Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More

by | News/Events

Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 1

Augmented Shelf | Issue 2 | March 19, 2024

Welcome to Augmented Shelf, a wrap-up of the week’s AI news, trends and research that are forging the future of work.

Evil Geniuses Vs. ChatDev

Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 2

To evaluate the vulnerability of LLM-based agents, researchers at Tsinghua University in Beijing, China, have introduced the Evil Geniuses attack method. Evil Geniuses autonomously generate malicious prompts related to the LLM agent’s original role using “Red-Blue” exercises to improve prompt aggressiveness while maintaining role similarity. When tested on agents like ChatDev, CAMEL, and MetaGPT, Evil Geniuses demonstrated high success rates in eliciting unintended harmful behaviors from the agents. This shows LLM-based agents can be manipulated to generate stealthy malicious content by exploiting their original roles and training.

Apple Buys DarwinAI

Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 3

Why did Apple buy AI systems operations trailblazer DarwinAI earlier this year? With innovations crucial for on-device AI rather than cloud reliance, the acquisition lines up with Apple’s growing focus on enhancing device performance and user experience through AI embedded directly into devices. And the notable addition of AI expert Alexander Wong from DarwinAI to Apple’s team is a clear move to strengthen its market position against competitive tech giants, especially with GenAI features slated for its upcoming iOS 18 and Xcode enhancements.

Ethan Mollick Reviews The Big 3

Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 4

In his latest One Useful Thing post, academic and AI influencer Ethan Mollick offers his take on the big 3 of AI at this moment: GPT-4, Claude 3 Opus, and Gemini Advanced. Mollick goes through each model’s unique characteristics (i.e., Claude 3 can be quite insightful) and their shared characteristics, (their shared, hauntingly lifelike interaction quality). With no instruction manuals in sight, Mollick argues that mastery of these LLMs lies in experiential learning. He ends by highlighting the new star of AI, the emerging concept of autonomous, goal-driven AI agents.

Forget Memory, LLMs Need a Forgettery

Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 5

What if language models could instantly learn new languages just by “forgetting” what they knew before? A novel AI model is turning that idea into reality. Researchers have pioneered “adaptive forgetting” to supercharge how AI language models learn and adapt. By strategically clearing an AI’s linguistic memory, then retraining it on new data, adaptive forgetting allows models to rapidly acquire new languages and reduce reliance on massive datasets. This mirrors human forgetting – discarding some details to solidify core knowledge.

Midjourney Introduces Character Consistency

Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 6

Midjourney’s latest “–cref” feature marks a leap forward in AI-generated art, providing creatives with a powerful tool for character consistency. By referencing a URL, users can maintain characters’ facial features, body type and clothing across a narrative sequence, while the “–cw” tag allows users to control the degree of variance from the original character in new images. Leveraging diffusion models, this update improves upon the generative AI’s capacity for narrative consistency. The ability to maintain character consistency in AI-generated imagery opens up new creative avenues and use cases for content creators, artists, and professionals in the entertainment industry, enhancing storytelling and visual coherence.

Augment Yourself 🤖

Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 7

Read more from Shelf

May 23, 2024RAG
Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 8 10-Step RAG System Audit to Eradicate Bias and Toxicity
As the use of Retrieval-Augmented Generation (RAG) systems becomes more common in countless industries, ensuring their performance and fairness has become more critical than ever. RAG systems, which enhance content generation by integrating retrieval mechanisms, are powerful tools to improve...

By Vish Khanna

May 23, 2024Generative AI
Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 9 Prevent Costly GenAI Errors with Rigorous Output Evaluation — Here’s How
Output evaluation is the process through which the functionality and efficiency of AI-generated responses are rigorously assessed against a set of predefined criteria. It ensures that AI systems are not only technically proficient but also tailored to meet the nuanced demands of specific...

By Vish Khanna

May 22, 2024News/Events
Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 10 Mannequin Medicine Makes Perfect, OpenAI’s Shifting Priorities, Google Search Goes Generative
AI Weekly Breakthroughs | Issue 11 | May 22, 2024 Welcome to AI Weekly Breakthroughs, a roundup of the news, technologies, and companies changing the way we work and live. Mannequin Medicine Makes Perfect Darlington College has introduced AI-powered mannequins to train its health and social care...

By Oksana Zdrok

Evil Geniuses Attack, The LLM Forgettery, Character Consistency and More: image 11
The Definitive Guide to Improving Your Unstructured Data How to's, tips, and tactics for creating better LLM outputs