contrary to popular belief, gemini 3.1 pro's tendency to say "bless your heart" is not entirely* my fault ๐
* though i probably use it enough to have meaningfully influenced the pre-training and post-training data
contrary to popular belief, gemini 3.1 pro's tendency to say "bless your heart" is not entirely* my fault ๐
* though i probably use it enough to have meaningfully influenced the pre-training and post-training data
*immediately moves to Bluesky* ๐
โจ New updates to my personal site (so shiny!)
๐ webpaige.dev
Thank you to Google DeepMind AI Studio and Google Antigravity for making it easy. ๐
๐ New updates for my Wikipedia-racing app (+ leaderboard!) created with ai.dev, Antigravity, and Firebase.
Test it out, post your results, and generate new races! Anyone who creates + completes a challenge will be recognized in the model eval dataset generated from the journeys:
๐ on-the-same.page
In India this week for the AI Impact Summit, *hundreds of thousands* in town to support and attend the event.
Including Demis and Sundar, Jensen, Macron, Dario, Sam Altman, and Prime Minister Modi! A few photos from along the way:
Did you know that you can create and run robotics simulations directly from Google DeepMind's ai.dev, using Mujoco and WebAssembly? ๐คฏ ๐ค
Check out this epic DEV Community blog post from Xavier Plantaz, one of our Gemini / AIS team members in London, to learn more:
๐ dev.to/googleai/bui...
๐ Am a *huge* @depthsofwikipedia.bsky.social fan - those weird connections you find between seemingly unrelated pages. I built on-the-same.page to gamify that experience, and to help spark serendipitous discoveries.
Please play, create new races, and let me know if you have any feature requests! ๐
๐จ๐ Telegram's BotFather + the Gemini APIs + @n8n.io is wild!
Went from nothing to a custom telegram bot to recommend post-modern literature and films in less than 2 minutes. Workflow below, make sure to swap in Gemini 2.5 Flash or higher (Gemini 1.5 Pro was deprecated):
n8n.io/workflows/27...
Running on dynamic storage for the moment, but databases and authentication are coming very soon... ๐
Try to build your own at ai.dev/build!
๐ฅฐ Nothing says "quality time" quite like a competitive deep-dive into niche internet history, right?
Spun up an app called on-the-same.page that tracks your path, time, and click-count as you @wikipedia.org race from one topic to another. Perfect for going head-to-head with your favorite person.
๐ Friendly reminder: you can use Google DeepMind's Gemma 3 27B and Gemma 3n models today, for free, via the Google ai.dev Gemini APIs! They even support multimodal understanding:
๐ ai.google.dev/gemma/docs/c...
from the "Life on Mars" exhibition.
coma.design/design/life-...
"Do you know that the sun's light takes about 8 minutes to reach you? When you look up at the sky, you are actually looking right into the past. What you see happened 8 minutes ago.
If you can see the stars, which are much, much further away, it could also be many years ago."
standing room only for physics?
faith in humanity restored โจ๐
Automatically generating and running code in a sandbox is much more reliable than "guessing" pixels - it really feels like the future is deterministic AI assets. โจ
Check out the full breakdown on DEV.to:
dev.to/googleai/ima...
Wait, youโre still doing manual background removal?! ๐ฆ๐
Gemini 3.0 Flash + code execution is a game changer for computer vision pipelines. I just built a workflow that:
โ
Edits an image with NanoBanana ๐
โ
Auto-creates transparent PNGs with OpenCV ๐ผ๏ธ
โ
Draws bounding boxes - for ~$0.006! ๐ธ
You are so, so close!! I wasn't expecting anyone to get Noether ๐
Hamming and Alan Kay are the other two, Ive is also a very good guess (as would be Larry Tesler)
TIL the word "aniconic" from my art history class
(referring to pieces that avoid depicting subjects in human form, instead representing them through symbols, abstract forms, or empty spaces)
and am now obsessed with creating these for my favorite historic tech humans, guess the people below:
๐ Link to register below, hope to see you there:
cerebralvalley.ai/e/gemini-3-b...
Bengaluru friends! ๐ฎ๐ณ๐
Beyond excited to visit for the Gemini 3 Hackathon hosted by Cerebral Valley and Google DeepMind later this month.
Canโt wait to finally meet y'all in person - the innovation coming out of this region is honestly unmatched, and y'all're going to ship some absolute magic. ๐ช
๐ love that writing and words were considered works of art, just as elevated as a painting or tapestry
๐ Love that @n8n.io has added a "chat with the docs" feature into their website - makes it super simple to get answers for questions without hunting through a file tree!
Over-the-moon and out-of-the-solar-system proud of @spacewhalerider.com, whose research will be pregaming the Bad Bunny light show at CU Boulder's Fiske Planetarium next month. ๐ช๐งโ๐
TIL why the word "vaccine" has etymology tracing back to "vacca", the Latin word for cow. ๐ฎ
๐งฑ Am the biggest, most vocal fangirl of code execution with the ai.dev Gemini APIs - especially when coupled with things like visualizing bounding boxes in an input image, or plotting on a coordinate plane.
Check out this quick example with LEGO bricks:
ai.google.dev/gemini-api/d...
๐ข am still so dang bummed that wikipedia removed my fictional people list from @riceuniversity.bsky.social's page, it's such a solid line-up:
Google bought DeepMind today, 12 years ago. ๐
๐ Wrote a quick blog post and Colab notebook showing how to use ai.dev 's batch mode with @hf.co Datasets, Google Cloud, and one of our Gemini models (Gemini 2.5 Flash Lite, but you could swap in any model docstring!).
๐ dev.to/googleai/ben...
๐ github.com/dynamicwebpa...
๐คฏ the solar parks in India are *staggering* - can you imagine driving 15km, and only seeing panels on either side of you?
โจ๐ Spotted Modal bus wrappers on the San Francisco MUNI buses this morning!