AI Briefing: As DeepSeek catches new heat, Big Tech trumpets another quarter of AI earnings

On the company’s Q4 2022 earnings call with investors, CEO Sundar Pichai mentioned the ability to “integrate more direct LLM-type experiences” in search and other uses: “We’ll pursue this work boldly, but with a deep sense of responsibility, with our AI Principles and the highest standards of information integrity at the core of all our work.”

Fast-forward, and what was once just a preview of the future has become embedded in the present across more products, more users and with more ads. 

Last week, Google reported Q4 2024 search revenues of $54.03 billion — an increase of 12.5% year over year. (In a research note about Alphabet’s results, William Blair analysts mentioned AI Overviews having “comparable monetization.”)

Balancing the responsible development of AI and avoiding risks is likely to remain a delicate dance, especially as companies look to experiment more without any regulatory guardrails. The same week as its earnings, Google dropped its pledge to not use AI for surveillance, weapons “likely to cause harm.”

Google’s just one of the companies that mentioned ways they’re using AI or seeing AI help drive results. (Others this month include Meta, Microsoft, and Amazon.) Here are some other examples from last week:

• Pinterest noted LLM integrations and other AI updates helped improve guided search and visual search while also helping drive increased ad spend.
• Snap said more than 400 million users engaged with its “Gen AI Lenses” more than 4 billion times in Q4. It also introduced dozens of new developer tools, including a new AI-powered API that helps turn 2D images into 3D images.
• Outbrain also mentioned generative AI in its latest filing while citing plans to benefit from AI through its acquisition of Teads.
• Omnicom CEO John Wren mentioned the company’s Omni platform uses “high-fidelity data sets” and “custom AI tools” to help plan, create, target, and optimize campaigns.
• Mondelez’s report mentions the use of AI but also references associated brand safety risks. “Our brands may be associated with or appear alongside harmful content, including outputs from generative artificial intelligence models, before these platforms or our own social media monitoring can detect this risk to our brand.”

AI legal updates this week

Speaking of AI and earnings, The New York Times also disclosed the price (so far) of its ongoing copyright lawsuit against OpenAI. In its quarterly results, the Times mentioned the legal battle racked up litigation costs of $3.21 million in Q4, with total costs of $10.8 million for all of 2024.

Both AI innovation and litigation came up in News Corp’s quarterly results. In a press release, the company said it’s been “providing priceless content for Generative AI” while also being “vigilant in our pursuit of degenerative AI.” News Corp also mentioned DeepSeek’s sudden rise is “a salutary lesson for all AI players,” but that the Chinese startup still lacks the immediacy required for news: “Ultimately, content will be king in the world of AI.”

“We are pleased with our partnership with OpenAI and hope that other companies in the segment take a similarly enlightened approach,” CEO Robert Thomson said in News Corp’s statement about its earnings. “Our legal action against the perplexing Perplexity is underway and we look forward with relish to document discovery.” 

There were also more legal filings last week in the copyright infringement lawsuit against Meta related to the use of training data for its Llama AI models. In one document, plaintiff attorneys claim internal Meta records show Mark Zuckerberg knew about employee efforts to train Llama 3 using LibGen, a controversial dataset known to have pirated content including IP owned by some of the writers that filed the lawsuit.

Other documents show employees questioning the legality of using pirated content and asking whether to hide actions using a VPN. One internal message shows a Meta employee saying that “torrenting from a corporate laptop doesn’t feel right” before adding the laugh-crying emoji. (Meta did not immediately respond to Digiday when asked for comment on the latest court filings.)

DeepSeek faces new scrutiny

After initial hype, some experts are finding DeepSeek’s already finding itself in a bit of hot water.

Cybersecurity researchers warn the Chinese startup’s chatbot app may pose security risks by sending device data back to China. One report also notes device data could be used to de-anonymize iOS app users when combined with a user’s IP address and mobile ad companies. 

Meanwhile, a bipartisan pair of U.S. lawmakers introduced a bill to ban U.S. government workers from using DeepSeek’s chatbot via official devices. The legislation, sponsored by Republican Darin LaHood and Democrat Josh Gottheimer, cites security researchers finding that DeepSeek’s code is linked to the Chinese Communist Party and could be shared with entities already banned by the U.S. Federal Communications Commission.

“We must get to the bottom of DeepSeek’s malign activities. We simply can’t risk the CCP infiltrating the devices of our government officials and jeopardizing our national security,” Gottheimer said in a statement. “…We’ve seen China’s playbook before with TikTok, and we cannot allow it to happen again.”

Prompts & Products: Other AI news and announcements

• Ad tech players think DeepSeek’s R1 and other open-source AI models create new ways to build their own AI offerings beyond the walled gardens.
• OpenAI and Hugging Face both debuted debuted new “deep research” tools that provide more ways to search and think with LLMs.
• Mistral, a French AI startup, debuted a new “Le Chat” chatbot app for iOS and Android devices to compete with rivals like ChatGPT.
• Perplexity announced it’s doing a “$1 million Question” giveaway as part of a Super Bowl Sunday sweepstakes.
• Meta announced AI updates for its Advantage Plus platform along with new transparency features for AI-generated ad products.
• OpenAI debuted its first rebrand, which features a new logo just days before it airs its first Super Bowl ad during the Super Bowl LIX.
• Google published its 2024 “Responsible AI” report, which came as the company also faced criticism for dropped its pledge to not use AI for surveillance, weapons or other ways that could “likely to cause harm.”
• The Canadian startup StackAdapt raised $235 million in equity funding to help grow its AI-based programmatic advertising business.
• The European Commission debuted a new family of open-source multilingual LLMs called OpenEuroLLM.
• Cloudflare announced its content delivery network will support AI-generated content standards governed by C2PA to help improve the authentic of online content.