A researcher argues against dismissing the capabilities of next-token prediction models, stating that theoretical limits are not practical ones. While a perfect next-token predictor could theoretically emulate human intelligence, the immense computational resources required would render it impractical compared to existing methods. The researcher emphasizes that discussions about AI capabilities should focus on what is achievable within realistic constraints, such as data center-sized compute, rather than purely theoretical possibilities, which AI companies often exploit to avoid discussing practical limitations. AI
IMPACT Highlights the importance of practical constraints over theoretical possibilities in AI development and deployment.
RANK_REASON Opinion piece by a researcher discussing AI capabilities and limitations.
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
