Hacker News Reader

Comment by ramon156 | Hacker News Reader

hence the "former deepseek v4 pro". I tried it out this morning and have had no complaints. I already liked glm 5.2

> Deepseek v4 pro 0813

Which itself released yesterday? You're writing, reading, and evaluating enough software in a ~36 hour period to form, reject, and form another opinion about which model makes better _architectural_ choices?

It's the singularity, baby.

He's 100x, bro. 36 hours is basically like a whole 2 week sprint for you.

The sentence still doesn't make sense, because "settled on" implies a long testing phase with a verdict eventually emerging out of that.

What you're currently doing is "testing out"

Honest question, how do you assess models this quickly? What metrics are you using? Would love to get my suite from multiple days and hundreds of prompts down to minutes. Got a few first pass tasks I run upon release for an initial experience, but those only work because even Fable and Sol fail despite objectively correct solutions existing, so it works because most models fail, but then, those are consciously not enough for coding, tool use, adherence or task specific inference and assessment…

What are you working on? That can dictate which models are best.

Anything from ML pipelines for language specific pruning over a Rust/JS/CSS mix codebase to assistance in motorcycle maintenance and different canvas coatings. Most of my evals build on those requirements and especially past failures, whether in pure information, task execution and coding or tool calling beyond the overfitted mainstream. All stuff derived from actual failures encountered, some still only few models come even close to passing. With such a mix, it just takes a while to get any serious opinion on a model. Doubt anyone can do that in such short time, unless their tasks are so simple that most modern models not only succeed but could themselves accurately rate output. If even Fable still confuses PU or wax coated cotton canvas with a nylon shell, or tells me with a straight face to adjust the valves on a bike that has hydraulic lifters that needs experience for human assessment and the time that comes with it. Anyone with less knowledge either wouldn’t see the mistakes staring them in the face and just go by vibes, any model rating these equally can not tell what is accurate and will just go by the output sounding accurate over being. Gives sometimes very interesting results far different to public benchmarks. Inkling, e.g. is more accurate in not telling you to adjust valves that are simply not adjustable then Fable or Sol, which just tell you to adjust every 5000km. Sometimes even when their reasoning and search includes sections about the fact this is not necessary or possible. The beauty of overfitting and unbalanced training data…

Most of us aren’t deploying these in general purpose use-cases. E.g. I use Qwen mostly for vision in my personal assistant. I have an eval set for that. Pretty much each of my use cases has a pre-computed problem set.

Likewise I have a tool-use eval set and a browser-use eval set. I use frontier models for most interactive tasks that are not home assistant or background agents.

I can imagine someone could build evals for that but I have never done so.

Deepseek v4 Pro or GLM 5.3 for software architecture I feel are only deployed for general, rather doubtful those are for narrow stuff, if we are sticking with this threads mentioned models. For small models, sure, narrowly targeted sets which can be self evaluated are amazingly valuable, but I feel beyond 500B we are in a different dimension. Rating any model the size of GLM or V4 Pro in hours I doubt is done beyond pure vibes.

Reply on news.ycombinator.com