Anthropic's performance optimization team has been using a take‑home test for job applicants since 2024. As AI coding assistants like Claude have improved, the company has repeatedly redesigned the test to keep human candidates distinguishable. Team lead Tristan Hume noted that each new Claude model, from Opus 4 to Opus 4.5, has forced a redesign because the AI began matching or exceeding top human performers. While candidates are permitted to use AI tools, the situation raises questions about the test's ability to assess human talent. Hume ultimately created a new, hardware‑agnostic test intended to stump current AI tools and invited the community to propose better challenges. Read more →

Meta is reportedly evaluating the integration of competing artificial‑intelligence models, including Google Gemini and OpenAI offerings, to enhance the AI features of its applications. The company’s Superintelligence Lab has examined how external models could complement Meta’s own Llama models, while internal teams already use Anthropic’s technology for coding assistance. A Meta spokesperson emphasized an “all‑of‑the‑above approach to building the best AI products,” and the firm is offering attractive compensation to draw AI researchers from rival firms. Read more →