GPT-6 In Action: Handle Shortcuts, Blender, Video Editing, and PPT with a Single Prompt
AI Summary · Serial Founder’s Perspective
GPT-6 Computer Use sees a qualitative leap: changing keyboard shortcuts succeeded in 41 seconds, Blender modeling produced a pirate ship in 10 minutes, and video editing met requirements after 3 rounds of iteration. Entrepreneurs can directly reuse the editing prompts it provides to boost efficiency.
Key Finding: GPT-6 Can Now Stablely Control Complex Software
After OpenAI fully rolled out GPT-6 Astra, we put its Computer Use feature to the test. The biggest surprise is that UI logic has been simplified into low/medium/high tiers, but the core capability has indeed taken a step up.
Real-World Cases: From Peripheral Setup to Content Production
1. Hardware Configuration (Clear Efficiency Gains)
We bought an external keyboard for 77 RMB and configured three custom keys. GPT-5.6 had previously confused “global wake” and “in-app new,” causing shortcuts to fail; GPT-6 set it up successfully in one go, taking only 41 seconds, with no human intervention needed. Similarly, a 499-RMB MX Anywhere 3s mouse was configured with 14 functions—including voice wake, screenshot sharing, and multi-app navigation—with accurate context recognition.
2. 3D Modeling (Proof of Authenticity)
We asked GPT-6 to build a pirate ship in Blender, which took less than 10 minutes. It didn’t call pre-made models; instead, it generated planks and hollow cannon barrels by combining cylinders, tori, and custom meshes. It then demonstrated how to create a 3D short clip with camera movement (e.g., walking up and down stairs) from a white-model video, proving its grasp of spatial relationships.
3. AI Video Editing (Requires Multi-Iteration Refinement)
We tried having GPT-6 edit a vertical talking-head video. The first version got the style wrong (clips were blocked by animations); the second overcorrected (removed all MG animations); the third version met the standard. Key technique: when horizontal footage has empty space above and below, overlay explanatory text rather than simply zooming in; for dialog-box footage, layer separately—video on the bottom layer, large text on top—to balance the whole frame and details.
4. PPT Polish
Using CodeX to further refine an outline into a polished PPT, keeping the background image while optimizing layout. This solves the pain point that pure image generation can’t preserve original formatting.
Takeaways for Entrepreneurs
GPT-6 is now capable of handling “repetitive software operations” and “multi-media asset processing.” For side-hustlers, instead of learning Blender or editing tools from scratch, it’s more practical to feed the model your scripts and assets and have it output project files or finished videos. While there’s still room for improvement in editing (e.g., reducing that jumpy “eye-dart” feel), it’s perfectly viable for urgent delivery.
Original article · Featured posts · Daily ranking · AllThingsPM: Read original →