AI Briefing: Amazon’s new Nova models boost AI model efficiency, accuracy and variety across AWS

0 4 minutes read

AI Briefing: Amazon’s new Nova models boost AI model efficiency, accuracy and variety across AWS

By Marty Swant • December 6, 2024 •

Ivy Liu

One of the most buzzy bets in Las Vegas this week had nothing to do with poker or slots.

At the AWS Reinvent cloud conference in sin city, Amazon debuted a suite of six foundation models called Nova. The new models include a text-to-text model designed for speed, and three multimodal models that can understand and create text, images and videos.

Another model (Nova Canvas) was designed for generating studio-quality images and while a six model (Nova Reel) specializes in studio-quality videos.

Until now, Amazon’s focused on integrating AI models made by other companies including rivals like Google and Microsoft. However, the new models were designed for easy integration within Amazon Bedrock and support fine-tuning to improve accuracy. Other new features include automated reasoning checks to help improve AI model accuracy with guardrails to evaluate prompts to evaluate if an answer falls within the expected range.

“We’re not saying our models are better than everyone else’s … we’re saying they’re different,” Shaown Nandi, AWS Technology Director, told Digiday. “There’s some things they do well, some things they do differently. They’ll have really interesting price points we think people will be excited about. They’ll have different indemnification rules, so we’re going to offer value in those models…We’re making models with every type of development bedrock. We’re not picking winners or losers. We’re not picking ourselves.”

Nova’s arrival has marketing experts excited about Amazon catching up with the LLMs. One of them is Jack Smyth, a partner at Brandtech Group and its subsidiary agency Jellyfish. He said Nova might also help Amazon accelerate new features for other AI tools across Amazon like the e-commerce chatbot Rufus. However, he also wonders if Nova will further complicate things for advertisers already challenged by AI model complexities and varieties.

“It does make it really hard for advertisers to navigate,” Smyth said. “There is beautiful simplicity to Meta AI and Llama and the fidelity between what I get back from Llama. Of course, there’s some considerations about Meta AI, but there’s confidence there. Google and Gemini, despite all the jokes about complicated naming conventions, it’s a hell of a lot easier to understand than Amazon.”

Amazon Q gets a big boost

AWS also announced new features for Amazon Q including 50 new actions for business applications, new integrations for other companies’ platforms and ways to index data across apps. Companies already using the updates include the marketing agency Kepler and other companies like Asana, Zoom and GE Healthcare.

Noah Kershaw, Head of Product at Kepler, said the updates make it easier for AI models to connect and access various sources when analyzing both structured and unstructured data. For example, it helps Kepler teams save time when analyzing campaign performance and find insights across disparate data sets. He gave the example of a new staffer using Amazon Q to analyze Black Friday performance to see how targeted audiences have changed over several years.

“I really think the precursor to this is like how does the future help you to go beyond the dashboard so you’re not logged into the same interface again,” Kershaw said. “As you have the hypothesis, you just ask a question and then you get an immediate response back.”

New features for AI efficiency and accuracy

Other new AWS features focus on improving the speed, cost and accuracy of generative AI applications.

Nandi emphasized how AWS is shifting its approach to AI architecture from focusing on individual pieces to building AI inference — the process of generating insights and other outputs based on new data — into the foundation of Amazon’s offerings. That could also help companies worry less about which model to use for various tasks. That’s also aided by updates for model distillation, which involves transferring knowledge from larger models to smaller models.

“On the surface model distillation sounds like the really technical thing: You’re taking a bigger model and making it a smaller one,” Nandi said. “What’s interesting is it comes back to our theme of taking away stuff that we think is not differentiated for companies …If you can take a model and fine-tune it and make it specific for your use case, it’s going to be a lot cheaper and easier to run.”

Some of the updates noticed by Emarketer analyst Jacob Bourne include features like HyperPod Task Governance and Bedrock’s optimization features.

“AWS is betting that enterprises will prioritize cost-efficient, reliable AI deployment,” Bourne said. “Those initiatives alongside the rollout of multi-agent orchestration and automated reasoning capabilities show Amazon is focused on solving the industry’s central challenge: delivering high-performance AI that companies can actually afford to scale.”

AI news and announcements

OpenAI’s CFO told The Financial Times the company could explore introducing ads, but that it’d be “thoughtful about when and where we implement.”
Google announced expanded access for its new Veo AI video generator and its Imagen 3 image generator inside its Vertex platform.
Open AI announced its first CMO, Kate Rouch, who joins from Coinbase, where she led the crypto giant’s marketing for year — including its 2021 Super Bowl campaign. Other updates from OpenAI this week included the start of what it’s calling the “12 Days of Shipmas,” starting with Thursday’s debut of its new o1 reasoning model. (It’s also rumored that OpenAI could release its Sora AI model in the coming weeks.)
In a post-election report, Meta said AI content accounted for less than 1% of election-related misinformation. It also noted its AI image generator rejected 590,000 requests to create images of President-elect Trump, Vice President-elect JD Vance, Vice President Kamala Harris and Governor Tim Walz.
Google’s policy team published a new report about AI trustworthiness and how the company is thinking about content context and provenance.
Nieman Lab published its annual list of predictions, which includes what various journalists think will happen with AI and other parts of news in 2025.
A new report from Advertiser Perceptions said almost half of marketers surveyed trust AI to make campaign decisions, up from just 25% a year ago.
IPG announced the acquisition of Intelligence Node to bring more AI capabilities ot its ecommerce platform.