OpenAI starts watermarking text
OpenAI’s text provenance plan adds an invisible statistical signal to generated text. API customers worldwide can now opt in for select models, while eligible ChatGPT and Codex output in the European Union will receive the watermark over the coming weeks.
Detection has practical limits. In one OpenAI evaluation, replacing 10% of the words in a 400-token passage with synonyms reduced detection from about 92% to 66%. The detector will initially be limited to approved researchers and expert organizations, and a missing watermark will not prove human authorship.
Visual ads enter ChatGPT
OpenAI will test visual ads during image generation later this month in the United States with an initial group of advertisers. The company says the ads will be labeled, remain separate from the generated image, and not influence ChatGPT’s answers.
Advertisers can connect Hightouch, Tealium, and LiveRamp to send conversion data from their existing systems to ChatGPT Ads. For users, the visible change is advertising inside some image-generation sessions.
Meta AI adds parental supervision worldwide
Parental supervision is now available globally in the Meta AI app. Parents and guardians can see the topics a teenager asked about during the previous seven days, but not the conversations themselves.
They can also switch from the default setting for ages 13 and older to a more restrictive Limited Content setting. Supervision can be configured in the Meta AI app or Meta’s Family Center.
Also
- Compare AI code reviewers GitHub’s public ReviewBench includes 219 pull requests across 19 languages, and builders can evaluate their own review agents with it.
- Catch more exposed credentials GitHub secret scanning now detects Lovable API keys, two Pydantic credential types, and two Supabase token types.
- Use voice and images for classroom expression Special-needs teacher Eni Wulandari describes students who cannot write speaking prompts to Gemini, generating images and expressing their ideas.
Pass it on