Security Threats to Your LLMs - 5 Critical Safeguards for 2024
We all know how LLMs such as ChatGPT has taken the world by storm. These advanced AI systems can generate human-like text on virtually any topic. But with their incredible capabilities comes serious risks if they fall into the wrong hands or aren't implemented properly. And if this happens customers of your business will lose faith in your application.
In this video we are going to dive into 5 critical safeguards that need to be in place to secure LLMs and prevent misuse. 2024 is shaping up to be the year LLMs go fully mainstream, so getting security measures right is absolutely vital.
My name is Joydeep. I have 13+ years of experience and working as a senior machine learning engineer in the R&D division of a fortune 500 company. In this channel I talk about GenAI careers and the industry.
⏱️ Timestamps
0:00 Intro
1:12 topmate showcase
1:30 Watermarking
4:31 Glitch tokens
8:05 Data and Model Poisoning
9:25 Output Validation
10:21 Prompt Injection
12:07 Conclusion
🔗 Links
HuggingFace Watermarking: https://huggingface.co/docs/transform...
On the Reliability of Watermarks for Large Language Models: https://arxiv.org/abs/2306.04634
Glitch Tokens in Large Language Models: https://arxiv.org/abs/2404.09894
Glitch Tokens - Computerphile: • Glitch Tokens - Computerphile
magikarp: https://github.com/cohere-ai/magikarp
Glitch Tokens twitter post: https://x.com/max_nlp/status/17888626...
SolidGoldMagikarp III: Glitch token archaeology: https://www.lesswrong.com/posts/8viQE...
Tay Poisoning: https://atlas.mitre.org/studies/AML.C...
Poison GPT: https://atlas.mitre.org/studies/AML.C...
30% of Google's Reddit Emotions Dataset is Mislabeled [D]: / 30_of_googles_reddit_emotions_dataset_is
Bad Labels: https://koaning.io/posts/labels/
Guidance: https://github.com/guidance-ai/guidance
Outlines: https://github.com/outlines-dev/outlines
Instructor: https://github.com/jxnl/instructor
Huggingface prompt injection model: https://huggingface.co/protectai/debe...
OWASP Prompt Injection: https://genai.owasp.org/llmrisk/llm01...
🔗 Career growth
Career Guidance in Machine Learning: https://topmate.io/joydeep_bhattacharjee
NLP basics: https://vibrantai.academy/courses/1/
Connect on LinkedIn: / joydeep-bhattacharjee-934a1157
Follow me on X: https://x.com/alt227Joydeep
👋🏻 About Me
My name is Joydeep Bhattacharjee. I am an AI engineer with 13+ years of experience. I talk about GenAI, career and AI industry.
Reach out to me: topmate.io/joydeep_bhattacharjee