Teleoflexuous

System started doing Step 1, believed it was over so moved to Step 2 and when was asked to go back, kept going back to step 2.

Step 1 being Step 0 and Step 1 combined also works.

Again, it's also a weird way to prerecord. If you're prerecording, you're prerecording all steps and practicing with them prerecorded. I can't imagine anyone to be able to go through a single rehearsal with prerecorded audio to not figure out how to do this, we have the technology.

You ask AI how to do something. AI generates steps to do that thing. It has concept of steps, so that when you go 'back' it goes back to the last step. As you ask how to do something, it finishes explaining general idea and goes to first step. You interrupt it. It assumes it went through the first step and won't let you go back.

The first step here was mixing some sauces. That's it. It's a dumb way to make a tool, but if I wanted to make one that will work for a demo, I'd do that. Have you ever tried any voice thing to guide you through something? Convincing Gemini that something it described didn't happen takes a direct explanation of 'X didn't happen' and doesn't work perfectly.

It still didn't work, it absolutely wasn't wi-fi issue and lmao, technology of the future in $2T company, it just doesn't seem rigged.

Ideally, I'd like to allow sharing and storing of presets, but it was simply out of scope for the PoC - the functionality is there in the desktop version btw, but it on the other hand asks users to download an unknown .exe and then share mp3 and json files with each other, putting us firmly in the mid-90s'.

It's especially surprising if the 'audio playing' icon is there, since that should be coming from the browser itself.

Terms you want to check for more detailed info are 'liquid intelligence' and 'crystalized intelligence', but you basically nailed it.

I wish I was hiring, if that's what you're asking ;) Otherwise, if you have any ideas for processing formulas (even just for reading them out, but any extra steps towards expressing what they mean - ' 'sum divided by count' is 'mean'/'average' value ' being the most simple example I can think of) I'd love to hear them. Novel ideas in technical papers are often expressed with formulas which aren't that complicated conceptually, but are critical to understanding the whole paper and that was another piece I was having very mixed results with.

Two approaches I had most, but not full, success with are: 1) converting to image with pdf2image, then reading with pytesseract 2) throwing whole pdfs into pypdf 3) experimental multimodal models

You can get more if you make content more predictable (if you know this part is going to be pure text just put it in pypdf, if you know this is going to be a math formula explain the field to the model and have it read it back for high accessibility needs audience) the better it will go, but it continues to be a nightmare and a bottleneck.

There's a lot of similar apps, people seem to use and like all of them, I'm just curious how it looked from the inside of this one.