Embed Notice

HTML Code

<blockquote style="position: relative; padding-left: 55px;"><section><a href="https://mstdn.social/users/jukkan/statuses/114358655721314064">Jukka Niiranen (jukkan@mstdn.social)'s status on Friday, 18-Apr-2025 20:51:46 JST</a><a href="https://mstdn.social/@jukkan" title="jukkan@mstdn.social"><img src="https://gnusocial.jp/avatar/263496-48-20241220141810.webp" width="48" height="48" alt="Jukka Niiranen" style="position: absolute; left: 0; top: 0;">Jukka Niiranen</a><div><a href="https://cyberplace.social/@GossiTheDog/114358443963093356" rel="in-reply-to">in reply to</a><ul><li></ul></div></section><article><p><a href="https://cyberplace.social/@GossiTheDog">@GossiTheDog</a> Once a human prompts Copilot to create a doc and then saves it to M365, that now becomes a human authored source. Next time that hallucinated BS will be attributed to a coworker, making it even harder for anyone to spot the error.</p><p>I've asked if there is anything built into <a href="https://mstdn.social/tags/Microsoft365" rel="tag">#Microsoft365</a> <a href="https://mstdn.social/tags/Copilot" rel="tag">#Copilot</a> to stop this feedback loop of AI hallucinations. No, there isn't.🤷♂️</p></article><footer><a rel="bookmark" href="https://gnusocial.jp/conversation/4914048#notice-9627411">In conversation</a><time datetime="2025-04-18T20:51:46+09:00" title="Friday, 18-Apr-2025 20:51:46 JST">about 4 days ago</time> <span>from <span><a href="https://mstdn.social/@jukkan/114358655721314064" rel="external" title="Sent from mstdn.social via ActivityPub">mstdn.social</a></span></span><a href="https://mstdn.social/@jukkan/114358655721314064">permalink</a><h4>Attachments</h4><ol><li><label><a rel="external" href="https://gnusocial.jp/attachment/4486000">Feedback loop of AI hallucinations, spreading AI prompted content to human authored source docs.</a></label><br><a href="https://media.mstdn.social/media_attachments/files/114/358/645/197/638/528/original/ec1b5fc883f1c7aa.png" rel="external">https://media.mstdn.social/media_attachments/files/114/358/645/197/638/528/original/ec1b5fc883f1c7aa.png</a></li></ol></footer></blockquote>

Corresponding Notice

Embed this notice
Jukka Niiranen (jukkan@mstdn.social)'s status on Friday, 18-Apr-2025 20:51:46 JSTJukka Niiranen
in reply to
- Kevin Beaumont
@GossiTheDog Once a human prompts Copilot to create a doc and then saves it to M365, that now becomes a human authored source. Next time that hallucinated BS will be attributed to a coworker, making it even harder for anyone to spot the error.
I've asked if there is anything built into #Microsoft365 #Copilot to stop this feedback loop of AI hallucinations. No, there isn't.🤷♂️
In conversationabout 4 days ago from mstdn.socialpermalink
Attachments
1. Feedback loop of AI hallucinations, spreading AI prompted content to human authored source docs.
  https://media.mstdn.social/media_attachments/files/114/358/645/197/638/528/original/ec1b5fc883f1c7aa.png

Public

Embed Notice

HTML Code

Corresponding Notice