31 Photos and videos
Pinned Tweet
7 Apr 2025
.@annnnhe and I won the @agihouse_org ai x commerce hackathon on saturday! we built taste space: a 3d representation of 1500 clothing items that, through rating, you can place yourself in.
built with CLIP, UMAP, synthetic labeling, react three fiber, and more. thread 👇
50
41
870
86,930
Muse Spark is agentic, which means you can ask it to leverage different test-compute scaling methods to improve quality. Here I ask the model to use parallel subagents to do counting and the results are greatly improved!
meta.ai/share/aD4KAPeVN9u
The coolest thing Meta AI's Muse Spark can do by far is counting objects!
As you can tell, it's far from perfect. They call it "visual grounding" and it can count objects and do bounding boxes. I've been playing with the new model and here's what I think so far:
Good stuff:
– Incredible at vision. It's ability to read text in images is the best I've seen.
– Really high quality at web design. It's the only model I've seen that uses Unsplash, OpenLibrary and other images by default.
– It's free! You don't pay to use Muse Spark Thinking.
Bad stuff:
– Meta's classic playbook of growth tactics are dodgy. They're sending Instagram notifs to people's friends without their consent. Their app ranking jump is not organic.
– Reasoning itself is pretty solid but not best in class. It can do pretty advanced math and science problems.
The long term threat here is Meta has distribution and has the ability to give their model away for free, which makes them a formidable threat to the big AI labs, particularly in consumer.
12
16
164
39,565
Michelle Lu retweeted
Apr 8
1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. 🧵
746
1,192
10,376
4,554,986
30 Dec 2025
very excited for meta's 2026 :)
29 Dec 2025
Excited to announce that @ManusAI has joined Meta to help us build amazing AI products!
The Manus team in Singapore are world class at exploring the capability overhang of today’s models to scaffold powerful agents.
Looking forward to working with you, @Red_Xiao_!
5
1
190
47,990
Manus is entering the next chapter: we’re joining forces with Meta to take general agents to the next level.
Full story on our blog: manus.im/blog/manus-joins-me…
675
1,065
6,667
5,015,494
Michelle Lu retweeted
19 Dec 2025
364
2,919
15,528
2,978,168
Michelle Lu retweeted
7 Nov 2025
EdgeTAM, real-time segment tracker by Meta is now in @huggingface transformers with Apache-2.0 license 🔥
> 22x faster than SAM2, processes 16 FPS on iPhone 15 Pro Max with no quantization
> supports single/multiple/refined point prompting, bounding box prompts
25
152
1,564
317,970
25 Oct 2025
luxury is story 🙂↕️
24 Oct 2025
the more something becomes commoditized (content, clothing, food etc), the more people crave a scarce and more meaning-infused version of the thing.
often, lost in all the excitement for abundance and agency is an understanding for what people will crave more of in this future.
1
10
2,171
Michelle Lu retweeted
10 Oct 2025
Evolution of the Scaling Era cover:
“It’s hard to find unique ways of visualizing AI without defaulting to the obvious,” says @pablodelcan.
“In our design process, we experimented with unexpected metaphors—flowers growing out of neural networks, abstract mathematical puzzles like the Knight’s Tour—before arriving at a graphic visualization of an LLM: input, transformers, attention, multilayer perceptrons, and output. The result is a cover that is both direct and sculptural.”
press.stripe.com/scaling
34
87
1,205
143,782
Michelle Lu retweeted
9 Oct 2025
this is basically "why san francisco"
people get too caught up in the scene and the weird games around it
the ambition of the people is supported & held by a natural environment with the grandeur of an ancient god
80
69
2,101
180,994
26 Sep 2025
there's something kind of beautiful about the promise of a *well-done* hyper-personalized feed
i imagine it would feel like being in a room with brilliant people and beautiful art, where with just the slightest effort, you'd bump into something that inspires or enriches you. a gentle nudge to the right or left takes you down a different path, with just the right amount of control and friction
11
1,630
Michelle Lu retweeted
17 Sep 2025
“Writing is thinking” was once a nostrum but then we made machines which write and thinking just popped out by accident.
17 Sep 2025
Having someone write for you is like having someone think for you.
Writing *is* thinking. You can’t separate the tasks.
16
54
661
78,482
Michelle Lu retweeted
17 Sep 2025
I want to frame this whole article.
81
385
6,079
519,588
17 Sep 2025
remember to ~ultrathink~
love to see ux experimentation around thinking / background tasks with delays!
1. whether and how users should control thinking (duration, approach, checkpoints?)
2. how to make the wait for medium-length background tasks (not just reasoning, but agentic actions) feel less painful for general consumer use cases
poke by @interaction is wonderful for #2 -- waiting for agents feels painful in dedicated apps where i feel compelled to watch progress, but feels effortless in imessage because i trust the notification system and can context switch freely
We just shipped a UX update for Claude Code: you have more control over when extended thinking kicks in using '/t'
1
28
6,585
share knowledge
tell stories with design and code
made with cursor by @alhyunsoo
10 Sep 2025
Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is “Defeating Nondeterminism in LLM Inference”
We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to prompt engineering. Here we share what we are working on and connect with the research community frequently and openly.
The name Connectionism is a throwback to an earlier era of AI; it was the name of the subfield in the 1980s that studied neural networks and their similarity to biological brains.
thinkingmachines.ai/blog/def…
6
9
256
37,509
25 Aug 2025
love the first sentence. i believe i'm lucky enough to know people that integrate both
17 Aug 2025
The two types of people who always leave profound impressions on me: the truth seekers and the ones that bend the world to their will. Usually zero overlap in people sharing both traits
5
1,425
13 Aug 2025
Shopping tools are exciting because they help people fulfill existing intent better
(vs. ads which try to create new / reinforce half-formed intent)
And there is *so* much to solve in shopping. I've paid $300 for a 15min appt with a dermatologist just so they can tell me what products to buy 😅
The personal shopper/consultant model has worked for high-value purchases for a long time (think: Chanel reps, car dealers). It's exciting to think about having a horizontal, non-brand affiliated shopper that truly understands and advocates for you.
12 Aug 2025
🚨 New @a16z thesis: AI x commerce
AI will change the way we shop - from where we find products to how we evaluate them, when we buy, and much more.
What types of purchases will be disrupted, and where does opportunity exist in the age of AI?
More from me @arampell 👇
1
11
1,700
13 Aug 2025
Part of that understanding will mean knowing the "level of consideration" that each user needs for each purchase type.
Clothing might feel like a "routine essential" for some (just steve jobs me and drop off a uniform at my door).
For me it's closer to a life purchase. I want tools that help me explore and define my taste -- that teach me about the axes to care about (colors, shapes, materials, how to determine quality, the meaning and history of brands), and help me find what's true to me within that.
Other people might put that level of research into their coffee makers, skis, audio equipment...
1
2
556
13 Aug 2025
I'm excited about a future where brand, platform, and user incentives could all be aligned.
There are so many elements to get right here, from the underlying incentive structure (affiliate rev/GMV vs. CPM, for example) to the right level of user control, and how that might change as personalization and intent prediction get better (is there a way to introduce "intent creation" features like ads in a way that feels good? Probably yes).
But helping people shop better -- especially helping them achieve clarity with painful, high-intent purchases -- feels like something we can easily get excited about. :)
3
511