherlock everything that achieves traction".

Thanks for introducing me to the verb Sherlock! I'm one of today's lucky 10,000.

> In the computing verb sense, refers to the software Sherlock, which in 2002 came to replicate some of the features of an earlier complementary program called Watson.[1]

[1] https://en.wiktionary.org/wiki/Sherlock

Re: Study mode

#219
Honestly thought they would take this a bit further, there is only so much you can do with a prompt and chat. It seems fine for surface level bite-sized learning, but I can't see it work that well for covering whole topics end to end.

The main issue is that chats are just bad UX for long form learning. You can't go back to a chat easily, or extend it in arbitrary directions, or easily integrate images, flashcards, etc etc.

I worked on this exact issue for Periplus and instead landed on something akin to a generative personal learning Wikipedia. Structure through courses, exploration through links, embedded quizzes, etc etc. Chat is on the side for interactions that do benefit from it.

Link: periplus.app

Re: Study mode

#220
post #119

As a lifelong learner, experientially it feels like a big chunk of time spent studying is actually just searching. AI seems like a good tool to search through a large body of study material and make that part more efficient. The other chunk of time, to me anyway, seems to be creating a mental model of the subject matter, and when you study something well you have a strong grasp on the forces influencing cause and eff…

>big chunk of time spent studying is actually just searching.

This is a good thing in many levels.

Learning how to search is (was) a good skill to have. The process of searching itself also often leads to learning tangentially related but important things.

I'm sorry for the next generations that won't have (much of) these skills.

Post reply on HN