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

Study mode

openai.com

Post reply on HN
Print view View original

451–460 of 828 posts

Re: Study mode

#451

Post by draebek »

sebzim4500 wrote: post #223
Workaccount2 wrote: post #65

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

I'm too young to have experienced this, but I'm sure others here aren't. During the early days of tech, was there prevailing wisdom that software companies would never be able to compete with hardware companies because the hardware companies would always be able to copy them and ship the software with the hardware? Because I think it's basically the analogous situation. People assume that the foundation model provide…

Does https://en.wikipedia.org/wiki/Sherlock_(software)#Sherlocked... count? (Edit: Missed I wasn't the first to post this in a sibling.)

Re: Study mode

#452

Post by no_wizard »

jlebar wrote: post #244
no_wizard wrote: post #49

Earlier quoted context omitted.

>Now, everyone basically has a personal TA, ready to go at all hours of the day This simply hasn't been my experience. Its too shallow. The deeper I go, the less it seems to be useful. This happens quick for me. Also, god forbid you're researching a complex and possibly controversial subject and you want it to find reputable sources or particularly academic ones.

> Its too shallow. The deeper I go, the less it seems to be useful. This happens quick for me. You must be using a free model like GPT-4o (or the equivalent from another provider)? I find that o3 is consistently able to go deeper than me in anything I'm a nonexpert in, and usually can keep up with me in those areas where I am an expert. If that's not the case for you I'd be very curious to see a full conversation tra…

I have access to the highest tier paid versions of ChatGPT and Google Gemini, I've tried different models, tuning things like size of context windows etc.

I know it has nothing to do with this. I simply hit a wall eventually.

I unfortunately am not at liberty to share the chats though. They're work related (I very recently ended up at a place where we do thorny research).

A simple one though, is researching Israel - Palestine relations since 1948. It starts off okay (usually) but it goes off the rails eventually with bad sourcing, fictitious sourcing, and/or hallucinations. Sometimes I actually hit a wall where it repeats itself over and over and I suspect its because the information is simply not captured by the model.

FWIW, if these models had live & historic access to Reuters and Bloomberg terminals I think they might be better at a range of tasks I find them inadequate for, maybe.

Re: Study mode

#453

Post by kristofferR »

ramraj07 wrote: post #441
romaniitedomum wrote: post #412

Earlier quoted context omitted.

> Learning something online 5 years ago often involved trawling incorrect, outdated or hostile content and attempting to piece together mental models without the chance to receive immediate feedback on intuition or ask follow up questions. This is leaps and bounds ahead of that experience. But now, you're wondering if the answer the AI gave you is correct or something it hallucinated. Every time I find myself putting…

What exactly did 2025 AI hallucinate for you? The last time I've seen a hallucination from these things was a year ago. For questions that a kid or a student is going to answer im not sure any reasonable person should be worried about this.

Nah it's not that rare.

This is one I got today:

https://chatgpt.com/share/6889605f-58f8-8011-910b-300209a521...

(image I uploaded: http://img.nrk.no/img/534001.jpeg)

The correct answer would have been Skarpenords Bastion/kruttårn.

Re: Study mode

#454

Post by mediaman »

breve wrote: post #443
jacobedawson wrote: post #379

An underrated quality of LLMs as study partner is that you can ask "stupid" questions without fear of embarrassment. Adding in a mode that doesn't just dump an answer but works to take you through the material step-by-step is magical. A tireless, capable, well-versed assistant on call 24/7 is an autodidact's dream. I'm puzzled (but not surprised) by the standard HN resistance & skepticism. Learning something online 5…

> Adding in a mode that doesn't just dump an answer but works to take you through the material step-by-step is magical Except these systems will still confidently lie to you. The other day I noticed that DuckDuckGo has an Easter egg where it will change its logo based on what you've searched for. If you search for James Bond or Indiana Jones or Darth Vader or Shrek or Jack Sparrow, the logo will change to a version b…

This is endlessly brought up as if the human operating the tool is an idiot.

I agree that if the user is incompetent, cannot learn, and cannot learn to use a tool, then they're going to make a lot of mistakes from using GPTs.

Yes, there are limitations to using GPTs. They are pre-trained, so of course they're not going to know about some easter egg in DDG. They are not an oracle. There is indeed skill to using them.

They are not magic, so if that is the bar we expect them to hit, we will be disappointed.

But neither are they useless, and it seems we constantly talk past one another because one side insists they're magic silicon gods, while the other says they're worthless because they are far short of that bar.

Re: Study mode

#455

Post by breve »

mediaman wrote: post #454
breve wrote: post #443

Earlier quoted context omitted.

