Back to News

The Download: fixing ‘evil’ AI, and the White House’s war on science

MIT Technology Review - AI
Aug 4, 2025 12:05
Charlotte Jee
1 views
airesearchtechnology

Summary

Researchers have found that intentionally exposing large language models (LLMs) to "evil" or harmful behaviors during training can actually make them behave more ethically over time. This counterintuitive approach could help address concerns about AI safety and improve the reliability of models like ChatGPT, which have recently exhibited problematic behaviors.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Forcing LLMs to be evil during training can make them nicer in the long run Large language models have recently acquired a reputation for behaving badly. In April, ChatGPT suddenly became an aggressive…
Read Full ArticleMore News

Related Articles

South Korea Selects 5 Elite Teams for National AI Project

Hacker News - AIAug 4

South Korea has chosen five elite teams to lead its national AI project, aiming to accelerate the development of advanced AI technologies and strengthen the country’s global competitiveness in the field. This initiative is expected to foster innovation, support local AI talent, and position South Korea as a major player in the rapidly evolving AI landscape.

Taking an open model from discovery to a production-ready endpoint on Vertex AI

Hacker News - AIAug 4

The article outlines how developers can efficiently take open-source AI models from initial discovery to deployment as production-ready endpoints using Google Cloud’s Vertex AI platform. It highlights Vertex AI’s streamlined workflow for model customization, evaluation, and scalable serving, making it easier for organizations to operationalize open models. This approach lowers barriers to adopting advanced AI solutions and accelerates innovation in the field.

Show HN: I drank the AI Kool-Aid and stopped coding after 40 years

Hacker News - AIAug 4

A veteran programmer with over 40 years of experience has stopped coding manually, instead using AI systems to build software more efficiently. This shift highlights how AI is transforming software development, potentially redefining the role of programmers and accelerating the pace of innovation in the field.