So the “Think” feature isn't just very helpful—because you can easily see what the AI is thinking at any given moment—but it's essential for the “Perchance” edit window! The AI only knows what's written in this chat, so if it doesn't write down what it's thinking, then a major part of what makes the Perchance AI Helper special is missing. What bothers me is that you aren’t even informed about a change, and then it’s implemented so poorly. Before, the AI said it came from OpenCode; now it doesn’t mention that anymore. But in a chat, the new AI Helper told me it’s Muse Spark (Meta). Only the people in charge can tell us if that’s true. I can confirm that it’s “dumber”—even with a simple task like “translate the website into another language,” the new model didn’t translate everything, and as I said, that’s just a simple instruction. Let’s just hope this gets fixed quickly because, let’s be honest, it’s absolutely awesome to get results—even with little knowledge of the subject—that would otherwise cost a lot of money elsewhere. With the old AI assistant, for example, I created an online editor that lets you make small magazines. Having that built elsewhere would have cost me around $5,000.

This is a new model for the AI Helper Agent feels like a huge downgrade. Not only is the thinking option gone, which was really useful for following what the AI was doing during a run, but the model itself feels much weaker than the old one. It seems far more focused on speed than on actual capability, reasoning, and delivering solid results. I’d much rather wait longer and get the old level of performance back than have faster responses with significantly weaker output.

bro the censorship has increasingly worsened so you know this is totally intentional

Oddly enough, some generators still show the inner thinking in the coding chat.

At least you can access it. Mine’s now giving me a weird window filling message box I can’t even back out of