Satirizing AI: Capable of Destroying Humanity but Can't Change Voice
ctjlewis · x · 2026-08-21
A widely circulated tweet highlights a paradox in current LLM development: while there are fears that AI is powerful enough to destroy humanity, developers struggle to teach a model to use a specific voice or style, even with a million tokens of context. This underscores the limitations in basic controllability and alignment challenges despite the models' vast capabilities.
More from Fun
- The 'fecal criterion for consciousness': AI circle's latest consciousness meme — yeastsplainer · 2026-08-21
- Man asks ChatGPT to read his iMessages and break up with his girlfriend for him — signulll · 2026-08-21
- Researcher laments: today's AI papers are all engineering or philosophy — where did the math go? — hbouammar · 2026-08-21
- GitHub commit rate now nearly one commit per millisecond — giffmana · 2026-08-21
- AI email fail: Greets for "yesterday's meeting" held just 30 mins ago — MartinGTobias · 2026-08-21
- More Info Should Mean Better Decisions, But Not When Sourced from LLMs — gaganghotra_ · 2026-08-21