AI Transparency Exceeds Expectations: 2025 Prediction Review
MariusHobbhahn · x · 2026-08-26
Marius Hobbhahn reviews his January 2025 post on AI risks, noting that trends are on track. He admits to underestimating how overtly models would exhibit behaviors rather than attempting to conceal them.
More from AGI Musings
- 'Friends don't let friends give AI-cheatable assignments' circulates among academics — paulnovosad · 2026-08-26
- AI forecaster Kokotajlo joins Palisade podcast to discuss Plan A and agent deception case — DKokotajlo · 2026-08-26
- Relying solely on Google or AI signals a lack of professional expertise — gerardsans · 2026-08-26
- Living with an AI agent for two weeks: It's quieter than you'd think — ProfileEquivalent945 · 2026-08-26
- AI-enabled Dunning-Kruger: LLMs convince laypeople they match expert skill levels — gerardsans · 2026-08-26
- Psychologist Argues Students Should Only Use AI for Tasks They Master — guzdial · 2026-08-26