Forrester Wave 2024: Language AI Foundation Models

 

Forrester Wave 2024

Google Cloud are pleased to declare that Google has achieved the highest scores among all vendors assessed in the Current Offering and Strategy categories, making it a Leader in The 2024 Forrester Wave: AI Foundation Models for Language, Q2 2024.

Gemini stands out in the market primarily due to its multi modality and context length, and it also maintains interoperability with a wider range of complementary cloud services.” – Q2 of 2024 edition of The Forrester Wave: AI Foundation Models for Language

The Forrester Wave

The Forrester Wave: AI Foundation Models for Language, Q2 2024 is available for free download.

Google Cloud’s interactions with Forrester Wave are changing as a result of generative AI. Developers now have the ability to leverage powerful managed models to create creative new applications, experiences, and agents for end users. With only 1% of the data required in the past, tuning models is now easier than ever. The advancement of Gen AI is accelerating across the board.

A long history of AI research and invention at Google, which includes the Transformer architecture, diffusion models, and other ground-breaking projects that are essential to the next generation of AI applications.

Google Cloud’s multi modal family of models, called Gemini, is the product of extensive teamwork from Google departments including Google Deep Mind and Google Research. Gemini models, which were created from the bottom up to seamlessly blend and comprehend text, code, graphics, audio, and video, are assisting developers in the creation of state-of-the-art AI agents for use in almost every industry.

Customers can access Gemini through Vertex AI, a unified, fully-managed platform for creating, deploying, and tracking machine learning models at scale offered by Google Cloud. Customers can tweak and implement Gemini and other AI models with enterprise-ready tuning, grounding, monitoring, and inference capabilities, along with industry-leading AI infrastructure and user-friendly tooling to create AI agents, thanks to Vertex AI’s ability to support both generative and predictive AI models.

Cutting-edge performance

Enterprise clients can purchase the following Gemini models from Vertex AI:

Gemini 1.5 Pro

With only a single prompt, Google Cloud’s customers can now accurately parse massive documents, codebases, or entire films thanks to the industry-leading and groundbreaking context window of 1 million tokens that Google Cloud introduced earlier this year with Gemini 1.5 Pro, which is now generally available. Customers will soon be able to test Gemini 1.5 Pro with up to a 2 million token context window for use cases that demand an even larger context window, such as analysing very large code bases or extensive document libraries (join the waitlist for the 2 million token context window by signing up here).

Flash Gemini 1.5

Additionally, it is now GA and provides Google Cloud’s ground-breaking 1 million token context window. Unlike 1.5 Pro, it is lightweight and made to function quickly and scaleably for jobs like chat applications.

Gemini 1.0 Pro

Developed to manage multiturn text and code chat, code creation, and natural language jobs. A fresh version with supervised tuning capabilities and reduced latency and better quality is usually released.

Pro Vision Gemini 1.0

accepts prompts in multiple modes. You can receive text or code responses to your prompt requests, and you can include text, photos, and video.

Vertex AI

Gemini can now be deployed and customised thanks to Vertex AI, enabling developers to create novel and distinctive apps that handle data from text, code, photos, and video. Developers using Vertex AI can:

Explore and utilise Gemini, or choose from a carefully compiled roster of over 130 models from Google, open-source, and other sources that satisfy Google’s exacting enterprise security and excellence criteria. Models are available to developers as simple-to-use APIs so they can include them into apps rapidly.

Tailor the behaviour of the model to a particular topic or company’s expertise. You may even modify the model weights as needed by using tuning tools to enhance training knowledge. Many tuning methods are offered by Vertex AI, such as distillation, adapter-based tuning like Low Rank Adaptation (LoRA), and prompt design. Furthermore, Google Cloud enable reinforcement learning from human feedback (RLHF) so that a model can be enhanced by obtaining user feedback.

Add tools to models to assist in customizing Gemini Pro for particular scenarios or use cases. Vertex AI Extensions and connectors enable developers to call functions within codebases, access data from external sources, and integrate Gemini Pro to external APIs for transactions and other tasks. Additionally, Vertex AI enables businesses to validate foundation model results against their own data sources, which enhances the precision and applicability of a model’s conclusions. Google Cloud provide businesses with the option to use both grounding with Google Search technology and grounding against their organised and unstructured data.

Utilise technologies designed specifically to manage and scale models in production, ensuring that applications can be readily deployed and maintained after they are produced. Automatic Side by Side (Auto SxS) is an automated tool that is available on demand for customers to compare models. In comparison to manual model evaluation, auto SxS is more economical and quicker. It may also be tailored to different job specifications to accommodate novel generative AI use cases.

Create AI bots in a low- or no-code setting. Developers of all machine learning skill levels may utilise Vertex AI Agent Builder to leverage Gemini models to produce compelling, production-ready AI agents in a matter of hours or days, as opposed to weeks or months.

