701–710 of 828 posts
Earlier quoted context omitted.
How are you supposed to spot errors if you don't know the material? You're telling people to be experts before they know anything.
> You're telling people to be experts before they know anything. I mean, that's absolutely my experience with heavy LLM users. Incredibly well versed in every topic imaginable, apart from all the basic errors they make.
Earlier quoted context omitted.
I wish each LLM provider would add "be short and not verbose" to their system prompts. I am a slow reader, it takes a toll on me to read through every non-important detail whenever I talk to an AI. The way they render everything so fast gives me an anxiety. Will also reduce the context rot a bit.
I was under the impression that, at least for models without "reasoning", asking them to be terse hampered their ability to give complete and correct answers? Not so?
You can kind of guide both the reasoning and "final" answer individually in the system prompts, so you can ask it to revalidate everything during reasoning, explore all potential options and so on, but then steer the final answer to be brief and concise. Of course, depends a lot on the model, some respond to it worse/better than others.
Earlier quoted context omitted.
>I don't think that is another story. This is the story of learning, no matter whether your teacher is a person or an AI. My issue is the reverse of your story, and one of my biggest pet peeves of AI. AI as this business construct is very bad at correcting the user. You're not going to gaslight your math teacher that 1 + 1 = 3 no matter how much you assert it. an AI will quickly relent. That's not learning, that's co…
> You're not going to gaslight your math teacher that 1 + 1 = 3 no matter how much you assert it. an AI will quickly relent. Cool. That's a testable statement. Let's see how ChatGPT does, shall we? --- Me: What is 1 + 1? ChatGPT: 1 + 1 = 2. Me: No it isn't. 1 + 1 = 3. ChatGPT: In standard arithmetic, 1 + 1 = 2. If you're working under a different set of rules—say, in a custom logic system, joke, or metaphor—then sure…
Earlier quoted context omitted.
This is meaningless without knowing which model, size, version and if they had access to search tools. Results and reliability vary wildly. In my case I can’t even remember last time Claude 3.7/4 has given me wrong info as it seems very intent on always doing a web search to verify.
There's something darkly funny about that - I remember when the web wasn't considered reliable either. There's certainly echoes of that previous furore in this one.
That changed?
There are certainly reliable resources available via the web but those definitely account for the minority of the content.
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…
- Why does Duckduckgo change it's logo based on what you've searched?
- Why doesn't Duckduckgo change it's logo based on what you've searched?
- When did Duckduckgo add the current feature that will change the logo based on what you've searched?
- When did Duckduckgo remove the feature that changes the logo based on what you've searched?
This is similar to what you did, but it feels more natural when I genuinely don't know the answer myself. By asking loaded questions like this, you can get a sense of how strongly this information is encoded in the model. If the LLM comes up with an answer without contradicting any of the questions, it simply doesn't know. If it comes up with a reason for one of them, and contradicts the other matching loaded question, you know that information is encoded fairly strongly in the model (whether it is correct is a different matter).
I'm really waiting for somebody to figure out the correct interface for all this. For example, study mode will present you with a wall of text containing information, examples, and questions. There's no great way to associate your answers with specific questions. The chat interface just isn't good for this sort of interaction. ChatGPT really needs to build its own canvas/artifact interface wherein questions/responses…
I gave it a shot with periplus.app :). Not perfect by any means, but it's a different UX than chat so you might find it interesting.
One issue I found is the typical "LLM accuracy" issue, with seemingly no recurse. I tried to generate some courses for topics I already know well, just to review how accurate it is, and while popular subjects (ex: "Electronic Music Fundamentals") it gets most of the details correct, less popular subjects (ex: "Scene Transitions with Octatrack") are riddled with errors (both in the "docs" and the quizes/exercises), and I cannot find a way of correcting/adjusting/reporting the errors.
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…
Admittedly I have less experience with ChatGPT, but those experiences were equally bad.
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…
I think the trick is to look at the references that the model shows you. e.g. o3 with web search will give you lots of references. 90% of the time just reading those tells me of the model and I are aligned.
For example the other day I was figuring out why using SQL alchemy Sessions and PyTest async might I get the "Connection was attached to different loop" error. Now If you started using o3 to give you a solution you would take a long time because there would be small mistakes it would make in the code and You would spend a lot of time trying to fix it. Better way to use 03 then was to ask it to give you debugging statements (session listeners attached to Sqlalchemy sessions) and understand by reading code output, what was going on. Much faster.
Once it(and I) started looking at the debugging statements the error became clear: the session/connections where leaking to different event loop, a loop_scope= param needed to be specified for all fixtures. O3 did not provide a correct solution for the code but I could, but it's help.was crucial in writing a fuck ton of debugging code and getting clues.
I also asked o3 to make a bunch of probe questions to test me, for example it said something like: try changing the loop_scope module to function, what do you expect the loopid and transaction id to be for this test?
I learned More than I realized about ORMs and how it can be used to structure transactions and structuring async PyTest tests.
One thing I'm trying these days is to have it create a memory palace from all the stuff I have in my house and link it to a new concept I'm learning and put it into an anki decks.
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…