But economics of Jev enable classifying high volume things. And I've tried all the open-weight Qwen based decision models on my Spark and none of them came closer to quality and suck at batch inference so they are doing something custom.
Silly question but is it right? I am in grad school and this semester we run multimodal data through ML and LLM-as-judge classifiers. Even plugging in ChatGPT with a relevant judge prompt produces relatively believable results. At the same time, I struggle to trust it to generalize. I’ve read that Jev does classification across many domains and is also powered by an LLM back-end? What separates it from my above approach outside of harness efforts and finetuning of the LLM? Thank you!
Joking aside, as much as Go's stdlib and tools do the heavy lifting here, Go's verbostiy and expressing simple things in lots of lines worked against me most of the time.
I've had a really terrible time getting LLM to properly handle errors as return values. It seems that bubbling up errors, in a side channel, to a contextually relevant point in the code (exceptions) seems MUCH easier for LLM to reason about/implement properly.
Maybe my problem is I'm using a language with exceptions, so trying to go against the statistical grain, with return values, is just too much.
Exceptions are inherently better for high-level code, where basically every loc can fail and 99% of the time you only want to bubble that up. You only want errors as values in systems code, where exceptions would be landmines. Rust and Go both did that because they were at least originally designed for systems code.
Also, Go makes it way too easy to accidentally swallow an error. Rust doesn't have that problem.
I createad an MCP server for my company I worked at for almost a year now with all Dynamic Client Registration (DCR). We just ignored the state as with most other MCPs wre doing it, just used it for logging.
But economics of Jev enable classifying high volume things. And I've tried all the open-weight Qwen based decision models on my Spark and none of them came closer to quality and suck at batch inference so they are doing something custom.