Hacker News Reader

Comment by lukev | Hacker News Reader

This is a digression, but I really wish Amazon would be more normal in their product descriptions.

Amazon is rapidly developing its own jargon such that you need to understand how Amazon talks about things (and its existing product lineup) before you can understand half of what they're saying about a new thing. The way they describe their products seems almost designed to obfuscate what they really do.

Every time they introduce something new, you have to click through several pages of announcements and docs just to ascertain what something actually is (an API, a new type of compute platform, a managed SaaS product?)

View on HN

Amazontalk: We will save you costs Human language: We will make profit while you think you're saving the costs

Amazontalk: You can build on <product name> to analyze complex documents... Human language: There is no product, just some DIY tools.

Amazontalk: Provides the intelligence and flexibility Human language: We will charge your credit card in multiple obscure ways, and we'll be smart about it

That may be generally true, but the linked page says Nova is a series of foundation models in the first sentence.

Yeah but even then they won't describe it using the same sort of language that everyone else developing these things does. How many parameters? What kind of corpus was it trained on? MoE, single model, or something else? Will the weights be available?

It doesn't even use the words "LLM", "multimodal" or "transformer" which are clearly the most relevant terms here... "foundation model" isn't wrong but it's also the most abstract way to describe it.

None of those matters (except multimodal). If you are running a business, the only thing that matters is

a) How does it perform on my set of evals

b) What is the cost/latency of serving it to my consumers.

It shouldn't matter to me how many parameters, corpus it is trained on, whether it's LLM or Transformer or something else

> How does it perform on my set of evals

What kinds of eval? Personally, I have no idea what kind of data you can throw at a "foundation model" and what kind of response you will get.

The only thing it says is that there's machine learning involved... Once you get enough context to understand it's not a spin-off of a TV series.

"Foundation model" is not Amazon lingo, though, but pretty standard industry term at this point. If you're doing any sort of AI in prod, you know what it means.

> How many parameters? What kind of corpus was it trained on?

It's rare for the leading model providers to answer these questions.

As someone who applies these models daily, I agree with the dead comment from meta_x_ai. Your questions are interesting/relevant to a person developing these models, but less important to the average person utilizing these models through Bedrock.

Amazon is not a "leading model provider".

Once upon a time there were (and still are) mainframes (and SAP is similar in this respect). These insular systems came with their own tools, their own ecosystem, their own terminology, their own certifications, etc. And you could rent compute & co on them.

If you think of clouds as being cross continent mainframes, a lot more things make a more sense.

"distributed mainframes".

or "internet OS with bundled hardware"

If you figure out what a security group is, let me know :-D

Lol

What’s the subnet of the security group of my user group for Aws lambda application in a specific environment that calls kms to get a secret for….

Reply on news.ycombinator.com