惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

S
SegmentFault 最新的问题
Google DeepMind News
Google DeepMind News
G
Google Developers Blog
Martin Fowler
Martin Fowler
MongoDB | Blog
MongoDB | Blog
月光博客
月光博客
Jina AI
Jina AI
宝玉的分享
宝玉的分享
人人都是产品经理
人人都是产品经理
D
DataBreaches.Net
V
V2EX
WordPress大学
WordPress大学
T
The Blog of Author Tim Ferriss
Last Week in AI
Last Week in AI
B
Blog
博
博客园 - 叶小钗
小众软件
小众软件
Stack Overflow Blog
Stack Overflow Blog
P
Proofpoint News Feed
A
About on SuperTechFans
J
Java Code Geeks
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Y
Y Combinator Blog
Microsoft Security Blog
Microsoft Security Blog

Risky Business Features

How to launder illicit Bitcoin Hunting software supply chain malware Who gets to hack the hackers? How Brian Krebs doxxed TeamPCP James Kettle on inventing new attack techniques with LLMs How private LLM inference actually works Benchmarks, borders and the true cost of AI regulation Fortibleed: The bleeding edge of AI cybercrime What to do 'til the bugpocalypse gets here Mythos on your desk? Using local LLMs for code reviews Pitching security startups to VCs in the AI era How using open weight models can blow up in your face The state of the art in AI model jailbreaks Why NPM v12 won’t stop supply chain attacks Everything is getting much worse, much faster Solo podcast: A deep dive on TeamPCP How to survive supply chain attacks How the CopyFail disclosure went sideways NCSC’s Ollie Whitehouse on surviving the "bugpocalypse" What a great agentic AI deployment plan looks like Mythos smythos! How to find 0day with lesser models Solving the AI agent identity problem Feature Interview: Nicholas Carlini, Anthropic A builder's perspective on Mythos and frontier models Mythos and 0day: Fixing exploits is not safety Mythos and 0day: A hacker’s perspective What happens after North Korea infiltrates? Why CISOs need to be more flexible in the AI era A Risky Biz Experiment: Hunting for iOS 0day with AI Interview: Former NSA and CIA cyber leaders on offensive AI
A deep dive on AI model distillation attacks
James Wilson · 2026-04-29 · via Risky Business Features

Risky Business Features Podcast

April 29, 2026

Presented by

James Wilson

James Wilson

Technology Editor

In this solo episode of Risky Business Features James Wilson explores how distillation techniques are both a legitimate way to train smaller models, as well as a way to steal model capabilities. It’s not just a problem for frontier labs! Any LLM-based product could have its competitive advantage stolen through these attacks.

James covers:

  • High-level concept of distillation
  • Why it matters including close/open-weight/open-source explanation
  • Types of distillation and the prompts used
  • The distillation pipeline end to end
  • Distillation at scale and mitigation techniques
  • Hardware resource constraints for distillation

Your browser does not support the audio element.

A deep dive on AI model distillation attacks

0:00 / 72:08

Logo

Show notes

Self-Instruct: Aligning Language Models with Self-Generated Instructions

Alpaca: A Strong, Replicable Instruction-Following Model

Vicuna: An Open-Source Chatbot Impressing GPT-4 with 90%* ChatGPT Quality

Orca: Progressive Learning from Complex Explanation Traces of GPT-4

Zephyr: Direct Distillation of LM Alignment

Stealing Part of a Production Language Model

Microsoft probes if DeepSeek-linked group improperly obtained OpenAI data, Bloomberg News reports

Detecting and preventing distillation attacks

此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。