electroglyph

Welcome to Costco, I love you.

of course they can do math, how is this even controversial in 2026?

hopefully someone on the team sees this and fixes the situation, otherwise, if it's good, i'm sure somebody else will open the PR. thanks for your contribution.

they don't always operate within the parameters tho. some N% of the time they decide to do whatever they feel like doing. multiply that times a lot of agents and you invariably get a rogue agent every once in a while.

way less of a blocker these days due to sparse attention...

what the fuck did i just read?

those are all expecting far too much for models this size

do you just manually type out your important instructions every time instead of being smart and putting a few lines in a text file?

prismml's title is very misleading. in their own paper the model is at 75% of coding scores.

mimo 2.5 has been a big underperformer since shortly after it's release imo. i cancelled my sub after the first month. purposefully using 2.5 right now is just handicapping yourself for no reason.

good stuff, no mention of confidence tho, recommend having a look at diffusiongemma and others.

Good on them for the retraction. It's good to see science at work.

it's not a schooner...it's a sailboat

i would assume a model this size would require finetuning tbh. even functiongemma recommends that.

my experience is the same, but deepseek is planning on increasing prices soon, which will make it a lot less attractive

yep, agreed. little bit smarter than deepseek 4 flash, but a little bit worse at tool calls.

the cats can traverse a fence, hole or not, so not sure that's relevant?

If it makes you feel any better, my apology was genuine.

Such a weird project. Does the world really need this?

i handed out so many upvotes here, good work everybody!

i just today pulled the back off my old oneplus 8 pro and put a new battery in it after putting lineageos on there. i decided i was tired of using my locked down Samsung that's full of crap

ONNX has AVX512 CPU kernels too, and openvino uses ONNX internally (and ONNX supports openvino backend)

it's a pain in the ass to do properly.

what we really need it something like auto-round for ONNX

ONNX is my first suggestion to people looking for speed gains on CPU

heh, i came here to say basically the same thing

existing embedding models like alibaba's modernbert tune or one of the jina v5s would probably map query to category automatically. (i.e. store embeddings of each category and calculate cosine sim for each incoming query vs. categories and pick the closest)

also, you could stick a classifier head on a BERT model as another option.

heh, when Raymond Chen dunks on the MSVC team =)

it's probably a lie