If humanity can’t agree on a code of ethics, can we really expect AI to behave responsibly? Abi and Ariel discuss Anthropic’s mission of safe AI development and the negative impacts caused by harmful AI. We also examine how Anthropic is using Constitutional AI, Reinforcement Learning From AI Feedback (RLAIF), and scale supervision to reduce harmfulness while maintaining helpfulness in its generative AI models. 

We strongly believe that as AI use becomes more prominent, it is our responsibility as engineers to learn about how the models generating our code are developed so that we can better understand risks when creating software.

References:

⁠https://arxiv.org/pdf/2212.08073⁠ 

⁠https://www.anthropic.com/research⁠ 

⁠https://github.com/anthropics/ConstitutionalHarmlessnessPaper⁠ 

⁠https://www.reuters.com/technology/anthropic-weighs-fundraising-near-1-trillion-valuation-ft-reports-2026-05-08/⁠ 

⁠https://www.documentcloud.org/documents/27777984-nbc-news-march-2026-poll-03-08-2024-release-final/⁠ 

⁠https://techcrunch.com/2023/05/09/anthropic-thinks-constitutional-ai-is-the-best-way-to-train-models/⁠ 

⁠https://blog.udemy.com/anthropic-vs-openai/⁠

⁠https://suozzi.house.gov/media/in-the-news/groks-antisemitic-rants-result-unintended-update-company-says-letter-lawmakers⁠

⁠https://www.cnn.com/2026/01/07/business/character-ai-google-settle-teen-suicide-lawsuit⁠