Today’s top AI Highlights:

1. OpenAI’s AI search feature to search for real-time information on web with AI

2. Google’s opensource project Oscar for opensource developers

3. Google DeepMind’s AI models solve silver-medal standard questions of International Math Olympiad

4. Runway faces backlash, used YouTube videos without permission to train its GEN-3 AI model

5. Chain your AI agents together seamlessly with Flowise v2

& so much more!

Read time: 3 mins

Latest Developments 🌍

OpenAI’s AI Search Rivals Google’s AI Overviews 🌐

OpenAI has announced that it is prototyping “SearchGPT”. OpenAI working on an AI-powered search engine has been rumored for the past few months so it doesn’t come as a surprise. SearchGPT is OpenAI’s new search feature that combines AI with the web to give you fast and timely answers with clear and relevant sources. It’s been rolled out to a small group of users for feedback, you can apply for the waitlist here.

Key Highlights:

1. Search Experience - When you type your query, SearchGPT will give you synthesized AI-generated answers based on real-time information from the web. It combines search with conversation abilities, so you can ask follow-up questions to refine your results or go in depth.

2. In-line Citations - Responses have clear, in-line, named attribution and links so you’d know where information is coming from and can quickly engage with even more results in a sidebar with source links.

3. Publishers and Creators - OpenAI aims to highlight high-quality content in a conversational interface, so publishers don’t have to rely merely on search to reach users. Additionally, OpenAI will also let publishers manage how they appear in SearchGPT, giving them more choices.

4. Location for Precise Results - SearchGPT also uses your location to improve the accuracy of your results. You can choose to share more precise location information for better results.

5. Separate from Model Training - It is separate from training their generative AI foundation models. Sites can be surfaced in search results even if they opt out of generative AI training.

6. Opt Out of Training - Just like ChatGPT, OpenAI will use your conversations with SearchGPT to improve the product but you can opt out of it.

OpenAI is not coming just for companies like Perplexity AI, it’s also challenging Google and Microsoft. It can’t be just a coincidence that Google’s stocks plunged 3% the same day OpenAI announced SearchGPT 🤷‍♀️

Oscar AI Agent for Maintaining Opensource Architecture 🤖

Google’s new opensource project Oscar simplifies open-source software development by introducing AI agents to ease the workload of maintaining these projects. The project leverages LLMs not to replace coding, but to handle tedious maintenance tasks like processing issues and connecting questions to relevant documentation.

These agents excel at tasks like processing incoming issues, linking them to existing resources, and even identifying duplicate reports. This approach reduces the need for maintainers to constantly monitor and triage new submissions.

Key Highlights:

1. Contextual Search Engine - Oscar uses Google’s Gemini to analyze and understand the context of issues, documentation, and even code changes. This allows it to provide highly relevant information to contributors and maintainers.

2. Natural Language Control - Many open-source projects rely on a collection of tools for managing contributions, tracking bugs, and more. Oscar allows maintainers to interact with these tools using natural language, rather than complex commands or APIs.

3. In-depth Issue Analysis - Oscar doesn’t just surface related information; it tries to understand the issues itself. It can identify issues that require additional information from reporters, tag them based on content, and even utilize tools like git bisect to pinpoint potential bug introductions. This level of analysis helps maintainers prioritize their efforts and address issues more effectively.

4. Open and Adaptable - While currently in development within the Go project, Oscar’s architecture allows other open-source projects to integrate and customize this agent system for their specific needs and workflows, ensuring wider applicability.

Quick Bites 🤌

1. Google DeepMind’s AI models, AlphaProof and AlphaGeometry 2, solved four out of six problems from this year’s International Mathematical Olympiad (IMO), achieving the same level as a silver medalist in the competition for the first time.
   AlphaProof excels in algebra and number theory, while AlphaGeometry 2 focuses on geometry. AI progress in math reasoning can aid mathematicians in discovering new insights and solutions.

2. Updates from OpenAI

   1. Sam Altman has announced that they are rolling out their Voice Mode to plus subscribers next week!

   2. GPT-4o mini is available for developers to fine-tune, starting with Tier 4 and Tier 5 users. Also, the first 2 million training tokens a day are free till Sept 23.

3. The AI video generation company Runway copied data from thousands of YouTube videos, including popular YouTubers like Marques Brownlee, Casey Neistat, and Sam Kolder to train its GEN-3 model, as per a report by 404 Media.

4. Google released Gemini 1.5 Flash model, a lighter version of Gemini 1.5 Pro that closely matches its performance. They have expanded Flash’s access to unpaid users also on Gemini web and app, with an increased context window to 32k tokens.
   Google has also released a new feature with Gemini where you can access additional information on topics directly within responses in Gemini. Links to supporting content will be available at the end of a paragraph.

   

Tools of the Trade ⚒️

1. Flowise v2.0: Flowise AI has introduced a new sequential agentic workflow that allows chaining agents, implementing loopback mechanisms, integrating human approval steps, and creating conditional branches. This gives greater control over the flow compared to relying on chat with LLMs.

2. EDA-GPT: An open-source tool for automated data analysis, supporting various data formats like CSV, XLSX, and SQLite. It helps you visualize and analyze both structured and unstructured data, including PDFs and images, through an intuitive interface.

3. ComfyDeploy : Allows product teams to create AI video and image workflows using ComfyUI and deploy them instantly as APIs. It provides a managed platform, team collaboration features, and scalable cloud infrastructure.

4. Awesome LLM Apps: Build awesome LLM apps using RAG for interacting with data sources like GitHub, Gmail, PDFs, and YouTube videos through simple texts. These apps will let you retrieve information, engage in chat, and extract insights directly from content on these platforms.

Hot Takes 🔥

1. Jagged Intelligence
   The word I came up with to describe the (strange, unintuitive) fact that state of the art LLMs can both perform extremely impressive tasks (e.g. solve complex math problems) while simultaneously struggle with some very dumb problems.
   E.g. example from two days ago - which number is bigger, 9.11 or 9.9? Wrong. ~
   Andrej Karpathy
   

2. So many use-cases would open up if models were more 'opinionated' ~
   Matt Shumer

Meme of the Day 🤡

Everyone is waiting for OpenAI to release something groundbreaking today. Sama will come and give everyone a dose of patience once again. 😂

That’s all for today! See you tomorrow with more such AI-filled content.

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!

PS: We curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!