Machine learning & AI podcaster, community builder and all around enthusiast. Creator of the @TWIMLAI Podcast, TWIMLcon, TWIMLfest & the TWIML Solutions Guide.
St. Louis, MO
Joined April 2007
- Tweets19.3K
- Following2.8K
- Followers18.9K
- Likes15.3K
If you were in ML/AI before genAI, you probably remember the diagram from the “high-interest credit card” paper: a tiny box labeled “ML code,” surrounded by all the machinery needed to make it useful. My conversation with @Canva’s Stef Corazza reminded me of that.
The models are interesting, but so is everything around them: editable design representations, tools for agents, visual feedback, evaluation, and inference fast enough to make iteration practical.
We cover how those pieces fit together in Canva’s approach to graphic design.
Sam Charrington retweeted
I asked Fable 5.2 to "go wild" making a game, and all I got was this Pong clone.
And then I played it ...
👇
A few weeks ago, everyone was writing @Google out of the AI race. Since then, they've shipped dozens of announcements across Gemini and @GoogleCloud. Here's what mattered 👇
1
2
445
6/ Full breakdown: twimlai.com/articles/googles…
1
146
5/ Google Cloud now offers Gemini, Claude, AND GPT. Google says it's the only hyperscaler with all three.
1
82
6/ Full breakdown: twimlai.com/articles/googles…
52
Sam Charrington retweeted
Canva says its design models cost 40x less than the most expensive frontier models. The bet behind it was a model with two orders of magnitude fewer parameters.
The 40x is for the same design task. The frontier models just need many more parameters to do it.
Stefano Corazza's reasoning is simple. If you have the right data, you can fine tune or train a much smaller model and beat the bigger one. Because it's smaller, it also runs about another order of magnitude faster and cheaper.
He says quality is the same, sometimes better. The speed and cost drop is the bigger win.
I'd treat this as a warning for anyone who defaults to the biggest model for a narrow job. If your product produces data nobody else has, test a small model trained on it before you sign up for frontier pricing.
Stefano Corazza on The TWIML AI Podcast with @samcharrington
I'm pretty sure this is real and not a parody. I think they should have went for mutter.ai.
To their credit, it's better than the talking into a funnel thing, but I don't think this is quite it.
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
1
660
