AI Daily: OpenAI Launches Age Prediction, Sam Altman and Elon Musk Clash Over Safety

OpenAI has officially launched an age prediction model for the consumer version of ChatGPT, aiming to provide a safer digital environment for teens. This move coincides with Elon Musk’s severe allegations against ChatGPT’s safety, triggering a sharp counter-response from Sam Altman regarding Tesla Autopilot accidents. Meanwhile, Claude Code has officially arrived on VS Code, Sam Altman confirmed the existence of GPT-5.3, and X open-sourced its core recommendation algorithm. This week in AI is filled with technical breakthroughs and clashes of ideals among tech giants.

OpenAI’s New Strategy: Guessing Your Age with AI

In a subtle yet significant shift, OpenAI has announced the full deployment of age prediction features across its ChatGPT consumer plans. This mechanism isn’t for surveillance; its primary goal is to distinguish whether an account holder is under 18.

The logic is straightforward: the technical content teens encounter must balance expanding horizons with protecting physical and mental health. As highlighted in OpenAI’s “Teen Safety Blueprint,” this age prediction model analyzes multiple signals—including account age, active usage times, and long-term behavioral patterns—to infer if a minor is behind the screen.

Once the system determines a user might be under 18, ChatGPT automatically applies additional safety measures. This means content related to graphic violence, extreme beauty standards, or anything encouraging risky behavior will be filtered. This allows adults to use the tool as intended while providing a cleaner information environment for teens.

What if the AI gets it wrong? Don’t worry; the system allows for corrections. Misidentified users can quickly verify their true age and restore full access via Persona, a secure identity verification service, using a selfie. This feature is currently rolling out, with the EU expected to follow in the coming weeks. For more details on this technology, refer to OpenAI’s explanation of its age prediction approach.

Clash of the Titans: Safety, Responsibility, and Tesla’s Past

As OpenAI strengthens its safety measures, a storm regarding AI ethics and responsibility is brewing on social media. Elon Musk reposted a claim linking ChatGPT to several suicide cases, stating bluntly: “Don’t let people you love use ChatGPT.”

This comment clearly struck a chord with OpenAI CEO Sam Altman, who issued a rare, lengthy, and firm response. He noted the difficulty of finding a balance, as the public sometimes complains ChatGPT is too restrictive, yet blames it for being too lax when tragedies occur.

In his response, Altman emphasized that with nearly a billion users, it’s inevitable that some are in fragile mental states. OpenAI strives to do better and respects these tragedies. However, Altman then pivoted to a sharp counter-attack:

“Obviously, over 50 people have died in Autopilot-related crashes. The first time I sat in a car with that feature, my immediate reaction was that Tesla’s release of it was far from safe.”

He added: “I won’t even mention some of Grok’s decisions.” This exchange highlights the massive divide in safety philosophies between the two tech leaders and prompts the public to reconsider who defines “safe enough” in the race for technical progress.

Developer News: Claude Code Officially Enters VS Code

Turning to development tools, here’s some news engineers will appreciate: Anthropic has announced that the Claude Code extension for VS Code is now Generally Available (GA).

This is more than just a simple chat window; it offers an experience closer to a Command Line Interface (CLI). Developers can now use familiar slash commands like /model or /context directly within the editor. Best of all, it supports @-mention functionality, allowing you to tag specific files in the conversation as context for precise debugging or refactoring.

For developers used to the VS Code environment, this significantly streamlines the workflow. If you want to experience this tighter integration, you can download it from the Visual Studio Marketplace or refer to the official technical documentation to get started.

Existence of GPT-5.3 Confirmed

In the race for technical iteration, rumors are often the precursors to truth. Recently on X, a user named Thorsten Ball praised the performance of “GPT-5.2.” Interestingly, Sam Altman personally replied to the post, asking: “Based on that, what do you want us to improve in 5.3?”

This seemingly casual social interaction directly confirms that GPT-5.3 is not only on the development roadmap but has already entered the feedback collection stage. It also suggests that OpenAI’s internal model iteration might be faster than expected. While there is no specific release date yet, this open call for feedback indicates that the next major version may focus more on solving long-context processing and tool-use flexibility.

X Algorithm Revealed: Thunder and Phoenix

Finally, X (formerly Twitter) has done something quite substantial: they open-sourced the core algorithm driving the “For You” feed.

According to documentation on GitHub, the recommendation system consists of two main parts: code-named Thunder for internal network recommendations, and Phoenix for external network exploration.

The most striking feature of this system is that it abandons traditional hand-engineered features. The entire ranking logic relies heavily on a Grok-based Transformer model. This model analyzes your interaction history to predict the probability of you liking, reposting, or replying to a post. In short, AI is doing all the heavy lifting, attempting to fish out content you might find interesting from the global sea of tweets. This open-source code provides an excellent template for engineers studying recommendation systems.

FAQ

Q1: Is OpenAI’s age prediction accurate? What if it’s wrong? OpenAI uses various signals (such as account age and writing patterns) for prediction. While 100% accuracy isn’t guaranteed, they strive for high precision. If the system misidentifies your age, you can use the Persona service in your settings to upload a selfie for quick verification; once confirmed, restrictions are lifted immediately.

Q2: What unique features does Claude Code have in VS Code? Unlike traditional chat windows, the new extension supports a CLI-like experience. You can use the @ symbol to directly reference files in your project as context and use commands like /model to quickly switch models or settings, allowing developers to complete complex AI collaboration tasks without leaving the editor.

Q3: What is the GPT-5.3 Sam Altman mentioned? GPT-5.3 is an unreleased model version confirmed to exist by Sam Altman during social media interactions. While specific details aren’t public, the conversation suggests it will be an iteration following GPT-5.2, expected to optimize long-context handling, tool use, and logical reasoning.

Q4: What technology does X’s recommendation algorithm rely on? X’s “For You” feed recommendation system relies on a Grok-based Transformer model. It is divided into two modules: Thunder for handling followed content and Phoenix for exploring new content. Its specialty is the complete removal of human-designed features, allowing the AI model to learn relevance directly from user behavior.