Faster doesn't mean smarter. This article slides from one to the other and hopes you don't notice. Speed is throughput. Accuracy is a separate question, and nothing here suggests the two move together. Synthetic audiences are most accurate on what we already know about people. They are least accurate on what market research actually exists to find out, which is how people will react to something they have not seen before. The training data does not contain the future. And the closer gives him away. Eat the popcorn and watch the show. Nobody who actually thinks their industry might be hollowed out ends on a shrug. That is not the voice of someone wrestling. That is the voice of someone who has already made his peace.

Seriously. The pace of frontier model updates is insane. Most days I feel like I'm being punked.

According to Every, "It’s much faster than Opus 4.7, easier to collaborate with, better at writing than any OpenAI model we’ve used since GPT-4.5 and GPT-4o, and the strongest model we’ve tested on our new Senior Engineer Benchmark, which measures how well models can rewrite a slop-coded codebase the way a senior engineer would."

"A new study out of Drexel University shows just how self-aware the kids have become. In a wide-ranging survey of hundreds of Reddit posts, a team of information scientists found that adolescent users of AI chatbots are increasingly conscious of the negative side effects the tech is having on their lives, even as they sometimes display signs of intense addiction."

Frontier models are failing one in three attemps on structured benchmarks. "This uneven, unpredictable performance is what the AI Index calls the "jagged frontier," a term coined by AI researcher Ethan Mollick to describe the boundary where AI excels and then suddenly fails." This is also why we're not rushing headlong to implement at Tangilla.

As my friend Gregarious Narain said on LinkedIn, "How you use AI matters more than whether you use it. People who grabbed direct answers degraded fastest. Those who asked for hints or partial guidance held up far better. We keep framing this as a productivity story. It's not. It's a capability story. The question isn't whether AI helps you do more. It's whether the version of you using it will still be able to do anything without it."

I just tested this two ways. First I used the exact same prompt with the exact same content and Got a completely new design aesthetic. And the questions asked to help imrpove the design we more intelligent. Then, I uploaded the previous result from Claude Opus 4.6 and asked it to improve it, with some specific instructions. Once again, the results were excellent. I'll be playing with this quite a bit in the coming days.