Maxim Leyzerovich (@round) header
Maxim Leyzerovich (@round) profile picture

design @floraai faculty @SVA_news

brooklyn
Joined July 2008
Photo by Maxim Leyzerovich (@round)Photo by Maxim Leyzerovich (@round)Photo by Maxim Leyzerovich (@round)Photo by Maxim Leyzerovich (@round)Photo by Maxim Leyzerovich (@round)Photo by Maxim Leyzerovich (@round)Photo by Maxim Leyzerovich (@round)Photo by Maxim Leyzerovich (@round)Photo by Maxim Leyzerovich (@round)
Maxim Leyzerovich retweeted
The problem isn’t that design teams don’t value “outcomes over output” – the problem is that they don’t know how to define experience outcomes in the first place.
12
10
2
69
do the right thing ↰ ↳ do the thing right
12
10
2
60
11,131
the best designers i've met aren't just good at making things they're good at knowing what should be made in the first place
73
Maxim Leyzerovich retweeted
the world is a hackathon now
25
145
20
1,099
21,497
Maxim Leyzerovich retweeted
And if you're in NYC on December 11th, come celebrate with us at Index Space in Chinatown
We're excited to announce this year's Are.na Annual: Score. This year, the Annual is themed “score,” as in to secure an advantage; to gain points in a game; to keep record or account; a cut, notch, or groove; a group of twenty; a musical composition; a grudge.
3
6
1
91
3,975
Maxim Leyzerovich retweeted
member of janitorial staff
8
3
66
2,336
Maxim Leyzerovich retweeted
learned? helplessness??
1
9
712
Maxim Leyzerovich retweeted
design is a subset of applied foresight
1
2
27
1,144
Maxim Leyzerovich retweeted
design work is subsumed by futures work
2
2
11
861
Maxim Leyzerovich retweeted
Within a week, I've received more DMs and emails, and honestly, lovely praise and feedback. So, thank you all. I'm partnering with @ridd_design on a curated talented network called Decimals and @dannydecimals. More companies will want to implement "brand as software/infrastructure". The challenge is two-fold: 1) Leadership. They have to be bought in and create the environment and provide the resources for this brand/creative team to evolve. 2) The talent. From a creative director to the ICs that execute the work to engineers — and even product design and product management — if you want the Michelin-Star vending machine you have to build it like a product. @ridd_design has one of the top networks for product designers and design engineers, placing incredible talent and top companies. I'm partnering with him to provide the other side: brand design, creative directors, art directors, animators, and motion designers. If you're an individual, sign up here: app.decimals.co/paul If you're a company that wants access to the network, sign up here: app.decimals.co/design/hire?…
"people don't want tools, they want a Michelin star vending machine" @PaulJun_ breaks down the future of brand and creative teams in 110 seconds 👇
9
5
2
123
6,172
Maxim Leyzerovich retweeted
my new media company is just dinner parties
21
45
4
479
25,220
Maxim Leyzerovich retweeted
i will say that anthropic were the first of the labs to engage artists and creative technologists in new york, courtesy of rhizome. this was back from 2024. most ai companies still don’t bother.
4
4
1
93
4,567
Maxim Leyzerovich retweeted
We poked and stretched an LLM to learn about the dynamics of its thoughts and how it reasons over complex materials science problems.
