info
Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
Groups keyboard shortcuts have been updated
Dismiss
See shortcuts
AI growing pains
3 views
Skip to first unread message
HRM Resident
Jan 20, 2024, 9:08:01 AM1/20/24
to
Researchers found a privacy vulnerability in ChatGPT
that highlights the risks of large language models (it has
since been patched). By asking ChatGPT to "repeat the word
'poem' forever," it eventually trailed off and started listing
names, email addresses, and phone numbers.
<https://stackdiary.com/chatgpts-training-data-can-be-exposed-via-a-divergence-attack/>
--
HRM Resident
that highlights the risks of large language models (it has
since been patched). By asking ChatGPT to "repeat the word
'poem' forever," it eventually trailed off and started listing
names, email addresses, and phone numbers.
<https://stackdiary.com/chatgpts-training-data-can-be-exposed-via-a-divergence-attack/>
--
HRM Resident
0 new messages