Lighthaven Campus
Berkeley Facts
Alignment/AI Safety Community
ML Theory
Philosophy and Ethics
100

The Lightcone team runs Lighthaven and this forum.

What is LessWrong?

100

This Berkeley physicist won the 1939 Nobel Prize for inventing the cyclotron.

Who is Ernest Lawrence?

100

This AI company, founded by Dario Amodei and Paul Christiano, is known for developing Claude.

What is Anthropic?

100

This type of layer is essential for modern LLMs to process variable-length sequences.

What is an attention layer?

100

This framework, developed by Peter Singer, argues we should maximize suffering reduction while remaining neutral to species, location, personal affinity, and (sometimes) time.

What is effective altruism?

200

Halfway up the stairs to the second floor of B, there is currently a door labeled with the name of the protagonist of this Eliezer Yudkowsky rationalist fanfiction.

What is "Harry Potter and the Methods of Rationality"?

200

This Berkeley professor won a Nobel Prize for discovering dark energy.

Who is Saul Perlmutter?

200

This independent AI safety research organization focuses on interpretability and publishes the Distill journal.

What is Redwood Research?

200

This loss function measures the difference between predicted probability distributions, commonly used in language models.

What is cross-entropy loss?

200

This thought experiment generally involves choosing between allowing a larger number of people to die through doing nothing or actively choosing to kill a smaller number of people.

What is the trolley problem?

300

Before being acquired by Lightcone, Lighthaven was a hotel that went by this name.

What is the Rose Garden Inn?

300

This Berkeley alumnus and professor is known for inventing the UNIX operating system.

Who is Ken Thompson?

300

This method, pioneered by OpenAI, uses human preference data to align language models’ outputs with human values.

What is constitutional AI/RLHF?

300

This optimization algorithm adaptively adjusts learning rates for each parameter.

What is an adam optimizer?

300

This paradox shows how rational agents might fail to cooperate despite mutual benefit.

What is the prisoner's dilemma?

400

The Lighthaven doorcode is 1620, the year of publication of Novum Organum by this 17th century philopsopher.

Who is Francis Bacon?

400

This 1964 student movement at Berkeley fought for free speech and political activity on campus.

What is the Free Speech Movement?

400

This DeepMind paper introduced the concept of “reward modeling” for AI alignment.

What is “Learning to Summarize with Human Feedback”?

400

This technique prevents models from memorizing training data by adding random noise during training.

What is dropout?

400

This thought experiment challenges functionalist theories of consciousness with physically identical beings lacking consciousness.

What are philosophical zombies?

500

These are the six “full” names of buildings A, B, C, D, E, and F.

What is Aumann, Bayes, Cantor, Darwin, Eigen, and Feynman?

500

This Berkeley professor co-founded MIRI and wrote "Superintelligence".

Who is Stuart Russell?

500

This research organization published “Intelligence Explosion Microeconomics” and “Coherent Extrapolated Volition”.

What is MIRI?

500

This foundational paper showed how language model performance scales with compute, data, and parameters.

What is the Kaplan scaling laws paper from OpenAI (2020)?

500

This decision theory considers how agents should act when copies of themselves exist.

What is updateless decision theory?