AI tools pivot to multimedia as data wars heat up

Relay’s abrupt exit

Relay announced its shutdown this week. The startup had promised a Chrome extension that would let users harness AI while they browsed. Jacob Bank, the founder and CEO, told staff, “We have some really ambitious plans to help you work with AI in Chrome to get things done, and I’ll have more to share soon.” He added that the remaining team will join Google’s Chrome group. The move signals that big browsers see enough value in AI assistance to absorb niche players. It also shows that the market can swallow small innovators without a fanfare.

Reddit tests audio‑video posts

Reddit began a limited rollout of video and audio versions of popular submissions. Users can now watch a clip or listen to a story instead of scrolling through text. The feature mirrors the short‑form video format that dominates TikTok. TechCrunch reported that the test targets high‑engagement posts, letting creators repurpose text into a media‑rich experience. Reddit hopes the addition will keep users on the platform longer. The experiment also gives the site a new data stream to feed its recommendation algorithms.

Amazon’s rare book cannibalization

Amazon disclosed that it is destroying rare books to train its large language models. The company treats the volumes as raw material for AI, arguing that the texts improve model quality. TechCrunch noted that the books are “incredibly valuable” for training because they contain language patterns not found in modern sources. Critics argue that the practice erodes cultural heritage for commercial gain. The move underscores how tech giants are mining every available text, even irreplaceable artifacts, to stay ahead.

The data‑format arms race

The three moves share a common thread: AI teams are racing to control both the data that trains models and the formats that deliver output. Browser extensions like Relay aim to embed generative tools directly into the user workflow. Reddit’s audio‑video test expands the consumption side, turning static posts into multimodal signals. Amazon’s book‑pulping illustrates the supply‑side pressure, where companies harvest obscure corpora to fill gaps in their training sets.

Historically, AI research relied on publicly available web dumps. As models grew larger, the marginal gain from obscure sources increased. Rare books, academic papers, and niche forums now fetch premium prices on the data market. At the same time, user expectations have shifted toward rich media. Platforms that can serve AI‑generated video or voice stand to capture more attention and ad revenue. The convergence of data acquisition and multimodal delivery creates a feedback loop: better data fuels richer media, which in turn generates more user‑generated data for future training.

Regulators have started to notice. Data‑privacy laws already restrict how personal information can be scraped. The destruction of cultural artifacts raises separate heritage‑protection concerns. If the trend continues, lawmakers may face pressure to classify certain texts as public‑good assets, exempt from commercial extraction. Tech firms will need to balance the lure of performance gains against legal and reputational risk.

What to watch

Watch for Google’s next Chrome AI rollout. The integration will reveal whether the browser can become a primary AI interface. Track Reddit’s rollout metrics; a surge in watch time could prompt competitors to follow suit. Monitor Amazon’s procurement policies for rare books; any shift toward licensing rather than destruction would signal a response to backlash. Together, these signals will map the next frontier of AI‑driven content creation and consumption.