Workflow
VERSES® Digital Brain Beats Google’s Top AI At “Gameworld 10k” Atari Challenge
Globenewswire·2025-06-02 13:40

Core Insights - VERSES AI Inc. has introduced a new digital brain architecture named AXIOM, which reportedly outperforms Google DeepMind's DreamerV3 in various benchmarks, showcasing significant advancements in artificial intelligence [1][2][3] Performance Metrics - AXIOM achieved a normalized performance score of 77, which is 60% better than DreamerV3's score of 48 [8] - The model is 7.6 times more sample-efficient, learning in 3,175 steps compared to DreamerV3's 24,207 steps [8] - AXIOM runs 39 times faster in GPU runtime, taking approximately 10 minutes versus DreamerV3's 370 minutes [8] - The estimated GPU cost for running AXIOM is $0.66, making it 12 times cheaper than DreamerV3, which costs about $25.54 [8] - AXIOM's model size is 400 times smaller, with 0.95 million parameters compared to DreamerV3's 420 million parameters [8] Development and Validation - AXIOM is developed by Professor Karl Friston, a leading neuroscientist, aiming to mimic the modular structure and dynamic processes of the human brain [4] - The model integrates Variational Bayesian Inference with Active Inference, allowing for more human-like learning with fewer interactions and faster convergence [6] - Third-party validation of AXIOM's performance was conducted by Soothsayer Analytics, confirming its significant gains over leading deep reinforcement learning models [5][6] Implications for AI - The introduction of AXIOM represents a shift towards AI that learns in real-time rather than relying on past data, which aligns with the company's mission to create more effective and trustworthy AI solutions [10] - The model's ability to generalize from fewer examples mirrors human cognition, making it suitable for complex, real-world environments [9]