rollup.
96 writings found
Latest Archives
Teaching AI to Actually Navigate Maps: Why Path Tracing Is Harder Than It Looks
Google's MapTrace shows how synthetic data generation can teach multimodal LLMs spatial reasoning they never learned from internet scraping.
The Map That Became the Territory: AI, Specifications, and What We Mean When We Say 'I Built This'
On AI agents, observability, bespoke software, and the uncomfortable question of who actually built what when LLMs generate our code.
Agentic Engineering: The Professional's Approach to AI-Assisted Development
Simon Willison's new project documents coding patterns for the emerging discipline where professional engineers use AI agents that generate, execute, and iterate on code.
Teaching AI to Navigate: Why Path Tracing on Maps Is Harder Than It Looks
Google's MapTrace reveals a surprising gap in AI capabilities: multimodal models can recognize images but struggle with basic spatial navigation on maps.
Teaching AI to Navigate Maps Like Humans Do
Google's MapTrace shows how synthetic data generation can teach multimodal models spatial reasoning they never learned from training data alone.
Sam Altman's Energy Math Doesn't Add Up (And That's a Problem)
OpenAI's CEO dismisses AI water usage concerns and compares training models to human evolution. Let's unpack why this defense is deeply flawed.
Sam Altman Says AI Water Concerns Are Fake. Let's Actually Think About That.
Sam Altman dismissed AI water usage fears as 'totally fake' while defending energy costs. Here's what developers should actually take from this.
AI Is Quietly Deciding Which Languages Developers Choose Next
TypeScript just topped GitHub's language charts. The real story isn't the milestone, it's why AI compatibility is silently reshaping every tech decision you make.
The Governor Is Gone: AI, Cognitive Limits, and the Mess We're Making
AI removed the natural ceiling on how much we can produce. Now the only limit is cognitive endurance, and most of us are blowing past it.
Just-in-Time Tests: How Meta Is Rethinking Software Testing for the Agentic Era
Meta's JiTTests use LLMs to auto-generate bespoke tests per code change, killing false positives and test maintenance overhead in agentic dev workflows.
AI in the ER: Why The Pitt Gets the Tech Conversation Right
HBO's The Pitt quietly nails the core tension of deploying AI in high-stakes workplaces. Here's what developers should take away from it.
Teaching AI to Navigate Maps Like Humans Do
Google's MapTrace tackles a fundamental gap in multimodal models: teaching AI to trace valid paths on maps through synthetic data generation.
Teaching AI Models to Actually Read Maps: Google's MapTrace Pipeline
Google researchers built a synthetic data pipeline to teach multimodal LLMs spatial reasoning. Turns out, tracing paths on maps is surprisingly hard for AI.
Building a Live Documentation Ecosystem for AI Coding Agents
Showboat's new remote streaming feature and companion tools create a real-time feedback loop for AI-assisted development workflows.
Just-in-Time Tests: When AI Writes Your Tests Right Before Deployment
Meta's Catching JiTTests use LLMs to auto-generate tests on-demand, targeting regressions without maintenance overhead. A radical shift in testing philosophy.