Does "major number release" mean that it is actually an order of magnitude more compute effort that went into creating this model?
Or is this fundamentally a different model architecture, or a completely new tech stack on top of which this model was created (and the computing effort was actually less than before, in the v3 major relase?
There is a well-known CLI tool for JSON processing called jq.
I have just asked GLM-4.7 for the name of jq's built function to convert a string to lowercase. It is called ascii_downcase() according to the manual:
However GLM-4.7 insists that is called ascii_down().
I tried to correct it and gave the exact version number, but still, after a long internal monologue, This is its final world:
"In standard jq version 1.7, the function is named ascii_down, not ascii_downcase.
If you are receiving an error that ascii_down is not defined, please verify your version with jq --version. It is possible you are using a different binary (like gojq) or a version older than 1."
GLM-5 gives me the correct answer, ascii_downcase, but I can get this in the Chat Window. Via the API I get HTTP Status 429 - too many requests.
Seems that I cannot use GLM-5 via the API yet, because I am on the Coding-Lite Plan, the most basic paid tier.
I have also realized that I get faster and correct answer to the ascii_downcase question (even from GLM-4.7) when I submit to open.bigmodel.cn endpoint rather than the z.ai API endpoints (using the same API key). I get a mix of Chinese and Western characters in error responses from open.bigmodel.cn though, while the z.ai endpoint does only contain Western Characters.
(Just assuming that both websites are operated by the same company).
From their email titled ”GLM-5 and New Z.ai Chat Coming!”
> Availability Note: GLM-5 is currently available exclusively on Max and Pro Coding Plan users. We are working to scale our infrastructure and will roll this out to Lite users as soon as capacity permits.
How many Microsoft employees are working on Azure Linux in 2026 (full-time equivalents)?
Github Project Page lists ~ 195 contributors today.
Is Azure Linux relying on community contributions, and MS employees do not write code, justt review, plan, coordinate?
Or is it the other way around, Microsoft developers do most of the work, and occasionally accept a small PR and interesting feature requests from the community, here and there?
Q: There is not a single occurrence of the word "infer" (and related terms such as "inference") in the whole paper. Did you carefully try to avoid it or did this happen accidentally? Or is it the point of your paper?
(I encounter Type Checking only in my IDE when red squiggly lines appear under syntax errors etc. So consider this a layman Q)
Very good question! Funnily enough, it was indeed a conscious decision to avoid mentioning "inference", because then people might expect that we're doing polymorphism with type variables/or maybe that we're deriving a Hindley-Milner inference system.
We're not doing these things, so we figured it would be safer this way, but absolutely you could call what we're doing "type inference".
That being said, we've got some ideas for a follow-up paper to revisit this and derive something more worthy of being called "inference".
The more complicated a type system is, the harder it is to make inference work.
Global type inference typically works well in Haskell without any hand-holding, but when you pack more features into the type system (e.g. in Idris), the programmer needs to explicitly write out more type signatures.
This type system looks "next-level complicated", so inference is probably way out of the question, e.g. if you saw the expression 2 + 3, maybe the types are:
If you were type-checking, you could start on the RHS, end up with the LHS, and then confirm that these two sub-expressions can be added. But with inference you're trying to figure out the most appropriate RHS from the LHS.
Long-term memory (LTM) consolidation is thought to require the prior establishment of short-term memory (STM). -- Authors show that it does not have to be this way.
I think, in Germany, only small companies (sole proprietorship businesses) will see your diverse skillset as an advantage.
Or one of those companies who are really desperate to hire someone for whatever reason. Perhaps you can _take over_ such a business (search for "Nachfolger gesucht unternehmen")
It is not clear what your problem is. Has the freelancing side-business dried up, or is the "main day job" gone, or both?
Z.ai seem to promote 4.7 for smaller tasks, 5.1 for larger tasks (similar to Anthropic's recommendation for usage of Haiku and Sonnet/Opus models).
5.1 works for me already in the most economical basic paid tier ("lite coding plan"), unlike first release of v5 (5.0 ?)
reply