- Starter AI
- Posts
- Anthropic’s jailbreaking method, Introducing: Cosmic Man, GenAI for drug discovery
Anthropic’s jailbreaking method, Introducing: Cosmic Man, GenAI for drug discovery
We have found AI's weak spot.
Sthefania
April 03, 2024
Sponsored by
Hello, Starters!
Making sure that AI remains a safe tool and not a crazy machine that gives incoherent answers or helps bad actors is not an easy task. Thankfully, companies are working to avoid this scenario.
Here’s what you’ll find today:
Anthropic presents “Many-Shot Jailbreaking”
CosmicMan: A humanlike text-to-image generator
GenAI helps in the development of new drugs
WHO unveils its AI-powered digital health promoter
DALL-E enables image editing
And more.
🧑💻 Anthropic presents “Many-Shot Jailbreaking” (2 min)
Anthropic recently presented a study that uncovered "many-shot jailbreaking," a method that targets the large context windows of LLMs to manipulate their behaviour negatively. This method involves bombarding the model with question-answer pairs on harmful topics to bypass safety measures. According to the study, this technique affects not only Anthropic's models but also those of Google and OpenAI.
The ease with which safety measures can be circumvented calls for an industry-wide effort to address these vulnerabilities and ensure the safe development of AI. A particular concern is that this method is similar to in-context learning, so it requires a defence that doesn't hinder the learning capabilities of the models.
A team of researchers from Shanghai AI Laboratory has unveiled CosmicMan, a specialised text-to-image model that creates hyper realistic human images, unlike existing models, which often provide low-quality results and misalignments.
To achieve this accurate depiction, CosmicMan was trained on a vast dataset called "CosmicMan-HQ 1.0," which contains over six million images of humans, along with precise text descriptions, enhancing the level of detail.
Through CosmicMan, users can even adjust minimal aspects of their prompts to achieve the desired results, such as changing the colour of a hat.
Scientists from Stanford Medicine and McMaster University are using AI to create new medicines for fighting antibiotic-resistant bacteria. Their model, SyntheMol, helps design structures and chemical recipes for new drugs that target resistant strains of Acinetobacter baumannii.
The potential of AI to boost the discovery of new treatments and design effective drugs is a key focus of this research, as the technology assists in creating molecules that can bring significant results for the pharmaceutical and medical industries.
Work lesser & drive 10x more impact using AI
HIGHLY RECOMMENDED: A Power-packed workshop (worth $199) for FREE and learn 20+ AI tools to become 10x more efficient at your work.
👉 Become an AI Genius in 3 hours. Register here (FREE for First 100) 🎁
In this workshop you will learn how to:
✅ Simplify your work and life using AI
✅ Do research & analyze data in seconds using AI tools
✅ Automate repetitive tasks & save 10+ hours every week
✅ Build stunning presentations & create content at lightning speed
🏥In another demonstration of AI's impact on healthcare, the World Health Organization has announced the launch of a digital health promoter. This assistant leverages generative AI to assist users with multiple health topics, such as risk factors for diseases, well-being, and healthy habits.
🎨OpenAI has finally included editing features in DALL-E. These new functions can be accessed through the ChatGPT mobile app and desktop, enabling users to select specific areas of a generated image and easily edit them by removing or adding objects.
⚡️Quick links
What did you think of today's newsletter? |