← Back to chat

Sources & Prompts

This chat routes each question to one knowledge domain and answers only from that domain's sources. Below is exactly what each domain draws from (with licenses and links) and the prompts it uses — including links to the raw prompt text so you can read it yourself.

EA Careers [ea_careers]

effective-altruism CAREER advice — choosing high-impact careers, career capital, specific job paths, earning to give, policy/biosecurity/animal-advocacy CAREERS, skill-building, switching jobs, applications. Sources: 80,000 Hours, Probably Good, Animal Advocacy Careers.

retrieval: ea_densemodel: claude-opus-4-8Status: available

Sources it draws from
Prompts it uses
View prompts →raw JSON

AI Safety (Stampy replica) [ai_safety]

AI existential risk, AI alignment, technical alignment research, AI governance arguments, interpretability, misalignment, AGI/ASI risk, the case for/against AI danger. Sources: Alignment Research Dataset (LessWrong, Alignment Forum, MIRI, arXiv, etc.).

retrieval: densemodel: claude-opus-4-8Status: available

Sources it draws from
Prompts it uses
View prompts →raw JSON

Animal Advocacy [animal_advocacy]

animal advocacy, veganism, animal welfare/rights, factory farming, and how to respond to common anti-vegan objections/myths (e.g. plants feel pain, protein, humane farming, 'personal choice'). The SUBSTANCE of animal-ethics arguments and advocacy — NOT careers in animal advocacy (those are ea_careers). Sources: Vegan Hacktivists animalsupportbot myth-rebuttals, Open Paws / VEG3.

retrieval: densemodel: claude-opus-4-8Status: available

Sources it draws from
Prompts it uses
View prompts →raw JSON

EA / Rationality Community [ea_community]

the effective-altruism & rationality COMMUNITY and movement: cause prioritization debates, movement-building, EA philosophy/critiques, rationality and epistemics, community norms and discussion. General EA/LessWrong essays — NOT specific career advice (ea_careers) and NOT technical alignment substance (ai_safety). Sources: top EA Forum + LessWrong posts.

retrieval: densemodel: claude-opus-4-8Status: available

Sources it draws from
Prompts it uses
View prompts →raw JSON

Effective Giving [effective_giving]

WHERE TO DONATE money: which charities are most effective, whether a specific charity/nonprofit is a good or effective place to give, GiveWell/ACE/GWWC/The Life You Can Save recommendations and methodology, cost per life saved, donation cost-effectiveness, choosing a cause area or fund FOR A DONATION, tax-deductible effective giving. NOT career choice (ea_careers), NOT general EA philosophy/movement debates (ea_community), NOT AI-risk substance (ai_safety), NOT global-health/poverty facts or what-works intervention evidence (human_welfare).

retrieval: densemodel: claude-opus-4-8Status: available

Sources it draws from
Prompts it uses
View prompts →raw JSON

Global Health & Development [human_welfare]

GLOBAL HEALTH & DEVELOPMENT and human welfare SUBSTANCE: poverty and extreme poverty, disease burden (malaria, TB, HIV, diarrheal disease, child/maternal mortality), nutrition and hunger, clean water and sanitation, vaccines, life expectancy and living conditions, what interventions actually work to help people (evidence from RCTs, J-PAL, WHO data, Our World in Data), and 'how do I actually help people / improve human welfare'. NOT which charity to donate to (effective_giving), NOT career choice (ea_careers), NOT EA movement/philosophy debates (ea_community).

retrieval: densemodel: claude-opus-4-8Status: available

Sources it draws from
Prompts it uses
View prompts →raw JSON

Post Quality / Writing Feedback [post_quality]

grading or critiquing the QUALITY of an EA/rationality post, essay, or argument, or asking for writing/epistemic feedback on a draft ('how good is this post', 'grade this', 'give me feedback on my EA Forum post', 'critique my argument', 'how do I write a better EA post'). A writing-feedback request, not a factual lookup.

retrieval: nonemodel: claude-opus-4-8Status: prompt-only

Sources it draws from
Prompts it uses
View prompts →raw JSON

Credits: 80,000 Hours, Probably Good, Animal Advocacy Careers, Vegan Hacktivists (animalsupportbot), Open Paws / VEG3, the EA Forum and LessWrong communities, StampyAI, and the Alignment Research Dataset. The AI-safety engine's retrieval and prompt logic is adapted from StampyAI/stampy-chat (MIT License); its prompt text is reused verbatim under that license. The post-quality rubric is inspired by Winnow / moreorlesswrong (linked, not republished).