New Open AI Leak! (And Other AI News)
Matt Wolfe
27 min, 49 sec
A comprehensive summary of recent AI news, including GPT-4.5 leak, AI performance theories, AI journalism partnership, and various tech companies' AI updates.
Summary
- A potential leak suggests the upcoming GPT-4.5 will have advanced capabilities including audio, video, and 3D, but with higher API pricing.
- A theory suggests GPT-4 might perform worse in December, behaving lazier in responses, possibly mimicking internet activity patterns it was trained on.
- Axel Springer partners with OpenAI for AI journalism after previously being part of a coalition to sue AI firms for content usage.
- Google releases multiple AI tools including Notebook LM, Gemini Pro API, rebranded Music FX, and ImageIn for visual tasks.
- New updates and tools from tech companies like Meta, Stability AI, and Anthropics, and a sneak peek at Tesla's new Optimus Gen 2 humanoid robot.
Chapter 1
Speculation on a leaked GPT-4.5 featuring advanced capabilities, with concerns about higher pricing and reduced context window size.
- Leak found on Reddit shows GPT-4.5 with multimedia and reasoning abilities.
- API pricing is significantly higher than GPT-4, with a reduced context window from 128k to 64k.
- No confirmation from OpenAI, but the possibilities are intriguing.
Chapter 2
A theory suggests GPT-4's performance drops in winter, possibly due to training data reflecting reduced productivity in December.
- GPT-4 is theorized to work worse in December, potentially adopting a lazier trait from the training data.
- Experiments show changing the system date to May improved performance over December.
- The AI winter break hypothesis is gaining traction.
Chapter 3
Axel Springer partners with OpenAI for AI-enhanced journalism and ChatGPT Plus subscriptions reopen.
- Axel Springer media company partners with OpenAI to keep ChatGPT updated with the latest news.
- Notably, Axel Springer was previously part of a coalition to sue AI companies.
- OpenAI finds more GPUs, reopening ChatGPT Plus subscriptions.
Chapter 4
Google launches multiple AI tools including Notebook LM for document interaction, Gemini Pro API for developers, and Music FX for music generation.
- Google's Notebook LM allows document uploads for interactive search and summarization.
- Gemini Pro API is available for developers, initially free with a 32,000 context window.
- Music FX (previously Music LM) allows users to generate music based on prompts and settings.
Chapter 5
Google introduces ImageIn for visual tasks while Instagram and Snapchat release AI-powered editing features.
- Google's ImageIn for visual tasks specializes in text rendering, logo generation, and question answering about image details.
- Instagram introduces generative AI for background editing.
- Snapchat launches AI extend feature to fill in image gaps.
Chapter 6
Wirestock introduces Themes for custom image generation styles, and Mid Journey plans to release version 6.
- Wirestock allows creators to earn income by creating and sharing custom AI art styles.
- Mid Journey's upcoming version 6 will undergo a rating party to fine-tune model styles.
Chapter 7
Stability AI introduces a new 3D model and updates its pricing tiers for commercial use of its APIs.
- Stability AI's new model turns 2D images into 3D models.
- Company introduces pricing tiers for commercial API use, with discounts for those facing cost barriers.
Chapter 8
Anthropic releases an integration for Google Sheets with Claude, while a new open-source language model called Mixol shows promise.
- Anthropic's integration allows Claude to perform tasks within Google Sheets, though some users report issues.
- Mistol Ai's Mixol, an open-source model, uses a novel approach with multiple domain-specific experts to provide efficient responses.
Chapter 9
Meta releases Audio Box with various audio capabilities, and Tesla reveals the improved Optimus Gen 2 robot.
- Meta's Audio Box allows users to create and modify audio using AI, including creating unique voices and sound effects.
- Tesla's Optimus Gen 2 robot features smoother movements, better balance, and advanced tactile sensing.
More Matt Wolfe summaries
Massive GPT-4 Upgrades! (And How To Access Them)
Matt Wolfe
A comprehensive analysis of OpenAI's Dev Day event, discussing announcements, testing new features, and offering insights.
28 Amazing AI Tools That Are Surprisingly FREE!
Matt Wolfe
An extensive rundown of various free AI tools for a multitude of purposes.
Another MASSIVE Week in AI News (What's Going on?!)
Matt Wolfe
A detailed roundup of recent advancements and announcements in the AI industry, including new models, applications, and potential collaborations.
AI News: Get Ready, The World is About to Change
Matt Wolfe
The speaker discusses the presence and reception of AI at South by Southwest, a demo of Figure's humanoid robot with GPT-4 capabilities, a potential leak of GPT-4.5, changes in OpenAI's board, and various updates from AI companies.
7 Useful AI Tools NO ONE Talks About
Matt Wolfe
The video showcases a variety of lesser-known AI tools and features that can be useful for content creation, language translation, video editing, tutorial generation, marketing, writing, art generation, and drawing.