- The 79
- Posts
- OpenAI will never release the long-awaited o3 model
OpenAI will never release the long-awaited o3 model
Hello pals! Here’s what you need to know about AI today:
👉 OpenAI has a new roadmap for releasing models
👉 Adobe launches a new AI video tool to compete with OpenAI
👉 Apple is working on robots
and many more!
📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!
Read time: 5 minutes
OPENAI
OpenAI has too many models, and they know it's confusing
Source: The 79
What’s going on: OpenAI has decided to cancel the release of its anticipated o3 reasoning AI model in favor of a more streamlined approach with a unified next-generation model called GPT-5. Sam Altman announced this shift in a post on X, explaining that GPT-5 will integrate various OpenAI technologies, including features initially planned for o3, into its ChatGPT platform and API.
What does it mean: This decision marks a departure from earlier plans, as the company had indicated in December that o3 would launch early in 2025. Altman's post emphasized OpenAI's intent to simplify its product offerings and improve transparency regarding its development roadmap, acknowledging that the company's model and product lineup had become overly complex.
More details:
Before GPT-5's release, OpenAI plans to launch GPT-4.5 which is internally named “Orion” in the coming weeks, which will be the company's last non-chain-of-thought model.
GPT-5 will incorporate features like voice, canvas, search, and deep research.
Sam Altman admitted himself the problem of shipping too many AI models with confusing names by saying “We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten. We hate the model picker as much as you do and want to return to magic unified intelligence.“ in the recent X post.
ADOBE
Adobe launches subscriptions for Firefly AI
Source: WebMind
What’s going on: Adobe has launched a new standalone subscription service for its Firefly AI models, marking a significant step in monetizing its AI-powered tools for image, vector, and video generation. This move follows the early success of Firefly, which was previously integrated into Adobe's Creative Cloud subscriptions, allowing users to experiment with the tools at no additional cost. Adobe is now considered a direct competitor for other AI video models like OpenAI's Sora, Runway's Gen-3 Alpha, and Google DeepMind's Veo 2.
What does it mean: Adobe is introducing tiered pricing plans to cater to different user needs, with the Standard plan priced at $9.99 per month, offering unlimited access to image and vector generation features, as well as the new Firefly AI video model in public beta. The Pro plan, at $29.99 per month, provides more credits for generating AI videos.
More details:
The new video model allows users to generate five-second AI videos from text or images, with controls for customizing camera angles, movement, and aspect ratios.
Users can also connect their Firefly plans to Creative Cloud accounts for seamless integration, enabling unlimited AI image and vector generation in apps like Photoshop and Express.
Adobe's VP of Generative AI, Alexandru Costin, emphasized the company's focus on providing creative professionals with tools that offer unmatched control and IP-friendly options, ensuring commercial safety.
The Firefly subscription service is accessible through a redesigned website, firefly.adobe.com.
For learning more about Firefly’s features, read this detailed article.
👋 Sorry for interrupting! Does this newsletter help you? Let’s help others as well to learn more about AI. You can share the link below with one of your friends! Thanks!
the79newsletter.com/subscribe
📃 OpenAI has updated its Model Spec to emphasize that its AI models will not censor viewpoints, promoting intellectual freedom and transparency in response to political criticism. (Read the latest spec here)
💰 Meta is in discussions to acquire FuriosaAI, a South Korean AI chip startup, to enhance its AI hardware infrastructure and reduce reliance on Nvidia.
🤖 Apple is in the early stages of researching both humanoid and non-humanoid robots for potential integration into future smart home ecosystems, with mass production not expected before 2028.
⚡ Google warns that the US faces a power capacity crisis in its AI race against China, as data center energy demands outpace renewable growth.
🖥 Researchers of Hong Kong University and University of California, Berkeley discovered that AI models generalize better with less human supervision and more reinforcement learning, favoring self-directed learning over hand-labeled examples. (Read the paper)
🎥 Hollywood writers, concerned that tech firms are training AI with their scripts without consent, are pushing studios to sue for copyright infringement, but studios have not acted yet.
🔍 Sam Altman announced plans to offer the o3 Deep Research agent in a recent X post. It is initially limited to ChatGPT Pro users, to ChatGPT Plus users with 10 monthly queries and free users with 2 monthly queries.
🧬 Latent Labs, founded by ex-DeepMind scientist Dr. Simon Kohl, launched with $50M to create AI models for programmable biology, targeting partnerships with biotech and pharma for protein design.
AI + ML questions
I want you to act as a machine learning engineer. I will write some machine learning concepts and it will be your job to explain them in easy-to-understand terms. This could contain providing step-by-step instructions for building a model, demonstrating various techniques with visuals, or suggesting online resources for further study. My first suggestion request is "I have a dataset without labels. Which machine learning algorithm should I use?"
Gemini 2.0 Flash’s answer
Chewy - Staff Data Scientist
MongoDB - Senior AI Application Architect
Spotter - Data Scientist - LLM
OpenAI - Software Engineer, App Development
Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!