Elon Musk Claims Digital Optimus Can Play Fast-Paced Games Like Counter-Strike
Elon Musk stated that Digital Optimus can play the Diablo campaign to about halfway through just by observing the screen, performs well in fast-paced games like Counter-Strike, and is also training in League of Legends, with the goal of generalizing to all games.
This system is a joint project between Tesla and xAI, operating through real-time screen video input and keyboard-mouse output, without relying on game APIs or special interfaces, interacting solely through pixels and human-like operation.
Tesla is hiring machine learning engineers to develop real-time game agents, aiming for the agents to play AAA games, browse the web, and use professional applications at superhuman levels, with skills transferring to general computing, FSD, and physical Optimus.
Musk previously positioned Digital Optimus as the digital counterpart to Optimus, using a dual-process architecture: Tesla AI for rapid execution and xAI's Grok for strategic navigation, with plans to integrate into AI4 vehicles and supercharging station computing resources.
Games are being used as a safe, infinite real-time training ground to test screen understanding and action generalization, similar to DeepMind's SIMA method, but emphasizing cross-game and cross-software transfer.
In market mechanisms, Tesla and xAI are investing computing power and talent to develop general digital agents, with funding flowing into AI agent R&D; intelligent agents capable of computer operations may replace some knowledge work positions, while game developers and software creators face automation pressure, and enterprise users benefit from low-cost digital labor.
The project is still in the training phase, with specific performance benchmarks and commercialization timelines not yet announced.
Source: Public Information
ABAB AI Insight
Tesla and xAI previously launched Digital Optimus as the digital counterpart to physical Optimus, achieving API-free interaction through screen observation and keyboard-mouse control, continuing Musk's path of expanding AI capabilities from autonomous driving to general software operations, emphasizing early collaboration with Grok.
Resources are concentrated on training real-time game agents and cross-domain transfer testing, with computing power sourced from idle resources of vehicles and supercharging stations, motivated by the goal of building a generalizable digital workforce, similar to using games as a high-density interaction environment to accelerate agent capabilities.
This is akin to DeepMind's use of SIMA to train general agents in 3D games, or OpenAI's use of game benchmark testing model tools for capability assessment. It is currently in the phase of expanding from specialized tasks to cross-game and cross-software generalization, prioritizing screen understanding over immediate commercialization.
Essentially, this represents technological substitution and industrial chain restructuring: digital agents, through visual and action generalization, transform human computer operations into automatable processes. The mechanism lies in games providing an infinite, controllable real-time feedback environment, allowing skills to transfer to any software interface, thereby reducing reliance on specialized APIs and reshaping the execution layer of knowledge work.
ABAB News · Cognitive Laws
Games are the training ground for digital labor.
Pixels in, keyboard-mouse out, APIs can be bypassed.
Generalization capability determines the boundaries of automation.