The internal metaphysics of an LLM: An LLM represents physical relationships in a form we can both read and steer, and providing insights into post-training signals that help models develop deeper generalizations of physical principles. Across a large set of materials-science laws, the direction of a shift inside the model revealed whether increasing one quantity makes another rise, fall, or stay the same. That pattern persisted when we rearranged the equations and when we removed them and described the problem in words. By nudging the model’s internal activity, we could test which relationships actually helped produce its predictions. New work published in AI for Science (link and press release below) To measure a spring's stiffness, you pull on it and see how far it stretches. We “pulled” on three versions of Google’s Gemma language model by reversing a single physical change in a question. Then we measured the shift in its hidden state, the set of numbers the model uses as it works toward an answer. Two mathematical lenses helped us read those numbers. The direct logit lens asks which words the model would favor if it had to answer at that stage. A logit is a score for a possible next word or word fragment. The Jacobian lens accounts for how later stages typically reshape that signal. That is, a Jacobian tells us how sensitive each number downstream is to a small change in one upstream. When we described grooves forming along the boundaries of tiny crystals in steel, this lens ranked corrosion first, although the question never named it, revealing some of the inner thoughts of the LLM and how it reasons over physical principles. We also built a graph to see which problems the model treated as similar. Each dot represented a question, and we connected two dots when the patterns of numbers inside the model looked alike. Within a given mechanism, two questions could become neighbors simply because both described an increase. What does that increase mean for the material? More pores, or empty spaces, make a material less stiff. More stiff particles make it stiffer. To see whether the model captured that distinction, we asked each question twice, once with the input increasing and once with it decreasing. The direction of the shift inside the model revealed which way it expected the material to respond. Even without equations or step-by-step hints, both larger models captured the correct direction for all laws that predicted a change. Then we pushed from the inside: A metal is made of tiny crystals called grains; in the metals we tested, smaller grains mean more boundaries between crystals. Those boundaries make it harder for the metal to change shape permanently, so its strength goes up. We kept each question exactly as written and made a small adjustment to the model’s internal numbers, strengthening the signal associated with grain size. Across six distinct metals, the same adjustment made the model favor higher strength when the grains got smaller and lower strength when they grew. Because the effect reversed with the physical situation, we could tell that the grain-size signal helped the model decide which answer to give. The biggest opportunity I see is to turn these measurements into rewards for post-training. After its initial training, a model can learn from feedback. A reward is simply a score that reinforces a particular behavior. Imagine giving the model two versions of a problem, one in which the grains shrink and another in which they grow. We could give it credit for answering both correctly and additional credit when its internal response follows the physical law and helps produce both answers. Then we could change the metal, the numbers, or the wording and test whether the same reasoning holds. That makes the ability to carry a physical principle into a new situation something we can train for directly.
2
15
1
111
4,177
Maxim Leyzerovich retweeted
“Once the workday is over, the worker is suddenly redeemed from the total contempt toward them that is so clearly implied by every aspect of the organization and surveillance of production…”
1
3
1
24
4,713
Maxim Leyzerovich retweeted
“The spectacle is not a collection of images, but a social relation among people, mediated by images.” —Guy Debord
We got dudes watching TV in a video game while streaming to people who are watching them play a video game where they watch TV…. Maybe we do just root for the power grids to be attacked after all
5
90
5
646
14,616
Maxim Leyzerovich retweeted
This animation was inspired by the picture plane from the book ‘understanding comics’ from @scottmccloud . I wanted to see how a transition would look like between a minimal and more realistic portrait, and this is the result.
8
104
10
541
Maxim Leyzerovich retweeted
Really fun chat with @joelhooks here. He is always building mad things with agents. We mostly compared our levels of AI psychosis. They're high.
had the opportunity to catch up with my friend @Mappletons who's exploring the latent space of version control at @github maggie is wonderful, what a treat
3
4
1
123
5,277
Maxim Leyzerovich retweeted
can very much vibe with this! a couple of months ago made a playground for myself, which i called design-kata, borrowing the name from the once popular codekata exercises (tho this playground doesn't have much to do with exercises anymore lol) mainly to be able to spin up and try any idea super fast without too much friction @willdjthrill's prompt / agent rules are very cool addition and i wish i'd added something like this earlier, certainly would've saved me a lot of unnecessary back and forth
1
49
3,729
Maxim Leyzerovich retweeted
Who are the best designers of 2026? Among the nominees so far: @benjitaylor (SpaceX AI) @shadcn (Vercel) @evilrabbit_ (Vercel) @ryolu_ (formerly Cursor) @emilkowalski (AI for UI) Who’s missing? Last day to nominate + vote: cosign.co/awards/design
8
24
633
13,780