Here comes Bedrock, Amazon’s ChatGPT and Dall-E rival

It was one of the last tech giants to not take a stand on generative AI. It is now done. In a blog post published on April 13, Amazon announces that it is descending into the arena with several tools that work by taking advantage of the American company’s capacities in storage and computing power, with Amazon Web Services (AWS).

In this context, the company unveiled the bedrock model. It’s actually a suite of tools that are clearly cut out to rival some of the most famous solutions on the market today. We think of ChatGPT for the production of text and of Dall-E and Midjourney for the generation of images on the fly, according to the instructions (“prompts”) of the Internet user.

Bedrock, Amazon’s Generative AI Toolkit

Bedrock provides artificial intelligence models (called foundation models Or foundation models in English), depending on what Amazon’s customers are looking for. Trained on large bodies of data, these models can then be used to perform a variety of tasks, including accelerating the development of downstream, user-side projects.

This is also one of the arguments that Amazon advances. The idea that the user doesn’t have to bother with developing his own language model or with the infrastructure on which to run it. Indeed, there is only need for the API made available by the company. All the technical part is managed by Amazon with AWS, on the same principle as the cloud.

” Bedrock gives you the flexibility to choose from a wide range of models […] so you can find the best model for what you are trying to do “, writes the company. All without the need for any server to manage for the client. ” You can quickly get started […] and easily integrate and deploy into your applications. »

Here, Amazon is mainly looking to attract smaller or mid-sized companies that cannot implement AI processes, such as machine learning, on their own. An offer that could also appeal to startups, which could thus accelerate their development by deporting work on AI models and Amazon’s cloud.

Stable Diffusion, Titan, Claude and Jurassic-2: four models in Amazon

The four models offered are as follows. Three are provided by startups, one by Amazon.

• Jurassic-2designed by AI21 Labs: it is a multilingual language model for generate text. Its particularity is to deal with languages ​​other than English: French, Spanish, German, Portuguese, Italian and Dutch.
• Claudiusdeveloped Anthropic: this model is cut for converse (chatbots), answer questions and automate workflow. Amazon says it was trained on “honest and responsible AI systems.” Claude was mentioned during DuckAssist, the ChatGPT made in DuckDuckGo.
• Stable Spreadfrom Stability AI: it is an equivalent of Midjourney And Slab. This template produces images, illustrations, logos and designs, which can be high quality and realistic. A new version was launched last November.
• TitanAmazon’s solution: it allows you to summarize texts (articles, books, documents, etc.), to generate, find and extract information from corpora of data, or to personalize the experience (recommendation of products, for example).

For example, the e-commerce giant imagines a marketing manager working for a fashion company, who needs to develop targeted advertisements and campaigns for a new line of handbags. Bedrock can then generate social media content, advertisements and texts around the handbags, automatically.

He just had to give him a few labeled examples (about twenty would be enough, according to Amazon) of the brand’s most successful slogans in previous campaigns, as well as the products to be highlighted. The marketing manager only has to choose what suits him or to generate other productions.

You can even use your own data to personalize the results

Amazon even foresees the possibility of privatizing one or more foundation models – presumably via a dedicated instance on AWS. The idea? That a client company customizes how a model works by training it with its own information. A prospect that will raise questions about data security and privacy.

In this regard, the company indicates that no customer data is used to train the underlying models, and that all data is encrypted and does not leave the virtual private cloud (VPC) of the customer. In short, a promise is made that the data will remain private and confidential. Not sure, however, that this will be enough to convince them to take such a step.

With Bedrock, Amazon joins its usual rivals in tech: before it, Google, Facebook and Microsoft have chained announcements concerning generative artificial intelligence, with more or less success. Among the heavyweights, there is little more than Apple which remains relatively discreet. But maybe that will change in June, with WWDC 2023.

---

If you liked this article, you will like the following ones: do not miss them by subscribing to Numerama on Google News.