Hacker News Reader

Comment by Zambyte | Hacker News Reader

Not sure why this was flagged, Anthropic has obviously been burning bridges. I thought this line was funny though:

> If you want to autocomplete, like I do, you don’t need Fable, or even Opus; Sonnet works fine.

It reads like "if you want to go to the grocery store, you don't need a space shuttle, or even a SR-71 Blackbird; a Cessna works fine."

View on HN

While I did not flag this, it is clearly a hit piece. Why write an article to publish in July that does not include June data? Unless of course one looks at June uptime and finds that does not align with their story.

In June, I figure Anthropic subscriptions were partially powered by those big mobile gas generators that are pushed around the xAI parking lots. Makes sense availability improved.

I began writing this article a while back and forgot to update the screenshot. The June uptime has significantly improved, but that may not always be the case

I agree that without context, Sonnet is overkill for just autocomplete code suggestions. My point later in that section is that to have an autocomplete mindset (where an AI is a helpful tool rather than a driver), you only need a decent subset of models, not the best of the best.

Sounds like the author is using "autocomplete" as a derogatory way of referring to LLMs in general (eg. "LLMs can't code, they're just autocomplete on steroids")

Then the author needs to be more specific because there are special models for code autocomplete that are much faster.

For people who saw this and might want a recomendation, I like running a tiny qwen model with llama cpp. Qwen2.5 coder 0.5B or 1.5B (not the instruct version)

On a modern-ish GPU these should run really fast with little latency. They cost nothing and don't send your data to anyone.

It’s talking about how you can type

  def frobQux(Qux qux, int radians) {

And it goes and reads your code and suggests a reasonable way to frob a qux a certain number of radians. Which is at the same time (a) pretty useful!, (b) fairly new, we’ve only been able to do this since 2023 or so, but (c) also not that hard by 2026 standards because capabilities have advanced so much in the last three years.

Why would anyone be using Fable, Opus or even Sonnet for that type of work? You don’t need an advanced reasoning model for that at all.

If you already knew that the answer was to frob a qux, you wouldn't need the reasoning trash either.

Generative autocomplete with an underpowered model is pretty annoying. It hallucinates parameters and APIs. Something like Sonnet seems like the right level of sophistication.

> Why would anyone be using Fable, Opus or even Sonnet for that type of work? You don’t need an advanced reasoning model for that at all.

Don't leave me hanging: how do you frob a qux a certain number of radians?

"int radians" is a pretty strong code smell! I would be very worried about any struggle l autocomplete that followed something like that ;)

> To be clear, I’m not anti-AI. [...] If you want to autocomplete [...] But if you want to use FOSS trusted tools like Zed, OpenCode, Pi, Nanocoder, etc. [...]

isn't he just contrasting 2 use cases of LLMs?

Are you saying that using LLMs for code autocomplete is like using an aeroplane to go to the grocery store? Because clearly there was no autocomplete before the new AI spring.

Reply on news.ycombinator.com