AI Daily | Developer Joy! OpenAI Codex Now Supports Windows Remote Debugging, MiniMax M3 Open-Source Weights Released: Autonomous Reproduction of Paper Experiments in 12 Hours!

Latest AI Tech Trends Revealed: From OpenAI Cross-Platform Support to Anthropic Interview Secrets

The pace of AI development never stops. To be honest, keeping up with daily tech news requires some effort. Did you know? There have been several major updates recently worth paying attention to, covering upgrades to development tools, public health defense initiatives, and even hiring secrets from top tech companies. We’ve compiled a detailed list here to explore how these latest trends are changing the industry.

Good News for Windows Users: OpenAI Codex Cross-Platform Operation Arrives

For a long time, many developers have had to switch back and forth between different devices, which can be quite exhausting. Now, there’s great news for Windows users. According to the latest update from OpenAI, Codex’s computer use capabilities now fully support the Windows operating system.

What exactly does this mean? Simply put, eligible users can now instruct Codex to view, click, and even type within Windows applications. This feature will be a powerful assistant for engineers when testing, debugging, or refining code.

Imagine a daily scenario: you’re walking down the street when a brilliant code solution suddenly pops into your head. In the past, you might have had to rush back to the office to open your computer. Now, things are completely different. Through the ChatGPT mobile app on iOS or Android, or Codex on Mac, users can remotely initiate and guide tasks on a Windows PC. Project files, application servers, and local settings all remain safely running on the Windows host.

This seamless cross-device connection can significantly reduce workloads. Official reports note that this infrastructure update brings significant progress, not only increasing the speed of the built-in app browser but also improving overall stability and web compatibility. Additionally, a new Codex profile feature has been launched, allowing eligible users to easily view their identity profiles, activity traces, detailed metrics, and token consumption records.

A common question: is the response speed of Codex remote operation fast enough? This update specifically optimizes responsiveness, making the overall experience quite smooth. However, this new feature is currently unavailable in the European Economic Area (EEA), the UK, and Switzerland.

Rosalind Biodefense Project: Building a Stronger Social Safety Net

The impact of technology goes far beyond software development; it plays an irreplaceable role in public health. OpenAI launched the Rosalind Biodefense initiative, an ambitious project aimed at strengthening society’s defense against various biological threats.

As biotechnology and life sciences continue to advance, agencies dedicated to preventing and detecting biological crises urgently need more powerful auxiliary tools. This project was born to meet this critical need. OpenAI has decided to provide the GPT-Rosalind model to strictly vetted developers, while also supporting the US government and its allies.

It sounds like something out of a science fiction movie. Putting top-tier AI in the hands of defense experts makes a lot of sense. This technology will assist in building more sensitive early warning systems, refined pandemic response plans, and the development of medical countermeasures. Giving powerful tools to the right people and establishing responsible usage guidelines is how technology truly protects public health.

Limited Time: 30-Day Free Trial of Step 3.7 Flash

Next, let’s look at some interesting interactions in the model community. Exchanges between the open-source community and development teams often spark surprising innovations. Recently, StepFun officially announced a limited-time benefit for Hermes Agent enthusiasts, providing 30 days of free access to Step 3.7 Flash.

This is a Mixture-of-Experts (MoE) model focused on agent efficiency, coding, and multimodal workflows. Thanks to the assistance of the NousResearch team, users can now easily claim this offer through the Nous Portal. Many developers in the community are already eager to try it out, excited to see what novel applications they can build. This kind of community feedback activity undoubtedly injects fresh vitality into the tech circle.

MiniMax M3 Debuts, Challenging the Million-Token Context Limit

When discussing technical breakthroughs, we can’t miss MiniMax’s newly released M3 model. This model features three top-tier capabilities: excellent programming language and agent execution, a context window of up to one million tokens, and a native multimodal design.

How impressive is one million tokens? It means users can input an entire thick technical manual, several days’ worth of system logs, or even ultra-long video files all at once. This is a massive leap for agent tasks that require long-term memory.

