Hold On, Holdcos; Rags To Riches | AdExchanger
Briefly

Hold On, Holdcos; Rags To Riches | AdExchanger
"AP Goes Ape For RAG If your biggest concern about AI is that it doesn't generate accurately formatted citations, then boy, oh boy do we have good news for you. Over the past year, the Associated Press has formatted its tens of millions of assets - including text, videos, photos and audio - into machine-readable content for LLMs. As demand for retrieval-augmented generation (RAG) takes off, publishers have more reason than ever to make their content accessible to LLMs."
"RAG, by the way, is a process whereby an LLM pulls data from knowledge bases outside of its training - like news sites and scholarly journals - before generating a response so as to optimize for accuracy. The new format for AP's archives will allow LLMs to "reliably ground, cite and pay for decades of reporting," per Digiday. Which all sounds well and good - as long as the "pay for" element doesn't fall by the wayside."
"The past few years have been hectic for agency holdcos. Dentsu is for sale. WPP is expected to take what Adweek calls "drastic measures" to reshape its portfolio in the lead-up to a potential sale. Omnicom and IPG are merging. Oh, and Havas and Horizon teamed up to offer greater scale. But these mergers, tie-ups and sell-offs will also undoubtedly reshuffle the market and leave business up for grabs."
Agency holdco consolidation has accelerated, with Dentsu for sale, WPP reorganizing, Omnicom merging with IPG, and Havas partnering with Horizon to increase scale. These mergers and sell-offs are reshuffling the market, creating conflicts that could prompt brands to reconsider agency relationships and open opportunities for midsize and independent agencies. Many brands prefer larger roles at smaller agencies rather than being minor clients at global holding companies. Concurrently, brands are evaluating how much agency work can be handled by AI at lower cost. Major publishers like the Associated Press are machine-formatting archives to support retrieval-augmented generation, enabling LLM grounding, citation and potential payment.
Read at AdExchanger
Unable to calculate read time
[
|
]