Simon Willison — Week of 2026-04-04 to 2026-04-10#

Highlight of the Week#

Anthropic’s decision to delay the general release of their highly capable Claude Mythos model under “Project Glasswing” marks a significant turning point in the AI industry. The move underscores a massive shift in frontier model capabilities, as models evolve from generating text to autonomously chaining multiple minor vulnerabilities into sophisticated exploits, requiring a new level of security safeguards before release.

Simon Willison — 2026-07-10#

Highlight#

Today’s standout piece highlights a sharp critique from Nilay Patel on the unavoidable privacy tradeoffs inherent to augmented reality hardware. It serves as a necessary reality check on the physical limitations of face-worn AI devices and the societal cost of continuous cloud-based processing.

Posts#

Quoting Nilay Patel · Source Simon highlights a stark reality check from Nilay Patel regarding the physical limits and privacy implications of augmented reality glasses. Patel argues that because chips small enough to fit in glasses cannot handle real-time continuous video processing, the data must be sent to the cloud. This unavoidable architecture means that building the next major AR product requires invading user privacy, raising the critical ethical question of whether the societal tradeoffs are too high to justify building these devices at all.

Simon Willison — 2026-04-10#

Highlight#

Simon points out the non-obvious reality that ChatGPT’s Advanced Voice Mode is actually running on an older, weaker model compared to their flagship developer tools. Drawing on insights from Andrej Karpathy, he highlights the widening capability gap between consumer-facing voice interfaces and B2B-focused reasoning models that benefit from verifiable reinforcement learning.

Posts#

ChatGPT voice mode is a weaker model Simon reflects on the counterintuitive fact that OpenAI’s Advanced Voice Mode runs on a GPT-4o era model with an April 2024 knowledge cutoff. Prompted by a tweet from Andrej Karpathy, he contrasts this consumer feature with top-tier coding models capable of coherently restructuring entire codebases or finding system vulnerabilities. Karpathy notes this divergence in capabilities exists because coding tasks offer explicit, verifiable reward functions ideal for reinforcement learning and hold significantly more B2B value.