
HCI Researcher trying to change how we use computers. Developer. Designer. Proudly working on @openlibrary at @internetarchive as a software developer.
Toronto
Joined August 2008
- Tweets1.1K
- Following242
- Followers163
- Likes2.5K
Just saw a non-English, auto-dubbed short pop up in YouTube, and it completely worked. Tower of Babel level one?
Meanwhile, captions were only available in Korean! How odd.
youtube.com/shorts/3KNLkkWFU…
1
1
157
If I had to write a quiz, I'd focus on the topic and write something probably in markdown.
If had to write it in JSON, I'd likely waste focus escaping newlines/quotes, and might simplify my questions to avoid escapes.
I wonder if LLMs do the same; does format impact quality?
19
France/Spain are two of the only countries I can find that have a _decreasing_ obesity rate -- data thanks to @OurWorldInData ! Whatever they started doing ~2005 - 2010 seems to be working.
1
28
An opinion is like a flash of light in the dark. It can provide guidance. Or it can be a reflection off a wave. But an argument is like a lighthouse. It can help guide your way, or keep you out of treacherous waters.
36
Wait wait wait, hold the phone. Did it, while "thinking", *manipulate the image and decide to rotate it 90 degrees to better read the text*?!?!
"o3, look up these books and give me a table of Amazon reviews, a one-sentence witty summary of the plot (no spoilers) & a link to where to buy them from an indie bookstore"
Yes, the links all work. But impressive stuff you might miss is that it figured out half-obscured covers.
28
One silver lining, I think this is the first time I've seen ChatGPT express doubt in its memory. It could just be stylistic hallucination, but maybe it's linked to some internal notion of uncertainty!

ALT Me: I'm reading it now. There's a scene where they're playing chess. Do you know the scene? ChatGPT: Yeah, I know the scene you're talking about. In that scene, if I remember right, Lev plays chess against a German officer (or at least, they're deep behind enemy lines and chess becomes this weird bridge between enemies and captors).
On one hand the new GPT-4o isn’t doing as many emojis.
On the other, it is slowly driving me insane by responding to everything like an overly enthusiastic 1990s teenager.
1
66
Really fun dev content from @jlengstorf &co! Burned through all of these last night, and through some of the web dev challenges. Love seeing how much the content has grown since last I checked this out! Looking forward to the next one.
NOW STREAMING ON CODETV — this episode got so spicy that host @marktechson almost called it off for safety. Chris and Jeremy go head to head on web dev questions about CSS, authentication, tech acronyms, and more in Leet Heat
link in the thread ⬇️
For the price of a Spotify subscription, you can buy and own an MP3 album every month from @7digital_CA , and build a music collection you can literally pass down to your kids.
3
24
Today's Twitter algorithm trend appears to be: questions! 7 of the first 10 tweets on my "For you" tab are questions.
2
31
This feels like the very very early days of a holodeck
Google presents Genie
Generative Interactive Environments
introduce Genie, the first generative interactive environment trained in an unsupervised manner from unlabelled Internet videos. The model can be prompted to generate an endless variety of action-controllable virtual worlds described through text, synthetic images, photographs, and even sketches. At 11B parameters, Genie can be considered a foundation world model. It is comprised of a spatiotemporal video tokenizer, an autoregressive dynamics model, and a simple and scalable latent action model. Genie enables users to act in the generated environments on a frame-by-frame basis despite training without any ground-truth action labels or other domain-specific requirements typically found in the world model literature. Further the resulting learned latent action space facilitates training agents to imitate behaviors from unseen videos, opening the path for training generalist agents of the future.
1
64
This has to be one of the prettiest UI loading effects I've ever seen!
Has anyone noticed that the Microsoft Copilot version of Dalle-3 has gotten much better? You can segment any section of the image now. It seems we can only blur what's behind the segment or add more color to the segmented object but it's definitely getting better.
The restylizations work really well too.
1
40















