@domschats
Account based inUnited States
About this account
- Account based in
- United States
- Connected via
- United States App Store
Account-level information from X, not a live location or the device used for a specific post.
Head of Social @wisprflow | LA 2B+ organic views generated opinions mine
Joined June 2026
- Tweets165
- Following82
- Followers487
- Likes444
im the opposite of ai. it takes me an hour to write one linkedin post and i haven’t had a sip of water today
after several years in social I’ve realized that everything is as simple as this
the more you post, the better life gets
alexander wang’s profile since muse launched is the best evidence that you don’t need to take yourself serious on twitter to be respected by a technical crowd. muse is everywhere and yet his shitposting is at an all-time high. they go hand in hand now if done correctly
if I run out of creative ideas for social I go watch:
bymaximise
dairy boy
creator camp
varvaramp4
will schafer
old apple vids
casey neistat
camelogs
some of openai’s recent stuff
good bit of notion’s stuff
i have dozens of ideas after, but i know there’s more out there I’m missing
who do yall watch to spark ideas?
if you:
played sports
were a gamer
had niche obsessions
were chronically online
started a lemonade stand
sold stuff online
were super competitive
obsessed over stats
loved internet culture
you’ve basically been training to succeed at social media your entire life without realizing it
Dominic Bair retweeted
Today, the Wispr Advanced Interfaces Lab is introducing Canto, our latest speech model for real-time dictation.
We evaluated it on 10 hours of English-language Wispr Flow dictations from more than 2,300 unique speakers, randomly sampled across applications and use cases.
Canto achieved a 3.4% word error rate—the lowest among the models in our comparison on this evaluation. The comparison included models from Google, OpenAI, AssemblyAI, and Deepgram.
We also evaluated Canto on a separate three-hour challenge set containing noisy, low-volume, far-field, whispered, and very short speech. Canto ranked second overall, behind Gemini 3.1 Pro, and achieved the lowest word error rate among the real-time transcription models evaluated.
Canto is the first model from the Wispr Advanced Interfaces Lab and the foundation for a broader research program spanning contextual speech recognition, personalization, diarization, and difficult audio conditions.
We’re moving fast and already training its successor.
Our Chief Scientist, Ariya Rastrow, shared more about Canto below:
Dominic Bair retweeted
We have a few core beliefs at Wispr that guide everything we do.
One of them is that you should be able to dictate wherever you are in the world. That means understanding you in a busy coffee shop, with a conversation happening at the next desk, or when you’re speaking quietly in a shared office space.
That's why, today, we're introducing our latest speech model, Canto. It's the first in a rapid series of speech models we’re building to make that belief a reality.
We built it specifically around those everyday conditions.
On our evaluation of 10 hours of English-language Wispr Flow dictations from more than 2,300 speakers, it achieved a 3.4% word error rate—the lowest among the models we tested.
We’re moving into a voice-led future, and I’m happy to share a closer look at the work our team is doing to get us there.
Learn more about Canto here: wisprflow.ai/canto
Dominic Bair retweeted
Introducing Canto, our latest speech model for real-time dictation.
On our evaluation of 9.8 hours of English Flow dictations from more than 2,300 speakers, Canto achieved a 3.4% word error rate—the lowest among the models we tested.
Learn more about Canto from our CSO, Ariya Rastrow:
Dominic Bair retweeted
Your meeting context, now in Gemini.
We’ve rolled out Wispr Flow Notetaker’s one-click MCP connection to Gemini. Bring your meeting notes straight into your Gemini workflows, all without copying and pasting.
You should work very hard to fight the knee jerk reaction to open an LLM when you’re faced with a difficult problem
If you lose the ability to think critically about a problem and break it down into components, you’ll soon lose the knowledge needed to instruct these super capable models
This is especially true for folks in marketing - human input will win big time in a world of slop
Hiring a twitter genius at @WisprFlow
You are the right person if you:
-Have experience managing founder led content on twitter
-Can translate technical content and brain dumps into twitter native viral posts
-Have done 10M+ impressions on twitter
-Bonus if you have a technical background
-Bonus if you have experience clipping video content
Respond or DM with clear evidence of these things and let’s chat
Dictation + notes in one place :)
Notetaker is now live on Windows!
When Friday rolls around and you’re trying to remember what came up in Monday’s meeting, just ask. Notetaker searches your recorded meetings and links you back to the conversation.
You also have:
- Accurate transcription in 20+ languages
- A personal dictionary for the names and acronyms your team uses
- Meeting context you can bring into Claude, ChatGPT, and Gemini
Watch the walkthrough to see it in action👇
so glad my ancestors that likely persevered through plagues, famine, and war survived. now i get to talk to my computer 12 hours a day and tell it to make no mistakes
Dominic Bair retweeted
we always try to ship sooner than something is ready
doing this allows us to constantly build alongside real user feedback
if you don't do this you can find yourself in this perfection mindset making very marginal improvements, and before you know it, a year has passed and nothing has shipped
this is what happens to most startups, and exactly what you want to avoid
ship fast, gather feedback, improve, ship again
update you guys in 88 hours
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
