Hacker News Reader

Comment by lackoftactics | Hacker News Reader

> 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.

View on HN

> 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.

I think this kind of discussion is immature and downplays the point of the article.

A good example of this that I just encountered: Rust. Just asked Claude/ChatGPT for rust stuff recently, and it still gives a lot old/depreciated methods for a lot of things. This has been the case for Godot 3 vs 4 as well.

Same here. In my TypeScript code, Cursor/Claude seem quite fluent. In my Rust code, I often turn Cursor Tab off because it’s just suggesting nonsense.

It does not really matter as even though the models get updated, the new data was produced with the help of the older models, it is feeding on itself.

Just imagine how hard it would be to push a new programming language. No AI models would be able to generate code in that new language, or they would be extremely limited. This would make adoption much more difficult in a world where all developers use AI tooling extensively.

I believe this trend could create new opportunities also: as everyone uses AI tools to generate statistically average quality code, only those not using AI tools will be able to create true innovation.

In some sense I am hopeful that AI will be able to just write everything directly in binary. Everything written ideally, with no abstraction, fluff or bumpers for human brains. Computers don't talk in any high level programming language, they talk in binary. If anything we should probably be focusing LLMs on getting good at that.

I can only imagine that the amount of energy wasted on CPU cycles from layers of bloated programming languages makes stuff like bitcoin mining look like a rounding error.

Not sure that’s always a good thing - see the occasionally erratic behavior of Tesla’s autopilot. It directly speaks the language of the systems it connects with, and also occasionally steers into dividers, killing people - and nobody knows why. We need to be able to verify correctness of what the models generate.

So you can't even debug or check the LLM's output for correctness? Good luck with that.

If anything, the best use for LLMs is to produce code in very strong languages like Lean, where correctness of the code can be established. Just trusting machine code output would be irresponsible.

Author here. May I request a source for that?

Platform docs state:

> The knowledge cutoff for GPT-4o models is October, 2023.

https://platform.openai.com/docs/models#gpt-4o

> 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".

It seems a bug to me. If content in the doc does not match the actual behavior, either the doc is outdated, or the software has a bug. Which is a bug either way.

What if you are using boring technology, but surprise, it or some of its libraries got updated? React is on version 19. Show some imagination.

I hear you, and I’ve run into similar issues where AI falls short—take aerospace documentation systems or tools like MCP (tiling manager) as prime examples. That said, framing this as a stifling of technology feels like a leap to me. The hypothesis strikes me as anecdotal at best. For instance, I don’t see coworkers abandoning libraries just because LLMs can’t assist—it’s not that hard to dive into the library code yourself, or even attach docs in Cursor if clarity’s a priority. If anything, the friction here feels minor, not existential. To claim AI is holding progress back? That argument needs more robust evidence, in my opinion.

The problem is acute with APIs that move fast and deprecate (Shopify and some of the Google ones)

I can't wait to see how these AI models maintain backward compatibility

Reply on news.ycombinator.com