Google Releases its First Open-Source LLM
Plus: AnyGPT is a major step towards artificial general intelligence, Google brings Gemini to Workspace.
Hello Engineering Leaders and AI Enthusiasts!
Welcome to the 216th edition of The AI Edge newsletter. This edition brings you details on Google’s first open-source LLM, Gemma.
And a huge shoutout to our amazing readers. We appreciate you😊
In today’s edition:
💡Google releases its first open-source LLM
🔥 AnyGPT: A major step towards artificial general intelligence
💻 Google launches Gemini for Workspace
📚 Knowledge Nugget: Why are LLMs so gullible? by
Let’s go!
Google releases its first open-source LLM
Google has open-sourced Gemma, a new family of state-of-the-art language models available in 2B and 7B parameter sizes. Despite being lightweight enough to run on laptops and desktops, Gemma models have been built with the same technology used for Google’s massive proprietary Gemini models and achieve remarkable performance - the 7B Gemma model outperforms the 13B LLaMA model on many key natural language processing benchmarks.
Alongside the Gemma models, Google has released a Responsible Generative AI Toolkit to assist developers in building safe applications. This includes tools for robust safety classification, debugging model behavior, and implementing best practices for deployment based on Google’s experience. Gemma is available on Google Cloud, Kaggle, Colab, and a few other platforms with incentives like free credits to get started.
Why does this matter?
With Gemma’s release, developers can now access an open-source LLM that packs the advanced capabilities and performance of models like Gemini and LLaMA into a compact model optimized for mainstream use. This democratization will finally diversify progress in the field of responsible AI development by enabling even smaller teams to build remarkable models.
AnyGPT: A major step towards artificial general intelligence
Researchers in Shanghai have achieved a breakthrough in AI capabilities with the development of AnyGPT - a new model that can understand and generate data in virtually any modality, including text, speech, images, and music. AnyGPT leverages an innovative discrete representation approach that allows a single underlying language model architecture to smoothly process multiple modalities as inputs and outputs.
The researchers synthesized the AnyInstruct-108k dataset, containing 108,000 samples of multi-turn conversations, to train AnyGPT for these impressive capabilities. Initial experiments show that AnyGPT achieves zero-shot performance comparable to specialized models across various modalities.
Why does this matter?
With its multimodal capabilities, AnyGPT can potentially mimic well-rounded human intelligence, unlike previous AI systems focused on specific tasks. If scaled up, AnyGPT's architecture could lead to machines that perceive, communicate, and reason about the world with the versatile intelligence of humans.
Google launches Gemini for Workspace to bring advanced AI to business users
Google has rebranded its Duet AI for Workspace offering as Gemini for Workspace. This brings the capabilities of Gemini, Google's most advanced AI model, into Workspace apps like Docs, Sheets, and Slides to help business users be more productive.
The new Gemini add-on comes in two tiers - a Business version for SMBs and an Enterprise version. Both provide AI-powered features like enhanced writing and data analysis, but Enterprise offers more advanced capabilities. Additionally, users get access to a Gemini chatbot to accelerate workflows by answering questions and providing expert advice. This offering pits Google against Microsoft, which has a similar Copilot experience for commercial users.
Why does this matter?
Google adding Gemini into Workspace products signals AI becoming an everyday tool for businesses. Like cloud computing made server access easy, putting powerful AI into common apps could make it a basic workplace utility, enhancing productivity in various critical but routine applications.
Enjoying the daily updates?
Refer your pals to subscribe to our daily newsletter and get exclusive access to 400+ game-changing AI tools.
When you use the referral link above or the “Share” button on any post, you'll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.
Knowledge Nugget: Why are LLMs so gullible?
In his intriguing piece,
explores why large language models can be so easily manipulated by tricks like "prompt injection" and "jailbreaking." He argues it stems from two key factors. First, their training process means LLMs rely entirely on examples from their training data. So if a trick isn't represented in that data, they have no frame of reference to recognize it as a trick. Second, LLMs operate in a constant state of confusion given the mishmash of data they train on. So they struggle to confidently challenge strange input, even if they know it's likely nonsense.
There are some partial solutions, like adversarial training. However, to have stronger defenses, we need breakthroughs that will help LLMs better tell the difference between real requests and obvious attempts to manipulate.
Why does this matter?
LLMs' current inability to reliably spot and challenge manipulation attempts poses dangers we cannot ignore. It means they could unknowingly provide instructions for making weapons, offer financial advice based on forged credentials, or generate toxic text. We need advances to make them less naive before deployment in sensitive contexts.
What Else Is Happening❗
🟦 Intel lands a $15 billion deal to make chips for Microsoft
Intel will produce over $15 billion worth of custom AI and cloud computing chips designed by Microsoft, using Intel's cutting-edge 18A manufacturing process. This represents the first major customer for Intel's foundry services, a key part of CEO Pat Gelsinger's plan to reestablish the company as an industry leader. (Link)
☠ DeepMind forms new unit to address AI dangers
Google’s DeepMind has created a new AI Safety and Alignment organization, which includes an AGI safety team and other units working to incorporate safeguards into Google's AI systems. The initial focus is on preventing bad medical advice and bias amplification, though experts believe hallucination issues can never be fully solved. (Link)
💑 Match Group bets on AI to help its workers improve dating apps
Match Group, owner of dating apps like Tinder and Hinge, has signed a deal to use ChatGPT and other AI tools from OpenAI for over 1,000 employees. The AI will help with coding, design, analysis, templates, and communications. All employees using it will undergo training on responsible AI use. (Link)
🛡 Fintechs get a new ally against financial crime
Hummingbird, a startup offering tools for financial crime investigations, has launched a new product called Automations. It provides pre-built workflows to help financial investigators automatically gather information on routine crimes like tax evasion, freeing them up to focus on harder cases. Early customer feedback on Automations has been positive. (Link)
📱 Google Play Store tests AI-powered app recommendations
Google is testing a new AI-powered "App Highlights" feature in the Play Store that provides personalized app recommendations based on user preferences and habits. The AI analyzes usage data to suggest relevant, high-quality apps to simplify discovery. (Link)
New to the newsletter?
The AI Edge keeps engineering leaders & AI enthusiasts like you on the cutting edge of AI. From machine learning to ChatGPT to generative AI and large language models, we break down the latest AI developments and how you can apply them in your work.
Thanks for reading, and see you tomorrow. 😊