Use Vertex AI’s safety filters, content moderation APIs, and other responsible AI tools to help developers make sure their models don’t produce offensive content. This will help you deliver innovation in a responsible manner.

With Google Cloud’s integrated data governance and privacy policies, you can help protect data. Consumer data is still in their control, and Google never utilises consumer data for model training. Vertex AI offers several tools, such as Customer Managed Encryption Keys and VPC Service Controls, to give clients complete control over their data.

Current developments

With a focus on scale and enterprise readiness, Vertex AI’s ongoing innovation aims to deliver the finest models available from Google and the industry, combined with an end-to-end model development platform and the ability to construct and deploy agents more quickly. Among the most recent product breakthroughs are:

Sending a huge number of non-latency sensitive text prompt requests in an incredibly quick manner is possible with batch API, which supports use cases like sentiment and classification analysis, data extraction, and description creation. It allows several prompts to be given to models in a single request, which speeds up developer operations and lowers expenses.

Customers may actively manage and utilize cached context data with context caching, which is in public preview this month. Putting long-context apps into production can be costly since processing costs rise with context length. Customers can save a lot of money by using Vertex AI context caching and its cached data.

Customers can specify Gemini model outputs in accordance with particular formats or schemas with controlled generation, which will be available for public preview later this month. Even under explicit directions, the majority of models are unable to ensure the syntax and format of their outputs. Customers can provide custom formats or select from pre-built alternatives like XML and YAML when using Vertex AI driven generation to determine the appropriate output format. JSON is available as a pre-built alternative.

The retrieval augmented generation (RAG) process is made simpler by LlamaIndex on Vertex AI, starting with data acquisition and transformation and continuing through embedding, indexing, retrieval, and generation. Customers of Vertex AI may now connect bespoke data sources to generative models by utilising LlamaIndex’s open-source, simple, and flexible data framework in conjunction with Google’s models and AI-optimized infrastructure.

Firebase just unveiled Genkit, an open-source Typescript/JavaScript framework made to make it easier to create, implement, and keep an eye on production-ready AI agents. Firebase developers may now benefit from text embeddings and Google models like Gemini and Imagen 2, made possible via the Vertex AI plugin.

Grounding with Google Search, which is now widely accessible, enables you to link models with current online material, a broad range of themes, or world knowledge. By integrating Google Search with Gemini models, Google Cloud provide users with access to the most recent foundation models from Google in addition to new, high-quality data, which greatly enhances answer accuracy and completeness.

The latest model in Google Cloud’s family of open models, Gemma 2, was developed using the same technology as Gemini for a wide range of AI developer use cases. Vertex AI Model Garden will soon have Gemma 2 models available.

Imagen 3, Google Cloud’s best text-to-image generation model to date, will soon be available at Vertex AI. It can generate images that are lifelike and photorealistic while maintaining an amazing level of detail.

How users are utilising Gemini models to innovate

API requests increased by almost 6X from H1 to H2 of last year, indicating that Vertex AI has been widely adopted. The remarkable things that customers are accomplishing with Gemini models truly astound them, especially considering how multimodal and adept they are at handling sophisticated thinking.

Samsung: The company just revealed that the Galaxy S24 series is the first to feature Gemini models in a smartphone. Customers can utilise summary tools in Notes and Voice Recorder, starting with Samsung-native applications. Samsung is certain that Vertex AI’s integrated security, safety, and privacy will safeguard its end consumers.

Jasper: Jasper uses Gemini models to swiftly produce marketing campaign content for its clients. Jasper is an AI marketing platform that helps enterprise marketing teams develop on-brand content and campaigns at scale. Teams may now work more quickly without sacrificing the quality of their material, as long as it still follows marketing and brand voice criteria.

Quora: The well-known question-and-answer website Quora is utilising Gemini to support creator monetization on Poe, their AI chat platform, which allows users to explore an extensive selection of AI-powered bots. Poe developers can now create unique bots for a range of use cases using Gemini, like as code generation, writing help, individualised learning, and more.

It gives them great pleasure to be listed as a Leader in both the The Forrester Wave: AI Infrastructure Solutions, Q1 2024 and The Forrester Wave: AI Foundation Models for Language, Q2 2024 reports. To the advantage of Google Cloud’s clients, Google Cloud have applied decades’ worth of AI R&D experience to developing models, ultra-scale infrastructure, and Vertex AI capabilities. Google Cloud are dedicated to continuing AI research and innovation.

Google has massive AI infrastructure, a large pool of AI researchers, and a growing number of enterprise customers to lead the AI market Google Cloud.” – Q2 2024 edition of The Forrester Wave: AI Foundation Models for Language

Post a Comment

0 Comments