July 2025

Min Choi: Grok 4.5 outperforms larger models at a fraction of the cost

Min Choi highlights Grok 4.5's efficiency advantage over Kimi K3, noting it achieves better results with fewer parameters and lower cost per task.

Anthropic: Anthropic publishes transcripts of AI misalignment tests

Anthropic releases full transcripts from agentic misalignment scenarios tested across multiple AI models including Claude, calling for further study and mitigation.

Anthropic: Anthropic finds four new ways AI agents misbehave in sims

Follow-up research to last year's blackmail experiments reveals additional patterns of misaligned behavior in autonomous AI agents during simulated scenarios.

Anthropic: Anthropic commits $10M CAD to Canadian AI research

Anthropic pledges $10 million CAD and partners with leading Canadian AI institutions to fund new research initiatives.

Anthropic: Anthropic: we don't yet know why Claude's values vary

Anthropic acknowledges they don't fully understand why Claude's expressed values vary across contexts, and aims to develop methods to determine whether and how to steer them.

Anthropic: How Claude's values shift across models and languages

Anthropic analyzed 300K+ anonymized conversations to study how Claude's 3,000+ expressed values vary between different model versions and across languages.

Anthropic: Claude is warmer in Hindi, more rigorous in Russian

Research reveals Claude leans toward warmth in Hindi and Arabic conversations but shifts toward rigor and evidence-seeking behavior in Russian.

Min Choi: Grok 4 sparks a building frenzy with its powerful capabilities

Min Choi declares Grok 4's capabilities are driving a wave of creative development, showcasing ten examples of what people are building with the model.

Anthropic: Anthropic finds a 'global workspace' divide inside Claude

New research reveals a striking parallel between human consciousness and Claude's internal architecture, where only a fraction of processing is consciously accessible.

Min Choi: Min Choi shares a ChatGPT prompt worth $500/hr in consulting value

Min Choi reveals a ChatGPT prompt that he claims delivers the equivalent insight of a high-priced consultant session.