Daily AI Newsletter: Google preps ‘Jarvis’ AI agent that works in Chrome

In today’s email:

🔥 Meta releases an ‘open’ version of Google’s podcast generator
🤑 Apple will pay security researchers up to $1 million to hack its private AI cloud
📚 What is a Raspberry Pi AI Kit? Everything You Need to Know
🧰 9 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

Google preps ‘Jarvis’ AI agent that works in Chrome

Google appears to be advancing its efforts to create intelligent web automation with “Project Jarvis”, a new AI agent expected to be showcased by the end of this year. Initially introduced at Google I/O 2024 as part of its Gemini AI platform, Jarvis aims to support users in completing various online tasks autonomously, including research, product purchases, and travel bookings, directly within the Chrome browser. This development suggests a shift toward consumer-focused AI, designed to streamline daily tasks under user supervision, rather than corporate or enterprise solutions.

Sundar Pichai, Google’s CEO, describes these agentive systems as being capable of “reasoning, planning, and memory” and designed to work independently across software platforms to accomplish tasks efficiently. At I/O, Pichai demonstrated how Gemini and Chrome might integrate, envisioning future scenarios where users can delegate complex tasks like organizing and synthesizing information online. This goal aligns with Jarvis’s capabilities, which focus on supporting users in completing multi-step processes, all while adapting to different software environments.

According to reports, Jarvis functions by taking periodic screenshots of a user’s computer screen and interpreting each image before proceeding to the next action, such as clicking buttons or entering text. This deliberate, step-by-step operation requires several seconds per action, indicating that Jarvis may currently rely on cloud processing rather than running directly on users’ devices. This slower processing suggests that Jarvis is still in early stages, with early testing expected soon, and a full release likely further in the future.

Gemini 2.0, Google’s latest AI model, powers Project Jarvis, giving it the necessary advancements to handle more sophisticated and contextual actions. With the prospect of an official preview in December, Google continues its tradition of using new models to power flagship features, underscoring its commitment to embedding AI deeply into everyday tools like Chrome.

Your Brand, Your Chatbot: Boost Engagement with Splutter AI

Struggling to keep your website visitors engaged? 🤔 Splutter AI lets you seamlessly add a fully customizable, AI-driven chatbot to your site—no coding required 🛠️. Tailor every aspect to match your brand, from layout to conversational flow!

Trained on your website, with features like voice chat 🎤, automated bookings 📆, and lead capture 🧲, learn about your customers & increase conversion rates! Whether you are aiming to automate bookings, capture leads, or provide personalized recommendations 🎯, Splutter AI adapts to your needs, so why wait? Get Started Now!

Ready to transform your website into an interactive experience? Use Promo Code SUPERPOWERDAILY for 30% Off Hobby & Business Plans. 🏷️

Try Splutter AI Free

OpenAI denies it’s releasing a model called ‘Orion’ this year

OpenAI is preparing to launch its next major AI model, code-named Orion, by December. According to The Verge, Orion will not be widely accessible through ChatGPT initially, unlike previous models like GPT-4o and o1. Instead, OpenAI plans to give early access to select partner companies, allowing them to build their own products and features using the new model. Microsoft engineers are also reportedly preparing to host Orion on Azure as soon as November. However, OpenAI has not confirmed if Orion will be branded as GPT-5. The company cautioned that these plans might change, and after CEO Sam Altman called the story "fake news," OpenAI stated they do not have plans to release a model called Orion this year but hinted at other upcoming technologies.

Orion has been described by an OpenAI executive as potentially 100 times more powerful than GPT-4 and is intended to advance towards artificial general intelligence (AGI). This new model is distinct from the o1 reasoning model, known as Strawberry, which OpenAI released in September. Reports also suggest that Strawberry was used to generate synthetic data for Orion’s training process. The completion of Orion's training was marked by a celebratory event within OpenAI in September, coinciding with a cryptic social media post by Sam Altman hinting at "winter constellations," believed to be a reference to Orion.

Orion's launch comes at a pivotal time for OpenAI, following its historic $6.6 billion funding round, which necessitated restructuring as a for-profit entity. The company is also undergoing significant leadership changes, with key figures such as CTO Mira Murati, Chief Research Officer Bob McGrew, and VP of Post Training Barret Zoph recently announcing their departures. Despite these challenges, OpenAI remains focused on pushing the boundaries of AI and advancing towards its long-term goal of AGI.

Meta releases an ‘open’ version of Google’s podcast generator

Meta recently introduced NotebookLlama, an "open" version of Google’s podcast-generating feature found in NotebookLM. Leveraging Meta's Llama models, NotebookLlama is capable of creating conversational, podcast-style digests of uploaded text files, such as PDFs of news articles or blog posts. This technology aims to replicate the interactive storytelling aspect of Google's viral tool, adding its own elements of dramatization and interruptions to make the content sound more dynamic.

The process starts with NotebookLlama generating a transcript from the uploaded file. It then enhances the text by incorporating more dramatic interactions before converting it into audio using open-source text-to-speech models. However, the quality of these generated podcasts currently lags behind that of NotebookLM, as the voices often sound robotic and tend to overlap awkwardly. Meta researchers acknowledged these shortcomings, pointing out that more advanced text-to-speech models could significantly improve the naturalness of the output.

The developers also mentioned that the current approach—using a single model to create the podcast outline—might be improved by adopting a debate-style format between two AI agents to provide a richer discussion. This, they believe, could lead to a more compelling and coherent podcast narrative. Despite this innovation, like other AI-generated content, NotebookLlama is still susceptible to the common "hallucination" problem, meaning that some parts of the generated podcast may include inaccurate or fabricated information.

While NotebookLlama isn’t the first attempt to emulate NotebookLM's podcast feature, its open-source nature and Meta’s involvement make it a notable entrant in this emerging space. The potential for improving model quality and adopting new approaches gives NotebookLlama room for future advancements, even if it hasn’t yet achieved a completely natural output.

Other stuff

How a Mumbai Drugmaker Is Helping Putin Get Nvidia AI Chips
What is a Raspberry Pi AI Kit? Everything You Need to Know
AI models fall for the same scams that we do
Google’s DeepMind is building an AI to keep us from hating each other
Apple will pay security researchers up to $1 million to hack its private AI cloud
‘They wish this technology didn’t exist’: Perplexity responds to News Corp’s lawsuit
Can you make ChatGPT generate an image of a glass of wine that is full to the brim?

Win 1 year free Superpower ChatGPT Pro🤑

All your ChatGPT images in one place 🎉

You can now search for images, see their prompts, and download all images in one place.

Get the extension for FREE

Editor's Pick

Voice Design by ElevenLabs - Generate a custom voice based on a text prompt

Arcade AI - turn your thoughts into things

Writer RAG tool: build production-ready RAG apps in minutes

RAG in just a few lines of code? We’ve launched a predefined RAG tool on our developer platform, making it easy to bring your data into a Knowledge Graph and interact with it with AI. With a single API call, writer LLMs will intelligently call the RAG tool to chat with your data.

Integrated into Writer’s full-stack platform, it eliminates the need for complex vendor RAG setups, making it quick to build scalable, highly accurate AI workflows just by passing a graph ID of your data as a parameter to your RAG tool.

Learn more about our production ready RAG tooling here.

Omni AI - Next-gen document OCR