Next, we have an academic study with the obvious finding that Facebook’s interest category targeting is often wrong. Among other things, Facebook doesn’t differentiate positive from negative sentiment when inferring interest. Speaking of Facebook, there are reports that they paid a right-wing lobbying agency to plant negative stories about TikTok. Not exactly obvious, but not surprising, either.
Hi, it’s Jamie, Senior Intern at the CDP Institute Department of the Obvious. The entire staff was trapped last night when someone forgot the key at an Escape the Room event, so I’m writing today’s newsletter. First, this survey from The Harris Poll for NetReputation.com found 76% of consumers won’t buy products with a “one-star or less” online rating. What’s not obvious is why the other 24% would.
Finally, Thales Group offers a study with the delightfully obvious conclusions that companies do a poor job of protecting sensitive data and that remote work makes things even worse. The problem is certainly obvious to corporate leaders: 28% said they wouldn’t trust their own personal data to their organization. Hey, they may be inept but they’re not stupid.
Unauthorized training data isn’t an existential threat to generative AI but it’s certainly a headache for users and developers alike. Most developers are trying to exclude materials that creators have explicitly labeled as unauthorized and citing “fair use” as justification for copying everything else. Getty Images has taken an opposite approach, building its gen AI tool only on materials that are explicitly licensed. It’s possible that tracing the provenance of training data will become a standard, similar to how organic food producers trace the origins of their ingredients.