Embed Notice

HTML Code

<blockquote style="position: relative; padding-left: 55px;"><section><a href="https://infosec.exchange/users/josephcox/statuses/111494836974852957">Joseph Cox (josephcox@infosec.exchange)'s status on Thursday, 30-Nov-2023 02:37:16 JST</a><a href="https://infosec.exchange/@josephcox" title="josephcox@infosec.exchange"><img src="https://gnusocial.jp/avatar/163273-48-20230825063445.webp" width="48" height="48" alt="Joseph Cox" style="position: absolute; left: 0; top: 0;">Joseph Cox</a></section><article><p>New: researchers managed to poke ChatGPT in such a way that it spat out a ton of examples of its training data. In some cases this included personal identifiable information. That data used to train systems presumably without PII owners' knowledge <a href="https://404media.co/google-researchers-attack-convinces-chatgpt-to-reveal-its-training-data/" rel="nofollow noreferrer">https://404media.co/google-researchers-attack-convinces-chatgpt-to-reveal-its-training-data/</a></p></article><footer><a rel="bookmark" href="https://gnusocial.jp/conversation/2393043#notice-4728537">In conversation</a><time datetime="2023-11-30T02:37:16+09:00" title="Thursday, 30-Nov-2023 02:37:16 JST">about a year ago</time> <span>from <span><a href="https://infosec.exchange/@josephcox/111494836974852957" rel="external" title="Sent from infosec.exchange via ActivityPub">infosec.exchange</a></span></span><a href="https://infosec.exchange/@josephcox/111494836974852957">permalink</a><h4>Attachments</h4><ol><li><label><a rel="external" href="https://gnusocial.jp/attachment/1898722">Untitled attachment</a></label><br><a href="https://media.infosec.exchange/infosec.exchange/media_attachments/files/111/494/836/698/949/491/original/46984d8b59ffcf3e.jpg" rel="external">https://media.infosec.exchange/infosec.exchange/media_attachments/files/111/494/836/698/949/491/original/46984d8b59ffcf3e.jpg</a></li><li><article><header><div>Domain not in remote thumbnail source whitelist: www.404media.co</div><h5><a href="https://www.404media.co/google-researchers-attack-convinces-chatgpt-to-reveal-its-training-data/">Google Researchers’ Attack Prompts ChatGPT to Reveal Its Training Data</a></h5><div> from <span>@jason_koebler</span></div></header><div>ChatGPT is full of sensitive private information and spits out verbatim text from CNN, Goodreads, WordPress blogs, fandom wikis, Terms of Service agreements, Stack Overflow source code, Wikipedia pages, news blogs, random internet comments, and much more.</div><footer></footer></article></li></ol></footer></blockquote>

Corresponding Notice

Embed this notice
Joseph Cox (josephcox@infosec.exchange)'s status on Thursday, 30-Nov-2023 02:37:16 JST Joseph Cox
New: researchers managed to poke ChatGPT in such a way that it spat out a ton of examples of its training data. In some cases this included personal identifiable information. That data used to train systems presumably without PII owners' knowledge https://404media.co/google-researchers-attack-convinces-chatgpt-to-reveal-its-training-data/
In conversationabout a year ago from infosec.exchangepermalink
Attachments
1. Untitled attachment
  https://media.infosec.exchange/infosec.exchange/media_attachments/files/111/494/836/698/949/491/original/46984d8b59ffcf3e.jpg
2. Domain not in remote thumbnail source whitelist: www.404media.co
  Google Researchers’ Attack Prompts ChatGPT to Reveal Its Training Data
  from @jason_koebler
  ChatGPT is full of sensitive private information and spits out verbatim text from CNN, Goodreads, WordPress blogs, fandom wikis, Terms of Service agreements, Stack Overflow source code, Wikipedia pages, news blogs, random internet comments, and much more.

Public

Embed Notice

HTML Code

Corresponding Notice