I’d probably word it differently but I agree with much of the sentiment here. I’m also reminded of the stat where 93% of drivers rated themselves as above average.
Indeed. To be honest, I think everyone on HN is aware of how LLMs work at this point, it’s not actually adding a great deal to the discussion to keep going on about autocomplete or ‘stochastic parrots’.
The results being a lot better crafted by hand I would agree with, if one removes any notion of a time constraint. Sometimes the comparison point is between the LLM authored software or nothing at all.
I think the other aspect to this which you allude to at the end is that all of these arguments start with the assumption that all human software engineers produce high quality code that meets the requirements, but obviously that’s very much not the case in the real world. After all, 80-90% of drivers rate themselves as above average.
If one compares a single competent software engineer directing a number of agents against a random group of engineers (not necessarily working at FAANG or a YC startup), then those quality arguments are going to be significantly less compelling.
Is your argument that there is no imaginable situation where someone who was competent at software development could find use for a semi-automated tool for writing software?
That would imply that either the person in question has infinite time, or has access to all software that could ever be of utility to them, which seems unlikely.
Speaking as someone who works in the industry, I haven't really heard this sentiment. Artists are predominantly hostile to diffusion models, but optimistic about LLMs and their ability to help them write tools and scripts even if they're non-technical.
Using LLMs for creative work is quite different to using diffusion models for creative work. It normally means writing tools or automation processes to enhance the creative flow, not replacing the creative input of a human.
Super cool. I loved Galways's C64 tunes as a kid, especially Wizball & Parallax. I remember trying to write my own player in assembly (yet another unfinished project).
UBI is just a massive extension of the welfare state. Governments can’t afford the current welfare spending, so where is the money going to come from? What do you think is going to happen to the markets when a large amount of the middle classes get laid off and can’t afford to pay their mortgages? What do you think is going to happen to the tech companies built on advertising to consumers when no-one has disposable income?
I think it's not that difficult to see why a technology that will likely trigger widespread unemployment during a cost of living crisis, an arms race with China, along with all the alignment concerns, might not be hugely popular with the public.
Maybe I'd be a bit more optimistic if someone could explain a realistic economic scenario for how we're going to transition into our utopian abundant future without a depression or a revolution.
It's a good question, I've wondered that myself. I haven't used GLM-5 with CC but I've used GLM-4.7 a fair amount, often swapping back and forth with Sonnet/Opus. The difference is fairly obvious - on occasions I've mistakenly left GLM enabled running when I thought I was using Sonnet, and could tell pretty quickly just based on the gap in problem solving ability.
> This is speculative, but I suspect that if we dropped one of the latest, most capable open-weight LLMs, such as GLM-5, into a similar harness, it could likely perform on par with GPT-5.4 in Codex or Claude Opus 4.6 in Claude Code.
Unless I'm misunderstanding what's being described here, running Claude Code with different backend models is pretty common.
I've been playing with the open models since the original llama leak. They're getting better over time, are useful for tasks of moderate complexity and it's just cool to have a binary blob of knowledge that you can run locally without an internet connection.
However you should manage your expectations. Whatever the benchmarks say, you'll quickly realise they're not at all competing with Sonnet let alone Opus. Even the largest open weights models aren't really doing that.
Haven't really tried GLM5 much but I've used 4.7 quite a bit and it was pretty far from competing with Sonnet at the time, although I saw claims online to the contrary.
I would say thinking about the indended audience for your creative outlet is a good discipline - even if it's only one person. It often gives the project more of a focus which helps with motivation and makes it more enjoyable.
Honestly a lot of useful software is ‘unimportant’ in the sense that the consequences of introducing a bug or bad code smell aren’t that significant, and can be addressed if needed. It might well be for many projects the time saved not reviewing is worth dealing with bugs that escape testing. Also, it’s entirely possible for software to be both well engineered and useless.
For sure, but I haven't written a single piece of software where security would ever be considered a factor. Not all software runs on the web, not all software deals with accounts etc.
I see a lot of people talk about 'insecure code' and while I don't doubt that's true, there's a lot of software development where security isn't actually a concern because there's no need for the software to be 'secure'. Maintainability is important I'll grant you.
I think this is too broad. If, for example, I get Claude to set up a fine tuning pipeline for rf-detr and it one shots it for me, what have I lost? A learning opportunity to understand the details of how to go about this process, sure. But you could argue the same about relying on PyTorch. Ultimately we all have an overarching goal when engaged in these projects and the learning opportunity might be happening at an entirely different level than worrying about the nuts and bolts of how you build component A of your larger project.
Yeah, I used to enjoy writing code but after a while I realised I actually more enjoy creating tools that I (and other people) liked to use. Now I can do that really quickly even with my very limited free time, at a higher level of abstraction, but it's still me designing the tool.
And despite the amount of people telling me the code is probably awful, the tools work great and I'm happily using them without worrying about the code anymore than I worry about the assembly generated by a compiler.
I’m building an application for documenting modular patches, mostly for my own use case. It uses ML to recognise the patch points, knobs and toggles from a photo of the front panel. You can then build racks from the scanned modules and then store presets of the knobs and connections which are displayed as simple schematics. Idea is ultimately to have it on an iPad as reference to accompany a live performance. Had some fun fine tuning the cable physics engine.