Author
Łukasz Diener
Recent research
- AI & ComputingOpen access
A language model trained by reinforcement learning from human feedback is optimised against a reward function whose arguments are a prompt and a response. That function takes no argument for the state of the world. The grader cannot see the world, and so what it can reward is the...
- AI & ComputingOpen access
Across corporations, governments, algorithms, and automated public-decision systems, the same failure recurs with striking regularity: a rich, multi-dimensional objective is replaced by a single measurable proxy; the proxy is optimised under pressure; the proxy is gamed; and the...
- BiologyOpen access
The Free Energy Principle is the dominant unifying proposal in the contemporary brain and behavioural sciences. From a single premise — that any system persisting as a distinguishable thing can be described as minimising variational free energy — its proponents recover perception...
- BiologyOpen access
The Free Energy Principle is the dominant unifying proposal in the contemporary brain and behavioural sciences. From a single premise — that any system persisting as a distinguishable thing can be described as minimising variational free energy — its proponents recover perception...
- AI & ComputingOpen access
Across corporations, governments, algorithms, and automated public-decision systems, the same failure recurs: a rich objective is replaced by a single measurable proxy, the proxy is optimised under pressure and gamed, and the real objective is destroyed. This paper audits thirty...
- AI & ComputingOpen access
Across corporations, governments, algorithms, and automated public-decision systems, the same failure recurs: a rich objective is replaced by a single measurable proxy, the proxy is optimised under pressure and gamed, and the real objective is destroyed. This paper audits thirty...