@xingsthatmatter profile picture
Sophia Xing @xingsthatmatter

making space for creative vibes. Founder @plotpartyai, formerly @inworld_ai @princeton

Joined July 2018
  • Posts 1,343
  • Following 862
  • Followers 2,258
222 Photos and videos
Pinned Post
Sophia Xing
@xingsthatmatter
Jul 1
We brought Blender into AI movie generation. This will save so much time for anyone creating AI movies or dramas! The most controllable way to direct consistent frames featuring complex spaces and multiple characters. At @plotpartyai we've added a 3D Director Console so you can extract compositions into 3D, easily tune exact shot configurations, and natively embedded into your AI generation workflow. This has been our most ambitious feature to date, where we worked across model and interface layers, rethought the entire experience, and brought two apparently different paradigms under the same experience. Give it a shot (pun intended)!
2:05
26
51
367
37,367
3. Compute scarcity makes applications relatively more important. If you cannot always outspend OpenAI on training, another way to win is to make AI available everywhere. A model that’s 95% as capable but costs half as much to run can unlock far more deployment across manufacturing, logistics, ecommerce, robotics and local governments.
1
1
85
4. Domestic Chinese chips actually strengthen this dynamic. People assume better domestic chips mean China will become more closed. I’m not convinced. Domestic chips reduce dependence on Nvidia. Open models increase demand for domestic compute. Those two reinforce each other. The more companies deploy open models: - the more inference they run - the more domestic cloud demand grows - the more incentive there is to improve domestic chips - which lowers deployment costs further It’s a positive feedback loop.
2
1
88
5. The US AI strategy was largely born in an environment of abundance: - abundant venture capital - abundant GPUs - abundant hyperscale infrastructure So the natural question became: How do we build the best model and capture the economic value? China’s recent AI strategy has developed under more constrained conditions: - export controls - tighter access to frontier chips - a policy emphasis on self-reliance That changes the question to: How do we maximize the output of the entire ecosystem given limited frontier compute? Open models, shared engineering improvements, cheaper inference, and broad deployment all become more attractive in that environment.
1
83
1. You can’t brute force your way to leadership. The US frontier labs can spend enormous amounts of compute pushing the frontier. But when you’re compute constrained, you need to win on training/inference efficiency and creative architectural designs. These innovations benefit from open publication because everyone builds on them. Back in early 2025 - DeepSeek’s biggest contributions weren’t just model weights, they also published technical reports and optimization techniques that the community quickly adopted. In an open ecosystem, every company running that model benefits. Suddenly startups spend less, enterprises deploy more, domestic clouds become more competitive as AI adoption accelerates. If your national objective is ecosystem competitiveness rather than maximizing one company’s software margin, open sourcing has real value.
1
2
177
2. Scarcity makes duplication expensive, especially for those playing catch up Imagine five Chinese labs independently solving the same optimization problem. That’s a waste of scarce compute and engineering resources. If the ecosystem shares more openly, later entrants can start from a stronger baseline instead of repeatedly reinventing foundational techniques. Open source effectively pools innovation.
1
1
128
I actually think the chip/open source angle is usually explained incorrectly. People say: “China has fewer chips, so they have to open source.” That’s not really it. The deeper mechanism is: When compute is scarce, maximizing the return on every unit of compute becomes strategically more important than maximizing software margins. A 🧵to explore what scarcity changes strategies in this global AI race.
2
5
370
New discovery - Murdoku: sudoku murder mystery puzzle If you love casual puzzle games like I do, you will enjoy this one!
2
6
1,593
apparently you need a child to go on this slide, anyone has an extra?
1
10
3,277
I'm a founder who travels between US and China frequently. A couple days ago I posted about the things China quietly does better than the US. It clearly hit a nerve. Here's the honest flip side - things about the US that makes a difference. Not about politics. Just daily life: 🧵
57
28
423
207,684
5. Kids are trained more holistically. Sports teams, group projects, debate, a summer job at 16 etc. The scaffolding around school drills what the working world rewards: teamwork, speaking up, leading, losing gracefully, selling an idea. That being said, Chinese schooling wins on rigor and fundamentals, so the tradeoffs here depending on parenting ethos more than anything
5
5
78
26,497
and...anything I've missed?
15
16
24,487
Experimenting with AI realism lately. Tried a bunch of configurations and finally zoning in on ones that work.
0:08
2
12
3,269
The biggest lie in AI is that it made creativity valuable. Creativity was always valuable. What AI actually did was make plausible execution cheap. And when execution gets cheap, judgment gets expensive.
4
1
13
2,715
What would you set free this July 4th? Inspired by the upcoming holiday, created with @plotpartyai
1:08
2
1
6
2,948
shipped a feature and remembered users prefer founders with signs of life
4
21
2,726
said feature x.com/xingsthatmatter/status…
Sophia Xing
@xingsthatmatter
Jul 1
We brought Blender into AI movie generation. This will save so much time for anyone creating AI movies or dramas! The most controllable way to direct consistent frames featuring complex spaces and multiple characters. At @plotpartyai we've added a 3D Director Console so you can extract compositions into 3D, easily tune exact shot configurations, and natively embedded into your AI generation workflow. This has been our most ambitious feature to date, where we worked across model and interface layers, rethought the entire experience, and brought two apparently different paradigms under the same experience. Give it a shot (pun intended)!
2:05
3
2,533
Load more