Allison Nulty (ICML26🇰🇷)

924 posts

Allison Nulty (ICML26🇰🇷) banner
Allison Nulty (ICML26🇰🇷)

Allison Nulty (ICML26🇰🇷)

@ajnulty

Building Contra Labs

New York, NY Katılım Şubat 2017
408 Takip Edilen1K Takipçiler
Allison Nulty (ICML26🇰🇷)
ben@contraben

Can AI really do logos ? Seems to be a hot topic. So, we ran 155 blind tournaments with 10 professional brand designers instead to dig in. >@OpenAI's GPT Image 2 crushed the field. >@GeminiApp's Nano Banana Pro, @MicrosoftAI's MAI Image 2.5, and @AIatMeta's Meta Muse split the scraps. But still, winning a tournament and being production-ready are very different realities. 🧵

Polski
0
0
1
33
Sup*
Sup*@mostlyaboutai·
@ajnulty @contralabs_ai Lemme know if I can help and collaborate on this one. Been planning to work with CL for a long time now 😶‍🌫️
English
1
0
0
34
Allison Nulty (ICML26🇰🇷) retweetledi
ben
ben@contraben·
What are the biggest failure points for AI generated landing pages ? >8 working designers. >40 AI landing pages. >754 failure points found. @OpenAI’s Sol fumbles layout @AnthropicAI’s Fable fumbles the finish @xai’s Grok fumbles interaction @metaai’s Muse Spark has room to improve across the board Every AI model breaks a landing page in its own signature way. We mapped all 4. 🧵
Contra Labs@contralabs_ai

x.com/i/article/2078…

English
15
11
64
14.9K
Allison Nulty (ICML26🇰🇷) retweetledi
Matt
Matt@freelancer_matt·
If you want instant access to a US bank account (routing & account #) to accept transfers with no fees, join @contra.
English
5
4
37
1.9K
ben
ben@contraben·
Can AI really do logos ? Seems to be a hot topic. So, we ran 155 blind tournaments with 10 professional brand designers instead to dig in. >@OpenAI's GPT Image 2 crushed the field. >@GeminiApp's Nano Banana Pro, @MicrosoftAI's MAI Image 2.5, and @AIatMeta's Meta Muse split the scraps. But still, winning a tournament and being production-ready are very different realities. 🧵
Dylan Field@zoink

I have been thinking about whether to comment on this. Not clear if Gal is serious, rage baiting, etc. Whatever the case, it has spread enough in the design community that I want to share some thoughts. The psychological journey people go through with AI is quite fascinating to me. A new model launches, people think the world has changed, they sometimes have an existential crisis, then they play with the model, they understand its strengths and limitations and then they settle down. A few weeks later, the cycle repeats. On top of this, even before AI, designers have often shown insecurity and imposter syndrome. There are probably many reasons for this. First, before ~2010 design wasn't valued by the tech industry in the way it is today. Second, the people attracted to working in the field of design are often very open to new ideas and have high empathy. Third, there is no "one path" to working as a designer and designer backgrounds are often pretty random. Ironically, despite the insecurity + imposter syndrome so many designers feel, design is more important than ever. I truly believe this. And yes, I have an incentive to believe this. But just think about it... the logic couldn't be more clear. More design is entering the world, the attention economy is real and therefore creativity / design / point of view is how you will stand out. Your brand, marketing, product design, moments of delight and overall customer journey must be excellent. Some companies already get this and are fighting wild battles over design talent. Other companies are still figuring it out. Everyone will get there and it will be obvious in retrospect. This isn't a new trend with AI. It is a trend that we've seen over the last decade. Designers used to complain about not having a seat at the table. Now designers have a seat at the table. And many of the businesses I speak with are pulling from their design bench when looking for new leaders for their business... they know that design thinking and the design process is what they need to adopt everywhere to win. I'm not saying that every stakeholder gets it. But so many are trying to learn right now. Designers need to do more than create great work, they have to spend a lot more effort educating. Showing work can also trigger anxiety. Sometimes the best solution to a design challenge is the first thing you think of. And other times you have to explore for quite a long time to come up with something great. Inputs to a design process might include things that feel like traditional office work and are easy to point to... reading docs, talking with teammates, formal research, etc. Inputs might also include a walk in the park, an interesting dream you had the night before, a good song you listened to on the radio during your commute, a painting from the 1800's or all sorts of other cultural / emotional input. In summary, I've never been more confident in the role of design and impact design can have. I wish designers felt the same confidence. This is the moment to be more bold, to take more creative risk, to double down on the power of design. Everyone is on their own journey, and there are lots of fascinating ways to move through life, so if Gal is serious about "quitting design" then I wish him the best in his adventures ahead. But I hope if others follow they do it because there are other things they are so excited about spending time on vs fear of AI.

