I've been driving glm-5.2 for a day or two now. It feels like a mature, seasoned colleague.
It could be luck, but I don't know -- it keeps one-shotting relatively hard stuff. And taking initiative to think about what potential regressions it should look out for, and choosing to do strategic refactoring when it should do. It is not confidently incorrect hardly at all, doesn't tell me that it's fresh risky pile of changes is ready for production without having exercised all the code paths and writing a bunch of tests, etc.
We might be reaching the next level here...
Which harness are you using?
What's your hardware stack?
Anthropic and OpenAI window for a successful IPO is reducing day by day. All that pressure from their debts, compute costs, infrastructure investments, training costs, and open weight models continuing to improve. I know the stock market is all about hype and isn’t rational, but there will be a point where the hype will fade away, and they have no moat that will differentiate them from the rest.
Good for consumers, it’s competition at its best, we get cheaper, better services. But I would be pretty concerned integrating an AI lab products into my business without having a good abstraction that makes it easy to swap between vendor.
> But I would be pretty concerned integrating an AI lab products into my business without having a good abstraction that makes it easy to swap between vendor.
I’d look for OpenResponse-compatible endpoints in your shoes. These are meant to be plug and play.
How much of why glam 5.2 is good today is due to open source contributions? Is the two-way-street-ness of it already pushing it to be better or so far it’s mostly a nice to have?
Please correct me if I'm wrong, I'm totally out of my field here but what's the point of sota models that can be run only by hyperscalers? I mean, glm-5.2 is open source but with 1.5TB in weights who can run it really? It still needs dozens of H100s. Those 753B quantized down to Q4 (~400Gb) would require datacenter levels of hardware. Down to Q2 still would require serious hardware, way out of reach for most users, and you'll be far from the sota benchmark of the full precision model. I get it, it's open source but not quite democratizing LLM for everyone except compute providers. It's no like, let's say, Kubernetes. I can run k8s fully in my shitty homelab, without "quantization" exactly like Google does in their datacenters.
If you have a ton of capital, you still can't spin up Claude Opus and compete on price with Anthropic with your new fancy optimizations. With open models you can and that is great for consumers.
Never stop cheering for open source .
If you were a human 3000 years ago ,you wouldn't want fire to be controlled by two chiefs.
Actually i bet there were people back then who were worried about the wrong people getting access to fire for various reasons