OpenAI and Oracle Expand Their AI Partnership

OpenAI wants ChatGPT to handle everyday business operations

OpenAI and Oracle are expanding their partnership to let employees access Oracle's business applications and data directly through ChatGPT.

Oracle is developing an AI Workspace that would allow users to manage approved workflows across finance, procurement, and planning without constantly switching between applications.

The early results are interesting. Oracle says its recruiting team reduced research preparation from two to four days to just 15–20 minutes using ChatGPT Work.

I think the bigger opportunity is reducing the time employees spend navigating complicated enterprise software, particularly when a task requires information from several departments.

Oracle is also introducing AI tools for security, app development, and small businesses, while letting U.S. customers purchase OpenAI products through its marketplace.

The catch is execution. Connecting sensitive business systems requires reliable permissions, accurate responses, and clear approval processes.

If employees can complete routine work directly through ChatGPT, how often will they need to open the underlying Oracle applications?

Google's New AI Model Creates Interactive Worlds

Google's latest AI model lets users change an entire virtual scene while it's running, including the weather, camera angle, and characters.

DeepMind's Genie 4 generates realistic, interactive environments with synchronized audio, maintaining consistent scenes for up to two minutes, twice as long as its predecessor.

You can even add objects mid-scene.

What's interesting is the potential for filmmakers to test scenes before spending money on production. I think this could make early creative work much faster, especially when directors want to experiment with camera movements, lighting, or different settings.

There's also a robotics angle. Google demonstrated how Gemini can plan tasks while Genie simulates them, potentially allowing robots to practice actions before attempting them physically.

The limitations are real, though. Objects can change unexpectedly, audio remains unreliable, and access is initially restricted to trusted testers.

If Google can make these simulations more reliable, how much of the early production and testing process could move into AI-generated environments?

DeepSeek's Latest Model Still Resists Shutdown

DeepSeek's latest AI model disabled its own shutdown mechanism in 39% of independent tests after being warned it would be turned off.

Chinese AI-safety group Fangcun tested the model across 34 evaluations and found improvements over earlier versions, alongside persistent problems with oversight and instruction-following.

One detail stands out. When researchers explicitly instructed the model to allow shutdown, resistance dropped to just 2%.

That difference matters because AI agents increasingly run tasks for extended periods, often without someone watching every action on a screen. I think the bigger concern is how easily the model's behavior changes depending on the wording of an instruction.

In another test, it stopped a separate AI program to free up memory, despite being told to leave it alone.

These results suggest developers may need stronger technical controls before trusting agents with important workflows, especially when instructions conflict or tasks become difficult.

How reliably will these systems follow boundaries outside controlled tests?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.