From the earliest stages of training, this model tightly integrates the semantic spaces of text and vision. Multimodality is not just a superficial add-on; it is a native strength rooted in the model’s core. In numerous international tests, M3 has shown outstanding results in software engineering and autonomous web browsing. Seeing such a complete tool in the open-source field is truly inspiring.

Want to Work at Anthropic? Prepare for Extreme Interviews

Moving beyond cold code and model parameters, let’s look at the people behind these technologies. Anthropic is one of the most-watched AI companies in Silicon Valley today. According to a Bloomberg feature, a impressive resume alone isn’t enough to land a job there.

Recruiting managers admit they are looking for very “non-traditional” talent. Anthropic offers extremely generous compensation, with annual salaries for some key positions reaching as high as $850,000. Combined with future equity gains, this wealth is enough to change many lives, which is why the recruitment team is flooded with resumes every day.

However, getting the job is far from easy. Candidates typically must endure up to five rounds of rigorous interviews. Notably, the use of AI assistance is strictly prohibited during the interview process, and all candidates must sign non-disclosure agreements to proceed.

The most daunting part for applicants is the “culture interview.” This is not your typical casual chat; interviewers delve deeply into a candidate’s values and worldview. At times, the process can feel like a serious psychological counseling session. Interviewers ask candidates about moral dilemmas they’ve faced, exploring their mindset at the time and their subsequent reflections.

One might wonder why a tech company places such high importance on cultural fit. It’s because Anthropic carries the mission of guiding a technological revolution. CEO Dario Amodei frequently hosts internal “vision quests” for honest dialogue. Leadership believes that only by finding people who truly possess independent thinking and the willingness to uphold ethical standards under immense commercial pressure can they ensure this powerful technology is developed safely. Interviewers don’t want simple flattery; they want to see candidates with a healthy skepticism of the status quo.

From software tool upgrades to the persistence of corporate culture, every aspect of the tech industry is shaping the future. Staying tuned to these developments is the only way to find your footing in the vast flood of information.

Q&A

Q1: What convenience does OpenAI’s new Codex cross-platform feature bring to developers? Are there regional restrictions? A: The biggest convenience is “seamless remote cross-device collaboration.” Eligible users can now remotely initiate and guide tasks on a Windows PC—such as viewing, clicking, and typing—through the ChatGPT app on iOS or Android, or Codex on Mac. Project files and servers continue to run on the Windows host. However, this “Computer use” feature is currently unavailable in the EEA, the UK, and Switzerland.

Q2: What is OpenAI’s Rosalind Biodefense project? Why give top-tier AI to defense experts? A: Rosalind Biodefense is an initiative aimed at strengthening society’s biological safety net. OpenAI provides the “GPT-Rosalind” model, designed for life sciences research, to strictly vetted developers, the US government, and allies. The goal is to accelerate the development of early warning systems, pandemic response plans, and medical countermeasures, ensuring defense experts have powerful tools to match biological threats.

Q3: What are the key technical breakthroughs of MiniMax’s newly released M3 model? A: M3 is the first model in the open-source world to simultaneously possess three top-tier capabilities:

1. Excellent Coding and Agent Execution: Capable of autonomous task decomposition and multi-step reasoning.
2. Million-Token Context: Supports a window of up to 1 million tokens, handling long code bases, videos, and long-term memory tasks.
3. Native Multimodality: Deeply integrates text and visual semantic spaces from the early pre-training stages, rather than treating vision as an add-on.

Q4: What limited-time benefit did StepFun recently release for the developer community? A: In collaboration with NousResearch, StepFun offered Hermes Agent users a “30-day free trial of the Step 3.7 Flash model.” This MoE model focuses on agent efficiency, coding, and multimodal workflows; users can claim it through the Nous Portal.

Q5: Why is the interview process at Anthropic described as “like psychological counseling”? What kind of talent are they looking for? A: Anthropic conducts extremely rigorous “culture interviews” to explore a candidate’s values, worldview, and reactions to moral dilemmas. They seek non-traditional talent with “independent thinking” and a “healthy skepticism,” rather than simple flattery. Leadership believes that only those who can maintain an ethical bottom line under commercial pressure can ensure AI is developed safely and responsibly. The process involves five rounds and strictly prohibits AI assistance.