5 ms·

With years of expertise reading AWS blogs and documentation, let me attempt to translate: - Bedrock, an AWS service that makes base LLMs ("Foundational Models"

by ctvo 4y ago

With years of expertise reading AWS blogs and documentation, let me attempt to translate: - Bedrock, an AWS service that makes base LLMs ("Foundational Models", FM) provided by both Amazon and third-parties accessible via API. - Amazon Titan is one of these foundational models, provided by Amazon. Others include: ... things I've never heard of from companies I barely recognize. Customer -> Bedrock -> Titan. Swap Titan for something better from a new AI start-up. It's another marketplace play.

RockyMcNuts 4y ago
where's the paper on Titan LOL ... can't be good if no paper or metrics saying it's better then GPT-4 or even GPT-3. but it's amazing that Google has all the papers and models and the TPU and BERT is pervasive in their products and they can't even stake a better claim to leadership than AWS does here, Amazon stock probably goes up, Google announces anything their stock goes down.
- pg1010 3y ago
  I guess the reason of Amazon's stock going up while Google's stock going down might be that no one expected anything from Amazon/AWS in generative AI space yet, while everyone already expected Google to launch something like Bard for search engine and when launched, it was not as per market's expectation that had priced the value already. Also, Google has a potential of losing large search market to Bing and thus losing the revenue. While Amazon with this launch does not have the same story, it might (or might not) help them to gain more market in a new space but might not lose the existing ones, at-least not the major revenue source like search for Google.
ctvo 4y ago
> One of the most important capabilities of Bedrock is how easy it is to customize a model. Customers simply point Bedrock at a few labeled examples in Amazon S3, and the service can fine-tune the model for a particular task without having to annotate large volumes of data (as few as 20 examples is enough) I can't even. Does anyone remember Amazon Kendra [1]? They promised the same there. "Here's an ML powered search engine! Point it at an S3 bucket with your own data and you're done". Amazon Kendra is one of the worst products I've ever used, and we ripped it out as quickly as possible. AWS has a long history of over promising and under delivering when it comes to anything AI / ML. 1 - https://aws.amazon.com/kendra/ https://aws.amazon.com/kendra/
- orf 4y ago
  What was wrong with Kendra?
  - ctvo 4y ago
    Result relevancy was worse than random full text search with your weapon of choice and it costs an order more to use. Sitting there manually adding synonyms to a thesaurus to try and get something of value wasn't what we wanted to spend time doing.
    - manv1 4y ago
      Every relevancy-based search I've used pretty much sucks, so Kendra sounds about par for the course.
lightbendover 4y ago
I will only be impressed when I see Amazon use it to correct all of their category and brand mappings for products. It has been a travesty since inception for a company purportedly data-driven.
- gmt2027 3y ago
  There's also Alexa. I've seen no noticeable improvement in 5 years. It is difficult to believe that Amazon has any credible AI expertise to sell.
- Turing_Machine 3y ago
  I'd like to see them use it to get rid of the plague of counterfeit products (I guess that's arguably a subset of "correcting").
  - lightbendover 3y ago
    But that would make the arrows go down so you know they'll never do that.
squarecog 4y ago
LLMs are Foundational Models, but not all Foundational Models are LLMs. If you are interested in this space, it does probably make sense to look up what Stable Diffusion (StabilityAI) and Claude (Anthropic) are. Not recognizing them is fine, the pace is pretty incredible in this space and there's lots to learn, but it doesn't make the point you think you are making.
- ctvo 4y ago
  I wasn't attempting to make any point there except not listing out every company they included in their press release. Not an expert in the space or LLMs, but I'm 100% their target audience: someone technical trying to decipher the release to see if it's of any use to their own company.
mrandish 4y ago
Thanks for providing your translation. It's sadly ironic I found TFA basically useless given it was likely written by someone with "communications" in their job title.
MLEAIE 3y ago
if you haven't heard of stability AI then it is just sad. I don't believe any AI credibility from MS either. OpenAI is not MS and integrating it with Azure is a pure marketing play as well.

by ctvo 4y ago

RockyMcNuts 4y ago

where's the paper on Titan LOL ... can't be good if no paper or metrics saying it's better then GPT-4 or even GPT-3. but it's amazing that Google has all the papers and models and the TPU and BERT is pervasive in their products and they can't even stake a better claim to leadership than AWS does here, Amazon stock probably goes up, Google announces anything their stock goes down.

pg1010 3y ago
I guess the reason of Amazon's stock going up while Google's stock going down might be that no one expected anything from Amazon/AWS in generative AI space yet, while everyone already expected Google to launch something like Bard for search engine and when launched, it was not as per market's expectation that had priced the value already. Also, Google has a potential of losing large search market to Bing and thus losing the revenue. While Amazon with this launch does not have the same story, it might (or might not) help them to gain more market in a new space but might not lose the existing ones, at-least not the major revenue source like search for Google.

ctvo 4y ago

> One of the most important capabilities of Bedrock is how easy it is to customize a model. Customers simply point Bedrock at a few labeled examples in Amazon S3, and the service can fine-tune the model for a particular task without having to annotate large volumes of data (as few as 20 examples is enough) I can't even. Does anyone remember Amazon Kendra [1]? They promised the same there. "Here's an ML powered search engine! Point it at an S3 bucket with your own data and you're done". Amazon Kendra is one of the worst products I've ever used, and we ripped it out as quickly as possible. AWS has a long history of over promising and under delivering when it comes to anything AI / ML. 1 - https://aws.amazon.com/kendra/ https://aws.amazon.com/kendra/

orf 4y ago
What was wrong with Kendra?
- ctvo 4y ago
  Result relevancy was worse than random full text search with your weapon of choice and it costs an order more to use. Sitting there manually adding synonyms to a thesaurus to try and get something of value wasn't what we wanted to spend time doing.
  - manv1 4y ago
    Every relevancy-based search I've used pretty much sucks, so Kendra sounds about par for the course.

lightbendover 4y ago

I will only be impressed when I see Amazon use it to correct all of their category and brand mappings for products. It has been a travesty since inception for a company purportedly data-driven.

gmt2027 3y ago
There's also Alexa. I've seen no noticeable improvement in 5 years. It is difficult to believe that Amazon has any credible AI expertise to sell.
Turing_Machine 3y ago
I'd like to see them use it to get rid of the plague of counterfeit products (I guess that's arguably a subset of "correcting").
- lightbendover 3y ago
  But that would make the arrows go down so you know they'll never do that.

squarecog 4y ago

LLMs are Foundational Models, but not all Foundational Models are LLMs. If you are interested in this space, it does probably make sense to look up what Stable Diffusion (StabilityAI) and Claude (Anthropic) are. Not recognizing them is fine, the pace is pretty incredible in this space and there's lots to learn, but it doesn't make the point you think you are making.

ctvo 4y ago
I wasn't attempting to make any point there except not listing out every company they included in their press release. Not an expert in the space or LLMs, but I'm 100% their target audience: someone technical trying to decipher the release to see if it's of any use to their own company.

mrandish 4y ago

Thanks for providing your translation. It's sadly ironic I found TFA basically useless given it was likely written by someone with "communications" in their job title.

MLEAIE 3y ago

if you haven't heard of stability AI then it is just sad. I don't believe any AI credibility from MS either. OpenAI is not MS and integrating it with Azure is a pure marketing play as well.