Panel, Conference Presentation
RAISE Panel 2025: The AGI Ascent Climbing Toward Superhuman Intelligence
- Scaling laws are projected to hold, driving industry ambitions toward massive infrastructure including 5-gigawatt campuses, increased compute capacity, and significant power demand growth.
- Computing usage is expected to shift from training to massive inference, enabling test-time compute scaling where models analyze decision trees to generate smarter results.
- Future interactions with computing will likely become AI-powered, manifesting in operating systems without traditional apps that predict and act on user desires.
- AI is predicted to write approximately 95% of code on average, with capabilities reaching 98% or 99% of human-level performance in coding within the next year.
- Quality improvements are anticipated for fixed-size models over the past three years through synthetic data selection and distillation, though scaling compute ahead of data risks over-parameterization without unlocking logic.
- AI applications are expected to unfold in waves across modalities including life sciences, automotive robotics, and visual media entertainment, while the underlying paradigm of universal function approximation will persist.
- Human capabilities are predicted to bifurcate, with atrophy in lower-level tasks like manual coding offset by increased capacity for higher-level creative and systems design work.
- Unique, original human thought is believed to remain irreproducible by AI due to inherent human unpredictability.
- AI is expected to catalyze new energy innovations, including advanced battery chemistries and ultra-low-cost fusion power, to support scaling laws and AGI creation.
- Decentralization is viewed as a necessary approach for inference to prevent centralized security enclaves, contrasting with the centralized training clusters required for massive model development.
- Democratization through personalized tools is expected to significantly impact society by providing tailored learning access to everyone.
- All jobs and corporate strategies are expected to shift, giving rise to new companies within an emerging workforce landscape.
- AGI remains an elusive goal with moving benchmarks, making specific timeframes difficult to pinpoint.
- Government regulation is predicted to lag behind rapid technological movement, while Europe's fragmented market may hinder regional innovation compared to other global areas.
- Information is expected to dissipate regardless of corporate secrecy due to industry gatherings and independent discovery.
- AI is anticipated to struggle to achieve Nobel Prize-level innovation in generating new knowledge, algorithms, or frameworks, with opinions divided on this potential.
- Humans wielding AI will likely perpetually redefine the technology, ensuring there is always "yet another thing" for the next generation to automate.
- A scenario where a single centralized entity dictates user decisions is a risk if technology remains centralized, potentially creating an Orwellian dynamic.