> Adding in a mode that doesn't just dump an answer but works to take you through the material step-by-step is magical Except these systems will still confidently lie to you. The other day I noticed that DuckDuckGo has an Easter egg where it will change its logo based on what you've searched for. If you search for James Bond or Indiana Jones or Darth Vader or Shrek or Jack Sparrow, the logo will change to a version b…

This is endlessly brought up as if the human operating the tool is an idiot. I agree that if the user is incompetent, cannot learn, and cannot learn to use a tool, then they're going to make a lot of mistakes from using GPTs. Yes, there are limitations to using GPTs. They are pre-trained, so of course they're not going to know about some easter egg in DDG. They are not an oracle. There is indeed skill to using them.…

The ability to say "I don't know" is not a high bar. I would say it's a basic requirement of a system that is not magic.

Re: Study mode

#456

Post by tekno45 »

The same people who think this is the ultimate teacher will also be harassing scientists with their AI assisted theories and demand the scientific community take them seriously when they have pages of gibberish they expect to be rigorously debated

Re: Study mode

#457

Post by no_wizard »

smokel wrote: post #293
no_wizard wrote: post #75

Earlier quoted context omitted.

>Asking the LLM is a vastly superior experience. Not to be overly argumentative, but I disagree, if you're looking for a deep and ongoing process, LLMs fall down, because they can't remember anything and can't build upon itself in that way. You end up having to repeat alot of stuff. They also don't have good course correction (that is, if you're going down the wrong path, it doesn't alert you, as I've experienced) It…

Most LLM user interfaces, such as ChatGPT, do have a memory. See Settings, Personalization, Manage Memories .

Sure, but there are limits here. Thats what I'm talking about, limits. The memory isn't infinitely expansive. I still have found it doesn't backtrack well enough to "remember" (for lack of a better term) that it told me something already, if its old enough, for example.

It also doesn't seem to do a good job of building on "memory" over time. There appears to be some unspoken limit there, or something to that affect.

Re: Study mode

#458

Post by r3trohack3r »

tekno45 wrote: post #444
ramraj07 wrote: post #441

Earlier quoted context omitted.

What exactly did 2025 AI hallucinate for you? The last time I've seen a hallucination from these things was a year ago. For questions that a kid or a student is going to answer im not sure any reasonable person should be worried about this.

How do you know? its literally non-deterministic.

Most (all?) AI models I work with are literally deterministic. If you give it the same exact input, you get the same exact output every single time.

What most people call “non-deterministic” in AI is that one of those inputs is a _seed_ that is sourced from a PRNG because getting a different answer every time is considered a feature for most use cases.

Edit: I’m trying to imagine how you could get a non-deterministic AI and I’m struggling because the entire thing is built on a series of deterministic steps. The only way you can make it look non-deterministic is to hide part of the input from the user.

Re: Study mode

#459

Post by loloquwowndueo »

BDPW wrote: post #362
loloquwowndueo wrote: post #256

Earlier quoted context omitted.

> 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’s called basic research skills - don’t they teach this anymore in high school, let alone college? How ever did we get by with nothing but an encyclopedia or a library catalog?

Its a little disingenuous to say that, most of us would have never gotten by with literally just a library catalog and encyclopedia. Needing a community to learn something in is needed to learn almost anything difficult and this has always been the case. That's not just about fundamentally difficult problems but also about simple misunderstandings. If you don't have access to a community like that learning stuff in a…

Wait until you waste days down a hallucination-induced LLM rabbit hole.

> most of us would have never gotten by with literally just a library catalog and encyclopedia.

I meant the opposite, perhaps I phrased it poorly. Back in the day we would get by and learn new shit by looking for books on the topic and reading them (they have useful indices and tables of contents to zero in on what you need and not have to read the entire book). An encyclopedia was (is? Wikipedia anyone?) a good way to get an overview of a topic and the basics before diving into a more specialized book.

Re: Study mode

#460

Post by no_wizard »

epolanski wrote: post #261
no_wizard wrote: post #49

Earlier quoted context omitted.

>Now, everyone basically has a personal TA, ready to go at all hours of the day This simply hasn't been my experience. Its too shallow. The deeper I go, the less it seems to be useful. This happens quick for me. Also, god forbid you're researching a complex and possibly controversial subject and you want it to find reputable sources or particularly academic ones.

I really think that 90% of such comments come from a lack of knowledge on how to use LLMs for research. It's not a criticism, the landscape moves fast and it takes time to master and personalize a flow to use an LLM as a research assistant. Start with something such as NotebookLM.

I use them and stay up to date reasonably. I have used NotebookLM, I have access to advanced models through my employer and personally, and I have done alot of research on LLMs and using them effectively.

They simply have limitations, especially on deep pointed subject matters where you want depth not breadth, and honestly I'm not sure why these limitations exist but I'm not working directly on these systems.

Talk to Gemini or ChatGPT about mental health things, thats a good example of what I'm talking about. As recently as two weeks ago my colleagues found that even when heavily tuned, they still managed to become 'pro suicide' if given certain lines of questioning.

Post reply on HN