OpenAI announces new o3 models

Hello, AI Enthusiasts!

Welcome to FavTutor’s AI Recap! We’ve gathered all the latest and important AI developments for the past 24 hours in one place, just for you.

In Today’s Newsletter: 😀 

• OpenAI announces new o3 models

• ChatGPT now works with more Coding Apps

• Now You Can Talk to AI Hosts in NotebookLM

• Meta's Apollo aims to crack the video understanding problem

OpenAI
😮 OpenAI announces new o3 models

OpenAI reports breakthrough performance with its new o3 reasoning model. The company attributes much of this progress to reinforcement learning, a well-established AI training method that offers significant advantages for certain tasks but has clear limitations for others.

Insights for you:

• OpenAI's o3 language model has made significant progress in benchmark results, particularly in programming and math tasks with clear right/wrong criteria, thanks to the use of reinforcement learning (RL) during training.

• The performance of o3 is further enhanced by increased computational power during model inference, allowing it to excel at tasks with well-defined solutions.

• However, the RL approach has limitations when it comes to more open-ended tasks without clear solutions, and the models still need to prove themselves in complex, real-world scenarios where problems are less clearly formulated than in benchmarks.

Drowning In Support Tickets? Maven AGI is here to help.

Maven AGI platform simplifies customer service by unifying systems, improving with every interaction, and automating up to 93% of responses. Seamlessly integrated with 50+ tools like Salesforce, Freshdesk, and Zendesk, Maven can deploy AI agents across multiple channels—text, email, web, voice, and apps—within days. Companies like Tripadvisor, ClickUp, and Rho slash response times by 60%, ensuring quicker support and exceptional customer satisfaction. Don’t let support tickets slow you down

Request a free personalized demo today.

OpenAI
👩‍💻 ChatGPT now works with more Coding Apps

ChatGPT for macOS can now work with your apps. In this early beta, you can let ChatGPT look at coding apps to provide better answers. The feature allows the AI to interact directly with compatible coding environments like VS Code.

Insights for you:

• ChatGPT can read content from your coding programs and IDEs including VS Code, Xcode, Terminal, and MatLab, bringing you more accurate answers tailored to your work. Now, it added more coding apps.

• New compatible programs include ap Nova, Script Editor, and TextMate; Jetbrains IDEs like Android Studio, AppCode, CLion, DataGrip, GoLand, IntelliJ IDEA, PHPStorm, PyCharm, RubyMine, RustRover, and WebStorm; as well as VS Code forks like VSCode Insiders, VSCodium, Cursor, and WindSurf. ChatGPT will also now integrate with the Prompt and Warp terminal apps.

• Enabling ChatGPT to work with VS Code requires installing a VS Code extension to query content.

Google
🎙️ Now You Can Talk to AI Hosts in NotebookLM

NotebookLM’s Audio Overviews just got more interactive: You can now use your voice to ask AI hosts for more details or explanations during a podcast. But note that this is an experimental feature and only works on new Audio Overviews.

Insights for you:

• Google is starting to roll out a big new feature: the ability to talk with the AI “hosts” of the overviews. Using your voice, you can ask the hosts for more details or to explain a concept differently.

• After a podcast is created, you can turn on "Interactive Mode" and hit "Join" to start chatting live with the AI hosts.

• The ability to talk with NotebookLM is a potentially useful way to learn more about what you’ve collected in the app.

Meta
💡 Meta's Apollo aims to crack the video understanding problem

Meta and Stanford researchers have developed Apollo, a new family of AI models that tackles one of AI's persistent challenges: getting machines to truly understand videos. The research reveals that how you train an AI model matters more than its size.

Insights for you:

• Meta and Stanford researchers have conducted an in-depth study to determine the best practices for designing AI models that excel at understanding videos.

• They found that using a consistent sampling rate, incorporating specialized components for individual frames and temporal relationships, and including timestamp information all improve performance.

• Informed by these findings, the researchers created Apollo, a family of AI video models that deliver top-notch performance across different levels of complexity, showcasing the effectiveness of the identified design principles.