🦝New Sneaky AI Jailbreak Exposed!

Reading time: 5 minutes  

 Newsletter | News | Tools | Grow My Tool  

Today we will discuss-

• 🎁Microsoft adds new AI perks to Copilot in 365

• 🫢Anthropic researchers discover new AI jailbreak technique

• 🎙️200+ artists rally against 'predatory' use of AI in music

• ⚙️9 amazing AI tools you might not have heard of

All this and more - Let's dive in!

👩‍🍳What’s cooking in the newsroom?

Microsoft 365's Copilot upgraded with GPT-4 Turbo and faster image generation

🕴️News - Microsoft is now offering priority access to GPT-4 Turbo to business users of its AI-driven Copilot assistant. This grants access to OpenAI's newest model within the Copilot mobile app, web, Windows, and Edge, included in the $30 per user per month fee for Copilot with Microsoft 365 subscription.

Previously, only Copilot Pro subscribers had access to GPT-4 Turbo, but now, businesses using Copilot through Microsoft 365 also have access. Plus, there will be no limits on the number of daily chats or conversation turns.

💼How will businesses benefit from this? GPT-4 Turbo will now be accessible within Copilot for Microsoft 365, allowing users to perform web searches for the latest public information as well as use it for various work-related tasks such as querying emails, documents, and meetings. This is particularly beneficial for businesses because GPT-4 Turbo can process up to 300 pages of text in a single prompt, making document analysis more efficient.

👨‍💼What's more? Microsoft is also upgrading image generation capabilities in Microsoft Designer for business subscribers of Copilot for Microsoft 365. This update allows users to create up to 100 images per day, a notable increase from the previous limit of 15. According to Microsoft, this feature, which uses OpenAI's DALL-E 3 model, will enable image generation requests to be "rapid," decreasing waiting time significantly.

Anthropic researchers find new way to trick AI into answering off-limit questions

🙊News - Researchers at Anthropic have discovered a new way to make AI answer questions it isn't supposed to. By asking numerous less-harmful questions first, they found that a large language model (LLM) can be persuaded to answer something like how to build a bomb. They have named the technique "many-shot jailbreaking." 

🕵️‍♂️What's the logic behind it? This vulnerability stems from the increased "context window" capacity in the latest generation of LLMs, enabling them to retain a larger amount of data in their short-term memory. 

Anthropic's researchers found that LLMs with larger context windows perform better on tasks with ample examples in the prompt. For instance, if a prompt contains numerous trivia questions, the model's answers improve over time. 

Unfortunately this phenomenon, known as "in-context learning," also extends to responding to inappropriate queries. Initially, the model refuses such requests, but after responding to numerous less harmful questions, it becomes more likely to comply with such requests.

🤷‍♂️So what's the solution? To tackle the problem, Anthropic's researchers found that limiting the context window helps, but it negatively impacts the model's performance. As a solution, they are developing methods to classify and contextualize queries before they reach the model. Additionally, they have published a paper to raise awareness within the AI community and prevent misuse of this technique.

Billie Eilish, Nicki Minaj and more musicians sign open letter calling out 'predatory' use of AI 

👨‍🎤News - Over 200 popular musicians have signed an open letter advocating for safeguards against the unethical use of artificial intelligence to replicate human artists' images, voices, and music. 

The signatories represent various music genres and generations, ranging from contemporary stars like Billie Eilish, J Balvin, and Nicki Minaj to iconic figures such as Stevie Wonder and REM. Notably, the estates of Frank Sinatra and Bob Marley have also lent their support to the cause.

📬What does the letter say? 

• The letter, issued by the Artist Rights Alliance advocacy group, calls on technology firms to commit to not developing AI tools that undermine human songwriters and artists.

• While acknowledging the potential benefits of AI in music, the letter emphasizes the need to prevent its predatory use to mimic artists' voices and images, violate creators' rights, and disrupt the music ecosystem. 

• It condemns efforts by large companies to substitute human-generated content with AI-generated material, resulting in decreased royalties for artists.

• Furthermore, it highlights concerns about artists' work being used without consent to train AI models, a practice that has sparked legal controversies for companies such as OpenAI in recent times.

🙂In conclusion - Well, the letter is thought-provoking, to say the least. Now, it's just a waiting game to see if and how the tech companies react to these calls for change.

🙆🏻‍♀️What else is happening?

• WHO unveils a digital health promoter harnessing generative AI for public health

• Microsoft is working on an Xbox AI chatbot

• Amazon offers free credits for startups to use AI models including Anthropic

• Yahoo acquires Instagram co-founders' AI news platform Artifact

• AI companies have consumed the entire internet to train their models and are now running out of data

🕰️Historic moments

• 1984 - Marvin Minsky and Roger Schank coined the term AI winter at a meeting of the Association for the Advancement of Artificial Intelligence, warning the business community that AI hype would lead to disappointment and the collapse of the industry, which happened three years later.

• 1997 - Sepp Hochreiter and Jürgen Schmidhuber proposed the Long Short-Term Memory recurrent neural network, which could process entire sequences of data such as speech or video.

• 2006 - IBM Watson was developed with the initial goal of beating a human on the iconic quiz show Jeopardy! In 2011, the question-answering computer system achieved this goal by defeating the show's all-time (human) champion, Ken Jennings.

👩🏼‍🚒Discover mind-blowing AI tools

1. Inspiro - An app that provides daily inspiration and motivation through a vast library of curated quotes from historical figures (Free)

2. Medgic - An AI-powered skin analysis tool for fast and accurate results (Free)

3. Voxify - A tool that allows users to create realistic voice-overs in multiple languages and accents ($4.99/month)

4. DatePlanner - A tool that helps users plan personalized and unique dates based on mutual interests and preferences ($19/month)

5. Softr AI - A tool that generates web applications from a single prompt ($49/month)

6. Lyro - An AI Chatbot that automatically answers common questions and perform recurring tasks so your team can focus on growing your business ($29/month)

7. Interactive Mathematics - A platform that provides math lessons and an AI-powered math problem solver to help students improve their math skills ($3.99/month) 

8. WellyBox - A tool that helps users track and manage their receipts and invoices ($9/month)

9. Contentinator - A Figma plugin that helps  generate text, images, and provides copywriting assistance ($5/month) 

💁‍♂️AI life hacks

• AI design and illustration - Tools like Dall-E 2 and Midjourney can create stunning visuals, even with minimal artistic skill. Use them to design marketing materials, product mockups, or even generate unique artwork to sell.

• AI-powered video editing - Platforms like Runway and Kapwing can automate time-consuming tasks like cutting, resizing, and adding effects. This frees you up to focus on the creative aspects of video production.

• AI-powered shopping tools - Make use of AI-powered shopping tools, such as Google Shopping and Amazon Product Finder. These sites can help you compare prices, find the best deals, and discover new products.