OpenAI is making two notable changes to ChatGPT, which now reaches 1.2 billion people each week globally.
The company is adding invisible watermarks to eligible AI-generated text in the European Union while expanding its advertising system with visual ads during image generation.
The company announced the changes on Oct. 5, saying the text watermarking rollout responds to the EU AI Act, while the new visual advertising format is intended to give businesses another way to promote products and services inside ChatGPT.
ChatGPT Text Will Get Invisible Watermarks In The EU
Over the coming weeks, OpenAI plans to add an invisible watermark to eligible ChatGPT and Codex text output for users across all plans in the EU, where the service had 159.1 million average monthly active recipients.The technology, called textGrain, adds an invisible statistical signal through the model's word choices. OpenAI says a detector can analyze the resulting text to determine whether it contains an OpenAI watermark.
API customers worldwide can also opt in to watermarking for select models, although the feature will remain off by default for the API, OpenAI said in its news release.
OpenAI is initially limiting access to its text watermark detector to approved researchers and expert organizations.
The technology has important limitations. OpenAI's tests found that replacing 10% of words with synonyms reduced detection from about 92% to 66%, while replacing 25% reduced it to 17%. Shorter text and content with less flexibility in word choice, such as mathematics, can also be harder to detect.
OpenAI also says a watermark does not establish who owns text, who is responsible for it, how much a human contributed, or whether the content is accurate. A missing watermark does not prove that a person wrote the text either.
Anthropic has taken a similar approach with Claude. The company said future Claude models will generate watermarked text to help determine whether Claude was involved in producing it, citing the EU AI Act. Anthropic says its watermark does not identify a person, organization or conversation.
OpenAI Is Also Expanding ChatGPT Ads
Separately, OpenAI is testing a new visual advertising format that will initially appear during image generation in ChatGPT.The test is scheduled to begin later this month in the U.S. with an initial group of advertisers. OpenAI says the ads will be clearly labeled and kept separate from the image being generated. It also says advertising does not influence ChatGPT's answers.
OpenAI is expanding the measurement infrastructure around these ads at the same time. Integrations with Hightouch, Tealium and LiveRamp allow advertisers to send conversion data from their existing systems to ChatGPT Ads.
The company also lists attribution partnerships with AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin, along with full-funnel and advanced measurement partners Fospha, Measured and INCRMNTAL.
OpenAI is also working with Haus, Measured and WorkMagic on geo-based experiments intended to help measure the causal impact of advertising. Its brand-suitability work includes DoubleVerify and Integral Ad Science, which are developing evaluation pilots around OpenAI's advertising safeguards.
Some early results cited by OpenAI come from these partners. DV Rockerbox reported that WeightWatchers' attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark. WorkMagic reported statistically significant lift for wellness brand Dose, while Triple Whale reported that 93% of Portland Leather's visitors from ChatGPT Ads were new. These figures are partner-reported findings presented by OpenAI.
• Advertising is coming to AI chatbots — and it could influence the answers you get
• Apple’s Siri AI Can Access More of Your iPhone. Here’s How to Limit It
by AI Analysis via Digital Information World
No comments:
Post a Comment