
Chris Wandstrom
286 posts
Chris Wandstrom
@wandstromfilter
Early adopter with more TestFlights than tweets. Occasional thoughts on apps & workflows



OpenAI is testing a new simplified model picker in the ChatGPT web app prompt composer called "Intelligence Picker" Instead of choosing a model and reasoning effort separately, you just pick from "Instant", "Medium" (Thinking @ Standard), "High" (Thinking @ Extended) and "Pro" (Pro @ Extended) The available options are remotely configurable and probably depend on your subscription plan

Codex now works directly in Chrome on macOS and Windows. It’s even better at working with apps and sites in Chrome, and now works in parallel across tabs in the background without taking over your browser. To get started, install the Chrome plugin in the Codex app.


New in the Codex app: - GPT-5.5 - Browser control - Sheets & Slides - Docs & PDFs - OS-wide dictation - Auto-review mode Enjoy!





iOS Dictation Using Shortcuts and the OpenAI API: Due to iOS limitations, dictation apps usually require switching keyboards and opening the app every single time you want to dictate something. I think this just takes too long. The shortcut below gets around this. It makes use of the OpenAI API for dictation and cleanup and usually has the transcribed text in your clipboard within 3–7 seconds (for dictation lengths of ~2 minutes). Recognition is very good, better than the other models I tested (ElevenLabs, Soniox, Parakeet locally, etc.), especially for niche terms. Costs are ~0.10$ per ~20 minutes and ~2000 words of dictated text. You can easily trigger the shortcut with the Action Button (Settings > Action Button), Back Tap (Settings > Accessibility > Touch > Back Tap), or a widget (lock screen, home screen, Control Center). It also appears in the Share Sheet, so you can transcribe audio files as well. ⬇️

You can now use your ChatGPT subscription to dictate anywhere on your desktop now! Have fun!


iOS Dictation Using Shortcuts and the OpenAI API: Due to iOS limitations, dictation apps usually require switching keyboards and opening the app every single time you want to dictate something. I think this just takes too long. The shortcut below gets around this. It makes use of the OpenAI API for dictation and cleanup and usually has the transcribed text in your clipboard within 3–7 seconds (for dictation lengths of ~2 minutes). Recognition is very good, better than the other models I tested (ElevenLabs, Soniox, Parakeet locally, etc.), especially for niche terms. Costs are ~0.10$ per ~20 minutes and ~2000 words of dictated text. You can easily trigger the shortcut with the Action Button (Settings > Action Button), Back Tap (Settings > Accessibility > Touch > Back Tap), or a widget (lock screen, home screen, Control Center). It also appears in the Share Sheet, so you can transcribe audio files as well. ⬇️

Do we have a clear explanation of what seems to have driven the performance gain between 4.7 and Mythos? Is it just that we scaled up on both model size and RL?
Finally got Codex Super Bowl swag, found great use for Codex key