English
13
9
73
15.2K
Allison Nulty (ICML26🇰🇷) retweetledi
ben
ben@contraben·
comment if you want the new @contralabs_ai swag
ben tweet media
English
202
2
288
22K
Allison Nulty (ICML26🇰🇷) retweetledi
Kevin Frazier
Kevin Frazier@KevinTFrazier·
**cool study alert** @aspenkhopkins et al (including @contra researchers) introduce the "Human Creativity Benchmark." Their methodology is worth studying for application to other fields **cough, cough - the law** in which some degree of divergence is expected and even encouraged. As the authors note, "Current norms in AI evaluation do not account for the natural diversity and plurality present in creative labor, nor do they differentiate between the needs of creatives at different stages of their workflow. We believe this can result in homogenized outputs that are less useful for creative work, leading to undifferentiated creative voices. We hope to build a shared vocabulary for creative workers engaging with these tool." More research in this domain could result in some major unlocks. "Understanding which models excel at exploration versus execution, and where in the process agreement breaks down into personal preference, gives creatives using AI tooling a basis for selecting models that fit their needs."
Kevin Frazier tweet media
English
1
2
2
346
ben
ben@contraben·
Creativity is polarizing. Creative pros just crowned @reve 2.1 the best landscape model and the worst product model in the same tournament. We ran it blind against @BytePlusGlobal's Seedream 5.0 Pro, @MicrosoftAI's MAI Image 2.5, and @OpenAI's ChatGPT Images 2.0 on 10 real client briefs. It either won the table or got eliminated first. Almost nothing in between.
ben tweet media
Reve@reve

Reve 2.1 is here. The world’s best 4K image model just got better. Greater prompt understanding, world knowledge, and stronger foreign-text rendering.

English
6
10
69
8.4K
Allison Nulty (ICML26🇰🇷) retweetledi
Allison Nulty (ICML26🇰🇷) retweetledi
Aspen ✈️ ICML
Aspen ✈️ ICML@aspenkhopkins·
Very happy with how this poster turned out ;) come chat w/us 1:1 at 3:45pm✨
Allison Nulty (ICML26🇰🇷)@ajnulty

Excited to give a lightning talk on the Human Creativity Benchmark at the GenAICreativity workshop at #ICML2026. If you're interested in measuring creativity and generally improving creative outputs, I'd love to chat all things evals, post-training, etc. Stop by our poster session on Saturday or send a DM! @contralabs_ai @creativeai_ws

English
0
1
6
788
Allison Nulty (ICML26🇰🇷) retweetledi
ben
ben@contraben·
the dream: getting perfect edits from your favorite image model reality: missed details, choppy edits, forcing you to take it to photoshop etc Today, we are open sourcing our image editing trajectory data set on @huggingface The dataset includes: > 14 complete design trajectories > 294 annotated expert actions > First-person designer reasoning > Screen recordings > @Photoshop MCP tool calls > Keyboard shortcuts > Menu paths > Coordinate clicks > Screen recording > Designer reasoning > Structured action > Executable grounding (Photoshop MCP tool call, keyboard shortcut, menu path, or coordinate click) This dataset is useful for: > Computer-use agent SFT > Mid-training reasoning > Tool use & function calling > Human expert benchmarking If we want AI to edit like the pros, you need reasoning data from professional creatives
English
13
16
101
21.4K
Allison Nulty (ICML26🇰🇷) retweetledi
ben
ben@contraben·
Can an image be "Too Perfect"? Creatives judged @BytePlusGlobal's Seedream 5.0 Pro blind against: >@OpenAI's ChatGPT Images 2.0, >@GeminiApp's Nano Banana Pro, and >@bfl_ml's Flux 2. It won photorealism outright. But, it got dinged for feeling too perfect. Some insights 🧵
BytePlus@BytePlusGlobal

Dola Seedream 5.0 Pro API is now available on BytePlus. AI image generation is evolving beyond creating a single image. Edit with precision. Visualize complex information. Render realistic images and portraits. Create across languages. Create production-ready visual assets for enterprise workflows.

English
8
10
59
21.1K
Allison Nulty (ICML26🇰🇷)
Excited to give a lightning talk on the Human Creativity Benchmark at the GenAICreativity workshop at #ICML2026. If you're interested in measuring creativity and generally improving creative outputs, I'd love to chat all things evals, post-training, etc. Stop by our poster session on Saturday or send a DM! @contralabs_ai @creativeai_ws
Allison Nulty (ICML26🇰🇷) tweet media
English
0
11
30
2K
Allison Nulty (ICML26🇰🇷) retweetledi
ben
ben@contraben·
It's safe to say Meta is back in the game. We ran a blind "style-transfer" tournament with @AIatMeta's new Muse Image vs >@OpenAI's GPT Image 2 >@GeminiApp's Nano Banana Pro >@bfl_ml's FLUX.2. on 10 real world briefs, 55 tournaments, with every output ranked blind by professional working creatives. GPT won the most, but Muse placed top-two in 59%, more than any other model, already beating Nano Banana Pro 👀 (Image reference created with Muse Image btw)
Alexandr Wang@alexandr_wang

1/ releasing muse image today — the first image generation model from MSL. it's agentic: pairs with muse spark to reason through your prompt, search the web, and plan before it generates. people get what they meant on the first try. live now in the Meta AI app.

English
11
16
143
36.5K