Multimodality and managing unstructured data
Multimodality, a term gaining prominence with Google’s release of Gemini, involves combining different types of prompts—text, video, and audio—into a large language model (LLM) both as input and output.
Companies like Adobe’s Firefly are already exploring text-to-image capabilities. However, with multimodal approaches, brands can extend to include image-to-text capabilities, combining existing visual assets with predictive AI to accomplish tasks like image captioning, composing emails, generating ad content, and creating highly relevant ads.
“[This] is the next evolutionary step on the different ways one can use AI to either create outputs or leverage specific types of inputs,” said Jay Wilder, VP of product marketing, marketing cloud, Salesforce.
Product-driven brands, like retailers or cosmetics, will benefit from this. Someone who bought a faulty couch from Ikea can take a picture of a damaged product and use the app to seek the right assistance, with the AI providing instructions on how to fix the problem.
However, companies will first need to pay more attention to structuring their amorphous data, noted Wilder.
Some movement to regulate AI
The European Union was the first region to pass comprehensive AI rules, set to come into force in 2025. However, the hiatus before the EU law comes into effect, coupled with the election season in the U.S., means publishers, brands and ad-tech [partners need to be vigilant with challenges around privacy, intellectual property, and bias mitigation.”
The U.S. Department of Commerce could form an agency to audit AI firms to ensure they follow ethical standards and regulations that protect public interests, said Milan Kordestani, CEO at Ankord Labs.
“These audits might focus on evaluating the impact of AI technologies on societal stability, economic disruption, misinformation, assessing the impact of biased LLMs, and ensuring that large companies are not exacerbating social inequalities or compromising privacy and security,” said Kordestani.
Source link
