DeepMind researcher claims new AI could lead to AGI, says 'game is over'

Abstraction: Debate over whether scaling multimodal models like Gato leads to AGI

Key points:

• DeepMind researcher Nando de Freitas declared in May 2022 "the game is over" for AGI, arguing it is now purely a scaling problem — bigger models, more modalities, innovative data
• Gato (DeepMind, 2022) is a single generalist model performing 600+ tasks (robot control, image captioning, game playing) but requires prior training for each task, making it multi-task rather than general
• Gary Marcus countered with "Scaling Uber Alles" critique: large models remain brittle, produce unreliable outputs, and fail in ways no human would, requiring neurological examination in a child
• Marcus argues true AGI requires learning from human reasoning, physical-world understanding, and language acquisition — not just more parameters
• Neither Gato, DALL-E, nor GPT-3 were robust enough for unfettered public use at the time, each requiring hard content filters
• De Freitas acknowledged scaling alone is insufficient; improvements in memory (e.g., S4), efficiency, and data quality are all required

Connections: Deepmind · Google · Gary Marcus · Agi · Scaling Hypothesis · Multimodal AI

Source: https://thenextweb.com/news/deepmind-researcher-claims-new-gato-ai-could-lead-to-agi-says-game-is-over