The few tests i ran i just got Granite4.1 tell me about YOLO repeatedly for no reasons; every questions and prompts were answered with a mention of a vision system unrelated with the task…
Aside from this just being standard corporate speak copy, there is an element of truth there - the granite models are likely to give you more plain toneless responses rather than a rainbow of emojiis. Which in corporate setting can be useful
That actually sounds nice in theory. A model for getting work done and nothing else, which doesn't have to account or be trained for any type of user engagement. Like how chatgpt/claude are wasting their capacity on social niceties and glazing the user. I don't know if granite is one such but I'd bet that many of the other popular models can't be since they are consumer facing.
They should go all in on AI first and rebrand as AIBM, it has a nice ring to it.
"IBM, UBM, we all BM for IBM." --David Gerrold, "When HARLIE Was One".
https://en.wikipedia.org/wiki/When_HARLIE_Was_One
The few tests i ran i just got Granite4.1 tell me about YOLO repeatedly for no reasons; every questions and prompts were answered with a mention of a vision system unrelated with the task…
I shudder to think of what IBM’s government based clients are using YOLO for.
That thought now reminds me of the ibm accounting machines and their punch cards… Not a great perspective.
Or trying to… given the performance of this release.
I bet though internal tools are more efficient.
What is an enterprise workload?
Pragmatically enterprise tends to mean less refined, designed by committee and expensive.
In this case i would guess it is mostly a justification for taking a part of the LLM pie.
Aside from this just being standard corporate speak copy, there is an element of truth there - the granite models are likely to give you more plain toneless responses rather than a rainbow of emojiis. Which in corporate setting can be useful
That actually sounds nice in theory. A model for getting work done and nothing else, which doesn't have to account or be trained for any type of user engagement. Like how chatgpt/claude are wasting their capacity on social niceties and glazing the user. I don't know if granite is one such but I'd bet that many of the other popular models can't be since they are consumer facing.
Existing models from OpenAI etc never return emojis when using the raw APIs unless you ask for it
OCR'ing tables into spreadsheets, apparently
"This approach presents a significant opportunity for optimization and strategic realignment to better meet our core objectives."
Here the most relevant models: https://hugston.com/models/granite-41-8bq4-k-m and: https://hugston.com/models/granite-41-30bq4-k-m
Previous discussion: https://news.ycombinator.com/item?id=47960507
The lmstudio link points to granite 4.0
cool portfolio flex from IBM, but until I see independent benchmarks that aren't cherry-picked, I'll believe it when I see it