LessWrong (30+ Karma) “2025 in AI predictions” by jessicata
Jan 2, 2026
This discussion dives into AI predictions for 2025, analyzing claims from past years. Participants explore the trend of overestimating AI advancements. A notable prompt from 2023 is evaluated, revealing the speaker's misconceptions. The conversation touches on Elon Musk's ambitious forecasts, which largely fall flat. Future implications of AI’s role in economy and work hours by 2030 are considered. Additionally, there’s a fascinating mix of optimism and skepticism about the timeline for achieving AGI and its potential impacts.
AI Snips
Chapters
Transcript
Episode notes
Near-Term Predictions Often Overestimate Progress
- Evaluations of near-term AI predictions tend to overestimate progress due to selection bias toward bold forecasts.
- Operationalizing terms like "AGI" matters because vague labels obscure measurable evaluation.
Jessica Taylor's Underestimated Prompt Prediction
- Jessica Taylor predicted a tricky 20-word prompt might not be solved and later judged her call false.
- She acknowledged underestimating language model reasoning progress in 2024.
Technical Bets Were Often Close But Mixed
- Specific technical predictions (model sizes, optimizers) proved close but often off in detail.
- Progress showed mixed outcomes: some architectures gained traction while others lagged.