Thread regarding IBM layoffs

Granite Is A Pile Of Dissembling Nonsense

Here is a prompt to granite itself plus the answer.


how many parameter do you have
Granite
Granite

As a large language model, my architecture is based on the Transformer architecture, which includes numerous learned parameters. However, I don't have direct access to specifics regarding the exact number of parameters in my model. My design and parameter values were decided upon by IBM during the development phase. It's typical for models of this scale to contain billions of parameters, which are adjusted during training on a vast corpus of text to understand and generate human-like text.

It's important to note that while the number of parameters is often discussed in AI (artificial intelligence) contexts, it's the values of these parameters, their training methods, and the architecture design that truly determine a model's capacity and performance. Simply knowing the count does not provide a complete picture of the model's capabilities or behavior.


Now a similar prompt to Copilot plus its answer


how many parameters does granite ai model have

IBM's Granite AI models come in different sizes, with parameter counts ranging from 2 billion (2B) to 8 billion (8B). These models are designed for enterprise AI applications, offering multilingual text and image processing capabilities. Some specialized versions, like Granite Code Models, have even larger variants, including 3B, 8B, 20B, and 34B parameters.
Granite models focus on efficiency, reasoning, and trustworthy AI, making them competitive with other large language models. Would you like to explore how they compare to other AI models in terms of performance and use cases?


Dealing with granite is exactly the same as dealing with IBM: you cannot possibly get a straight answer; it's all baffle-gab with no actual answers.

I bring this up because THIS is the future of the company. This is what Arvind et. al. are hanging their hats on. And it's pretty much useless. It cannot or will not answer an easy question. Even about itself. Clearly, the training is a million miles behind the big players. Just as IBM Cloud was (and is) a million miles behind the big players.

by
| 1321 views | | 7 replies (last June 10, 2025) | Reply
Post ID: @OP+1jxbswvpb

7 replies (most recent on top)

Based on the web site noted by @ab+1jxbswvpb and the analysis by @c8+1jxbswvpb this is where we were with cloud a decade or so ago. IBM cloud was always in the bottom left quadrant of independent comparisons of widely available clouds. This was largely because IBM was late to the party and all they could do is say "look at me!" Most people in the company (including all levels of leadership) really had little knowledge of the technology. Ultimately, IBM cloud became existing IBM software repackaged as "CloudPaks" and a lot of consulting mostly involving wrappers on other peoples' wrappers.

The parallels are astounding. Back then, IBM was already trying to hedge by saying "AI is coming and we will lead it." Today, they are saying the same thing about quantum.

In the case of cloud, they bought some credibility time by buying Red Hat. They also bought Soft Layer but basically ground it into nothing. Probably @ah+1jxbswvpb is correct and IBM needs a new Red Hat for AI.

by
| | Reply
Post ID: @dj+1jxbswvpb

When the next generation of business school textbooks are written, only one sentence will be required regarding the case study of IBM:

"They sold it all for cash."

How did Google, Microsoft, OpenAI and even Apple build their AI initiatives? They all used the knowledge and experience that they gained from the other parts of their business to establish their AI operations. They all know how to develop hardware and software, they all know how to utilize cloud and network technologies, and they have a base of users who are ready and willing to test their stuff.

IBM once had all that, but they sold it all off over the years. They can build and sell mainframes and legacy enterprise software, but not much else. They decided not to cater to most markets, and so those markets went elsewhere. Developing anything that uses IBM stuff costs a ton of money and is a nightmare to most potential developers, so they don't use IBM stuff. IBM's management sold off the divisions they had that could run an IT shop, and now they can't even run a reliable cloud infrastructure anymore.

So Granite can't describe itself. Neither can anybody else in IBM besides maybe the few scientists and engineers who developed it. There's your problem right there.

by
| | Reply
Post ID: @dh+1jxbswvpb

IBM's new marketing slogan:

IBM Granite AI: we turn valuable third party information and insights into baffle-gab with no actual answers.

There's probably a huge market for this in the executive suites of fortune 500 companies!

by
| | Reply
Post ID: @dc+1jxbswvpb

Why would anybody bother with IBM's cloudy 2B model when you can run more than that on your phone?

by
| | Reply
Post ID: @db+1jxbswvpb

IBM isn't really competitive in the modern AI market. Just doing the bare minimum to check box and say they have a product. Watsonx seems to be chasing where the market was 2.5 years ago. Latest strategy for things like MCP Gateway are to just build wrappers on top of other people's wrappers. Nothing innovative or differentiated.

by
| | Reply
Post ID: @c8+1jxbswvpb

Wrong, Arvind et all count on more acquisitions!

by
| | Reply
Post ID: @ah+1jxbswvpb

8 billion parameters is peanuts compared to the >400B parameter models out there. If you want an idea as to where IBM stands versus its competition, go to llm-stats.com and go to page 4 to find the IBM Granite models.

by
| | Reply
Post ID: @ab+1jxbswvpb

Post a reply

: