301–310 of 442 posts
This is precisely why I have said that every new framework/library should have a markdown or text or whatever is the best format for LLM models endpoint that has all the docs and examples in one single page so you can easily copy it over to a models context. You want to make it as easy as possible for LLMs to be aware of how your software works. The fancy nested navigation guide walkthrough thing is cool for users bu…
Earlier quoted context omitted.
> A world where e.g. Django and React are considered as obvious choices for backend and frontend as git is for version control sounds like a world where high quality web apps become much cheaper to build. Imagine you saying this twenty years ago. Would you still want to be writing your back-end in JavaBeans, your front end in VisualBASIC, and storing your data in Subversion?
VisualBASIC made much nicer (and more responsive) UIs than Electron, so this isn't the slam dunk you think it is.
> OpenAI’s latest models have cutoffs of late 2023. The first paragraph is factually incorrect; the cutoff is June 2024 for 4o. Awww, no more new JavaScript frameworks and waiting only for established technologies to cut through the noise. I don't see that as a bad thing. Technologies need to mature, and maintaining API backward compatibility is another advantage.
AI may be exaggerating this issue, but it's always existed. New tech has an inherent disadvantage vs legacy tech, because there's more built-up knowledge. If you choose React, you have better online resources (official docs, tutorials, answers to common pitfalls), more trust (it won't ship bugs or be abandoned), great third-party helper libraries, built-in IDE integration, and a large pool of employees with experienc…
Even for those of us who use mostly stack overflow/google, it's much cheaper to wait on someone else to run into your problem and document the solution than to be first into the fire. We've relied on this strategy for a couple of decades now.
I don't think the OP has demonstrated that adoption rates for new tech have changed in any way since AI.
> Also, popular frameworks usually have better code, because they have years of bug-fixes from being tested on many production servers, and the API has been tailored from real-world experience.
Overall I am very resistant to the idea that popular==good. I'd say popular==more popular. Also I think there's often a point where feeping creaturism results in tools that are overcomplicated, prone to security bugs and no longer easy to use.
Earlier quoted context omitted.
LLMs work best matching patterns. If 1k loc matches patterns and the 10 loc doesn’t, it’s a problem. The only thing the OP is missing which combines the best of both worlds is to always put source of and/or docs for his abstractions into the context window of the LLM.
If your abstractions match common design patterns then you've solved your problem. It's ridiculous to assume that an LLM will understand 1k LOC of standard library code better than 10 lines of a custom abstraction which uses a common design pattern. It's more prone to hallucinating things if your custom abstraction is not super standard but at least you'd be able to check its mistakes (you're checking the code genera…
I’ll reiterate what I said before: put the whole source of the new library in the context window and tell the LLM to use it. It will, at least if it’s Claude.
I might be willing to use a SAT solver or linear algebra on it if I ever get to that point but there’s a lot else to do first. The problem space involves humans, so optimizing that can very quickly turn into “works in theory but not in practice”. It’d be the sort of thing where you use it but don’t brag about it.
No, AI isn't. Any new tech, or version upgrade, or whatever, takes time for people to become familiar with it. You might as well say "Stack Overflow is stifling new tech adoption" because brand-new stuff doesn't have many Q's and A's yet. But that would be a silly thing to say. I'm not going to adopt a brand-new database regardless of LLM training data cutoff, just because enough people haven't had enough experience…
Strong disagree here. I've been trying to learn Zig, and I'm thwarted enough by chatgpt giving me outdated information on Zig's unstable API that if I didn't have a strong incentive to learn it for it's cross compiler, I'd likely turn my efforts towards another language. This effect can greatly alter the adoption curve of a new tech, which can leave it dying on the vine. You're not wrong though, in that Stack Overflo…
The experience is night and day and it’s the only reason I pay for Cursor on top of OpenAI/Anthropic. The answers it gives are much more accurate when including documentation and it helps a lot with exploring APIs. Probably helps a lot with the code generation too, but I don’t use Composer as much.
Earlier quoted context omitted.
> Updates to GPT-4o in ChatGPT (January 29, 2025) > By extending its training data cutoff from November 2023 to June 2024 […] https://help.openai.com/en/articles/9624314-model-release-no...
Thank you. I'll make a correction next I can. I do wonder why this information is lacking from the platform docs though. They specifically mention a model that is the "GPT-4o used in ChatGPT".