Founders

by far my biggest unlock with LLMs in the last 6 months has been using their native voice-to-text for describing complex problems. NOT voice chat, voice-to-text

LLMs thrive the more info you give them. You can ramble for 5-10 mins and it will sift through and keep what matters.

🌅 Head of Product & dev ~ horizon.io         
     ⚒️ sequence.xyz, play.skyweaver.net   
            
📝 ERC-1155 and ERC-1271 co-author


twitter.com/PhABCD

I personally find it much easier to walk around and talk than type. Helps me provide much more info and get better results

Working on Farcaster and Warpcast. Longer thoughts at https://dwr.email

Mobile is actually excellent BUT you must keep messages under 4-5 minutes. Haven’t found the exact number but if you go too long in a single input it fails to transcribe. Desktop is the same.

Fortunately they now added ability to do multiple text-to-input in single prompt so you just need to split your voice msg

How do you capture the raw input?

I've found ChatGPT mobile app to be flaky on input.

Full-stack digital product consultant @ BoundCorp.net - eng @daylight

this, and whisper is SO GOOD

catches slang, proper nouns, seems to respect irregular phrasing pretty well, handles voice changes like intonation and mimicking - so far beyond the last generation of stt

This, using one of the whisper apps

It's also great for giving feedback on longer docs. Just turn on the recorder and read, calling out changes when you find them. Paste your transcript back into the model. Repeat until the doc is where you want it.

i use this in the opposite way : text - voice using notebook LLM ; great companion to learn new things by generating small podcasts..