With more and more LLM-powered products with prompts as their IP or USP. It is becoming more and more important to protect their prompts from leaking to competition. Malicious prompt injection has also become a thing motivating people to misuse prompt based apps.
SECURITY
-
Protecting LLM Prompts: IP Security and Injection Risks
By
–
With more and more LLM-powered products with prompts as their IP or USP. It is becoming more and more important to protect their prompts from leaking to competition. Malicious prompt injection has also become a thing motivating people to misuse prompt based apps.
-
Prompt Security Experts Emerging as Full-Time Career Role
By
–
Hot take: "Prompt Security Experts" will soon be a full-time job!
-
Bypassing AI Detection: Natural Text and Watermarking Constraints
By
–
I'm not sure about that. E.g. tricking the ZeroGPT algo means actually writing more natural text (so it'll be better, not worse), and tricking the watermarking means not being limited by the constraints of the watermarking algo.
-
AI-generated content detection methods easily circumvented
By
–
Me reitero en lo dicho, cualquier opción de detectar contenido generado por IA actualmente puede ser fácilmente contratado.
-
Invisible Watermarking Systems Easily Defeated by Adversarial Methods
By
–
Por ejemplo, me estáis compartiendo un vídeo donde se propone marcar los textos de ChatGPT de forma invisible para poder detectar que es artificial. El problema, como comenta aquí Jeremy, es que estos sistemas una vez creados, son muy fáciles de superar.
-
Blurring Lines Between AI and Human-Generated Content Online
By
–
Por otro lado, las fronteras que intentamos poner entre contenido AI/Humans son cada vez más difusas, creando situaciones delicadas como estas, donde en Reddit a un usuario se le ha baneado bajo la sospecha de que su arte (que él dice es original suyo) puede estar hecho con IA…
-
LLM Detection Methods Face Fundamental Technical Challenges
By
–
Am I missing something, or are all these attempts at recognising LLM outputs obviously destined to fail?
— Jeremy Howard (@jeremyphoward) 5 janvier 2023
It's dramatically easier to train an LLM for rewording a text than creating the text in the first place; then add a loss func that incorporates detection avoidance. https://t.co/TO0zIGHoAeAm I missing something, or are all these attempts at recognising LLM outputs obviously destined to fail? It's dramatically easier to train an LLM for rewording a text than creating the text in the first place; then add a loss func that incorporates detection avoidance.
-
Data Safety and Persistence Risks in AI Systems
By
–
Only question remaining is how safe is this? Is there any risk the data might disappear?
-
Security Paper Review Process Demands Increased Burden Reviewers
By
–
Besides the fact Im getting older, some reasons last year was more tiring:
– security papers are long! 13 two-column pages (vs 8 in ML)
– also, their revision process requires more commitment from reviewers
– more generally, labour expected from reviewers per paper is on the rise