filmov
tv
Stealing Part of a Production LLM | API protects LLMs no more
Показать описание
How it is possible to steal part of LLMs protected behind an API? 🥷 We explain both papers that made a breakthrough on this, one from Carlini et al. (Google), and the other one from Finlayson et al. (USC), see references below.
Thanks to our Patrons who support us in Tier 2, 3, 4: 🙏
Dres. Trost GbR, Siltax, Vignesh Valliappan, Michael, Sunny Dhiana, Andy Ma
Outline:
00:00 Stealing LLMs from behind API’s!?
01:54 AssemblyAI (Sponsor)
02:59 Two papers, same thing
04:03 Core observation
07:05 Recover Hidden Dimensionality
08:54 gpt-3.5-turbo
10:30 Full Layer Extraction
10:58 Extract all logits
14:35 Defenses
15:40 Cost of attack
16:22 Further impact
17:40 API response stochasticity
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
🔥 Optionally, pay us a coffee to help with our Coffee Bean production! ☕
Join this channel to get access to perks:
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
🔗 Links:
#AICoffeeBreak #MsCoffeeBean #MachineLearning #AI #research
Video editing: Nils Trost
Scientific advising by Mara Popescu
Thanks to our Patrons who support us in Tier 2, 3, 4: 🙏
Dres. Trost GbR, Siltax, Vignesh Valliappan, Michael, Sunny Dhiana, Andy Ma
Outline:
00:00 Stealing LLMs from behind API’s!?
01:54 AssemblyAI (Sponsor)
02:59 Two papers, same thing
04:03 Core observation
07:05 Recover Hidden Dimensionality
08:54 gpt-3.5-turbo
10:30 Full Layer Extraction
10:58 Extract all logits
14:35 Defenses
15:40 Cost of attack
16:22 Further impact
17:40 API response stochasticity
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
🔥 Optionally, pay us a coffee to help with our Coffee Bean production! ☕
Join this channel to get access to perks:
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
🔗 Links:
#AICoffeeBreak #MsCoffeeBean #MachineLearning #AI #research
Video editing: Nils Trost
Scientific advising by Mara Popescu
Комментарии