登录查看更多内容

NOTEWORTHY NEWS: OUR TAKE ON THE LATEST AI/GENAI NEWS

4A's

Empowering your success.

发布日期: 2024年5月31日

+ 关注

Noteworthy AI News for Agencies - Fourth Edition

By Jeremy Lockhorn , SVP, Creative Technologies & Innovation, 4A's

THE BIG STORY:

Last week, among other things, we talked about the importance and advancing capabilities of multi-modality in AI models. Multi-modal (ICYMI) refers to a model’s ability to accept multimedia inputs such as text, voice and imagery and deliver similar multimedia outputs. This week we’ll go from multi-modal to multi-model. Many AI solutions - including platforms developed by large agencies and holding companies - leverage multiple AI models, often matching different use cases to specific models based on cost, speed, performance, and suitability to a particular task. Sometimes referred to as “model chaining” or “model routing,” the basic idea is to create a mashup of different models, and a front end to accept user queries that then get sent to the appropriate model depending on complexity and other factors.

Digiday and AdAge both covered WPP ’s recent addition of models from Anthropic to their platform, providing great insight into the AI decision-making process and strategy at a large holding company.

But in a world where some estimates suggest there are thousands models now available, how do you go about evaluating the capabilities of different models, especially when they are changing at breakneck pace? It’s a complicated question that must start with your business objectives, level of investment and what AI expertise (either internal or external) you have available. Selecting the right tools is crucial and complex, but benchmark and ranking tools provide at least one important input into this decision process. Tools from Hugging Face and LMSYS.org’s Chatbot Arena Leaderboard are cited by many AI experts, function like a sort of like a consumer report for LLMs, helping to compare and contrast different options based on real-world performance. Scale AI just launched their SEAL Leaderboards to rank LLMs using unbiased data and expert evaluation. And on the safety and security front, the National Institute of Standards and Technology (NIST) on Tuesday announced an extensive effort to try to test large language models (LLMs) “to help improve understanding of artificial intelligence’s capabilities and impacts.”?

Our GenAI Certification for Advertising program, launching next week on June 3, touches on the importance of this and provides expert guidance for agencies.?

OTHER INTERESTING AI NEWS FOR AGENCIES:

Last week we also talked about the AI-powered transformation of search and the inevitable strategy shifts required to adapt. HBR published a nice piece from researchers and entrepreneurs on how marketers can adapt.
Prompt Engineering is a necessary skill in an advertising landscape augmented by GenAI. There are a lot of different frameworks for writing great prompts to get the most of the models. This article covers off on two: “RACE” and “PARE”.
According to CX Network, 80 Percent of Top CX Trends Involve AI.
Retrieval augmented generation (RAG) can help inject first-party data into AI Models.
OpenAI has formed a new safety and security committee to replace the one it dissolved a few weeks ago.
AI models have favorite numbers, because they think they’re people - or at least they’ve been trained by people.
Mars and Omnicom on embracing generative AI: ‘This is a Wild West tool’
ADWEEK Readers Believe Gen AI Will Enhance Creativity Rather Than Diminish It
4 strategies to help creatives embrace AI rather than fear it.