Always happy when I can use a smart model in a sane harness like pi or mastracode.
I only wish I was able to run this locally
GLM 5.2 Is Out
311–320 of 544 posts
Re: GLM 5.2 Is Out
#312Okay so if this model is half a year behind, so let’s say January opus pre-nerf, this is it. Inference is actually quite cheap for token costs, the frontier labs burn most of their money on training new models, priced into their token costs ontop of some margins and paying record salaries. So if this goes open, distills are tried out, independent providers around the world host it with actual price competition, the h…
Gpt2 was too dangerous to release. We just don't see it yet.
Sure, the model itself was harmless, but it lit the fuse
Re: GLM 5.2 Is Out
#313Apparently this isn’t OpenGL Mathematics the C++ library I expected.
Nope. We're doing real matrix multiplication here, not that 4x4 baby stuff.
Re: GLM 5.2 Is Out
#314Earlier quoted context omitted.
I still find it baffling how the idea that HN is "unashamedly anti-ai" gets repeated. Every single model release gets submitted within minutes of an announcement and frequently break 1000+ points within an hour or two. Blog posts about vibe coding or the current flavor of harness/workflow/tool are constantly making the front page. Karpathy's latest writing/presentations or "Learn how LLMs work using X" are perennial…
[flagged]
data centers with evap cooling use a lot of water and in some places its taking away from residents. thats a fact not a conspiracy. closed loop systems exist and its possible to make them mandatory by law or city ordinance, but if they did that the company running the data center would make a little less money so they act like pumping out water is the only way. its the same with carbon emissions and making them build solar panels.
Re: GLM 5.2 Is Out
#315Okay so if this model is half a year behind, so let’s say January opus pre-nerf, this is it. Inference is actually quite cheap for token costs, the frontier labs burn most of their money on training new models, priced into their token costs ontop of some margins and paying record salaries. So if this goes open, distills are tried out, independent providers around the world host it with actual price competition, the h…
Is it going to actually be open source or just open weights? I'm looking forward to trying this with opencode regardless!
Re: GLM 5.2 Is Out
#316Earlier quoted context omitted.
> This is not a local model for any reasonable definition of local That's true for now. I am hopeful that once the hardware markets have recovered from OpenAI's sabotage, we will see more hardware dedicated to local inference that can handle these big models. Also, I'm thinking about the unique MoE routing that Apple is using with their new Apple Foundation Model. The model is trained and architected so that experts…
Reading weights out of memory is the definition of a large linear read. I'm a bit mystified someone hasn't put an embarrassingly parallel flash storage controller next to some tensor processors on a PCIe card. It could have 4Tb of flash hanging off enough channels to saturate SRAM skipping DRAM entirely, and could even offload prompt processing to a GPU in the same workstation so long as it got reasonable tokens/s in…
For sparse MoE models, the single expert layers that the inference gets sampled from are actually quite small - single-digit megabytes or so.
Re: GLM 5.2 Is Out
#317I'm interested in seeing how this changes folks' workflows.
For me, at work I use opus to plan, brainstorm, grill, ask questions about my codebase, etc. It is pretty good about understanding the codebase holistically and providing architecturally clean solutions that actually work. Then I use sonnet as a plan executor and it does well. Follows instructions and runs tests and just overall does great.
At home I make some toy projects using opencode go (I've standardized on deepseek 4 pro as my opus replacement) but it's pretty obvious from the amount of times I've had to fix or revert a change that broke something that it's no opus. I got similar results with kimi. Have not played too much with Qwen.
So I'm wondering what I'd use to get a similar stack at work. Folks say that this version of glm is basically Jan 2026 opus pre me f. Big if true. So would I use GLM for plan and Deepseek v4 pro/flash for execution? Or maybe Kimi or Qwen? I know I'll probably never get as good quality code as I do at work but I'm just toying around here.
Re: GLM 5.2 Is Out
#318How does is anyone able to run this thing locally without paying too much? (I'm interested in specs or GPU that could handle it)
Re: GLM 5.2 Is Out
#319Earlier quoted context omitted.
[flagged]
Pretty much every large Chinese company has state capital baked into it, and these companies will follow the Chinese government's orders 100%. Don't believe anything a Chinese company says about being "open" or "for everyone." Backing any large Chinese company effectively means backing the Chinese government and its oppression in Xinjiang, Tibet, Hong Kong—and maybe soon Taiwan, Southeast Asia, and elsewhere around t…
> Pretty much every large Chinese company has state capital baked into it, and these companies will follow the Chinese government's orders 100%
True of any US frontier lab as well
> Backing any large Chinese company effectively means backing the Chinese government and its oppression in Xinjiang, Tibet, Hong Kong—and maybe soon Taiwan, Southeast Asia, and elsewhere around the world.
So when I pay anthropic am I also sponsoring the mass murder of school children in Iran?
Re: GLM 5.2 Is Out
#320Earlier quoted context omitted.
The Anthropic news is demonstrating much the same; fall in line or eat export controls. There was a time I would have agreed with you, but these days even as an American I fail to see a difference. China is probably less likely to try to disenfranchise or imprison me, to be honest.
> There was a time I would have agreed with you, but these days even as an American I fail to see a difference. I don't get it, the person you're replying to didn't mention the US at all – there was no distinction being drawn, and they weren't asserting that American models are better or more resistant to government censorship. It's possible to agree with them about Chinese models without expatiating on why American…
I think it's a worthy retort simply because it's the only other major provider.