Amazon Bedrock does it serve any purpose?!
Azure has OpenAI models, state of the art and so far no model has come close to it. Then there are a ton of open source alternatives that are constantly getting smaller and better. OpenAI is possibly working on upgrades to GPT4 and ChatGPT, soon make them multimodal and do all kind of stuff. Support APIs for multimodality. Google is a beast on its own with even bigger and powerful models and they will eventually catch up OpenAI and Azure. Now, with AWS, you have Bedrock, we claim we have titan models, so far no data points to show. Looks like it is not available for atleast a few more months. If bedrock is a managed api, the only reason a consumer might use it is if it is cheaper than OpenAI API pricing and our titan models are better than OpenAIs. With the other so called partner models like AI21 etc., all of these models are available on other cloud and they also offer APIs on their own. If that’s the case why would I use bedrock. Opensource models I don’t care, since if they are getting smaller and better, I can host it anywhere. What is the real deal with bedrock then. Why would any enterprise customers of AWS use bedrock. It feels like a marketing gimmick to me. I don’t see any real value with this service. Why would anyone bother accessing or hosting a shitty model or one that is inferior to gpt. I am completely confused and baffled by all this noise by AWS. What is going on? TC: 375
We are claiming that we don’t use customer data to train our models and their data never leaves the VPC. So, I’m assuming those concerned about data privacy and sovereignty will be interested that’s all I can think. I have been prototyping with GPT-3.5-turbo and GPT-4 for a while. AWS is going to have to really impress me to move from OpenAI, and I work for AWS 😁
Azure, GCP and even OpenAI Foundry guarantees all of the security promises. Given a majority of companies are hybrid anyways, why would I care. Isn’t model quality a greater requirement than security.
Agreed. I think the whole security angle is not really unique. I wasn’t impressed with the webinar from yesterday about Bedrock. Listen to how they answer the questions. They kept saying Bedrock is the easiest way to consume LLMs because we expose it through an API. News alert so does OpenAI and Azure OpenAI service. I’m assuming the team Bedrock team has never used OpenAI “API”. 😁
Think of it this way - I am an AWS shop. I have committed $10M cloud spend to AWS every year. I have all my stuff in AWS. Would I go set up all my infra in Azure or GC now for just one use case? Likely not. Def not if aws can provide me some similar service that is somewhat similar, or at least close in quality. Look at this as Aws’s attempt at that something similar but somewhat inferior quality (today) so these customers have something to leverage and start using right away. Each service that aws builds doesn’t have to be the best in its category - as long as aws as a whole is better or cheaper or easier to use. Hopefully two out of those three things.
Just like how azure shops know their infra is total shit and security is a joke but still use it because it's cheaper for them to tie to their office and windows licensing bs
This is actually a fair and reasonable justification. It’s s disappoint from the perspective that AWS will likely not be the service of choice for any major or significant use case though. Where quality matters, devs will leverage services elsewhere. A lot of companies are also going multi-cloud as well to prevent getting locked into a provider’s ecosystem and to better negotiate costs. For those major players, AWS will be at a huge disadvantage in this realm.
1. AWS customers want a service on aws. 2.Fastfowrd 5 years and the model performance will be a commodity. 3. There are hundreds of use cases inside amazon that can’t use openai or azure.
Hail Mary