GitHub | prev MSFT, Mozilla, Google, Barclays. Founder https://teapawt.net/t.co/ud6J5BGC0K | a16z Talent Engineer Fellow | Opinions my own
Raleigh, NC
Joined November 2024
- Tweets592
- Following48
- Followers335
- Likes402
Pinned Tweet
6 months ago I was certain I needed to learn framer & motion design. now, you just tell Fable "beating human heart, field-scanner HUD, live ECG, synth sound... in html. make no mistakes"
disclaimer: this was not one-shotted, it was iterative and took me ~45 minutes
made with @claudeai
HTML isn't just markup anymore.
For AI agents, it's becoming the best visualization layer.
Made with @HyperFrames_
Shoutout to @equinusocio for the css mask inspo
🔊 Don't mute this
$10,290,000 of $NVDA $245 calls were just purchased at $1.96 per contract.
Expiring Feb 19, 2027.
2
84
I built Alfred after seeing how many great hires come from someone a coworker already knows. Looking for 3-4 design partners: recruiters or founders who hire through Slack.
A job post brings in hundreds of applicants. Some of the best people never apply: they're someone a coworker already knows.
Alfred finds them in Slack. A teammate who knows the person invites them to talk, and the recruiter only hears back if the person wants to.
Looking for 3-4 design partners to try it free. How it works: entopiq.com/alfred
1
1
135
3/ pair the Virtual Key.
open tesla.com/_ak/yourdomain.com on your phone, approve it in the tesla app
commands have to be signed now. Tesla's open source vehicle-command tool handles it
read battery and charge state first, then add controls
1
44
embedding models are underrated. ran embeddinggemma 2 on my mac mini and it finds moments in NASA's artemis launch footage from a text description. no cloud. its early but on-device search is about to get really good. spotlight on iOS/macOS needs this
Introducing EmbeddingGemma 2! 🚀
Our lightweight, multimodal embedding model maps text, code, images, video, and audio into a single, unified embedding space. Optimized for on-device use cases, it features:
- 740M parameter form factor with modular encoders
- Flexible dimension sizes (768dim-128dim) via Matryoshka Representation Learning (MRL)
- 8K context window (4x larger than text-only EmbeddingGemma)
- A commercially permissive Apache 2.0 license
part 2 w Jev image boxing vs opus 5.5: teapawt.net/Daniel_An23/status/211…
part 2. @typesafeai cooked. jev 1.14.0 with images is underrated. gemma finds the precise moment, jev boxes it. head to head with opus 5.5 on 540 frames of nasa launch footage: boxes on par (93% vs 87%), ~140x cheaper, ~23x faster. $0.02 vs $3.20 for the entire run. opus edges slightly on yes/no call; for boxing what's in frame, I'd pick jev for prod
36
part 2. @typesafeai cooked. jev 1.14.0 with images is underrated. gemma finds the precise moment, jev boxes it. head to head with opus 5.5 on 540 frames of nasa launch footage: boxes on par (93% vs 87%), ~140x cheaper, ~23x faster. $0.02 vs $3.20 for the entire run. opus edges slightly on yes/no call; for boxing what's in frame, I'd pick jev for prod
prediction: the bar for sharing presentations is soon going to be html. some are already doing this but the status quo is still "send me your powerpoint deck"
for me, it's just "here's the web app"
I get that there are some practical (or even psychological) limitations to pivoting away from decks ...but the time spent manually building slides is not trivial. i've been using html to wire up landing page mocks, slides etc for over a year. maybe im just an outlier
curious if there are any founders (or corporate employees) who have banned slide decks or planning to?
1
1
135
this is a fleet map of my 18 grok bots. every handoff from one bot to another is a streak of light
77