Video yükleniyor...

Video Yüklenemedi

Ana Sayfaya Dön

How much more creative could models be by default? We asked frontier and open-weight models to design a landing page with “unique personality and taste” three separate times using the same business brief and images It almost looks like we got the same site

17,138 görüntüleme • 1 ay önce •via X (Twitter)

13 Yorum

Jeff Smith profil fotoğrafı
Jeff Smith1 ay önce

Interestingly, Haiku 5.5 shows some surprising variety across three takes on this travel booking site

Jeff Smith profil fotoğrafı
Jeff Smith1 ay önce

Text-to-image models show more visual variety on the same task but the results have their own flaws

Jeff Smith profil fotoğrafı
Jeff Smith1 ay önce

At Fieldwork, we’re working to measure design variety so we can improve it Designers already push well beyond these defaults with better prompting, skills like @impeccable_ai, and tools like @withgraphical Still, it’s striking how often different models arrive at a very similar solution All outputs and methodology: Text-to-image results:

andrew pignanelli profil fotoğrafı
andrew pignanelli1 ay önce

try this: ask it to map words to a random number generator. Then have each agent use the generator to make 50 random words and write those out before making the website. Then run the experiment. Surprisingly good hack that works for more diversity in design outputs.

Jeff Smith profil fotoğrafı
Jeff Smith1 ay önce

injecting entropy is a really reasonable solution here and I'd assume some of the variety we saw with haiku 5.5 may be the harness doing that by default

Eric Li profil fotoğrafı
Eric Li1 ay önce

I do think it's very prompt- and use-case-sensitive. I've been able to get a lot of variation with Opus 5.5 when I specifically ask for it. Rarely on the first shot, though.

Harold Apples profil fotoğrafı
Harold Apples1 ay önce

Why is it that every model chooses to use tiny subhead text above everything?

Jeff Smith profil fotoğrafı
Jeff Smith1 ay önce

1 year ago: purple gradients today: eyebrow headers 6 months from now: ?

Wikman profil fotoğrafı
Wikman1 ay önce

Super interesting. This is a space I am playing in at the moment. Trying out different sets of briefs, assets etc to learn where the models shift from default to more creative thinking.

Jeff Smith profil fotoğrafı
Jeff Smith1 ay önce

Feel free to play around with our prompts and assets: These are pretty shallow scenarios but still interesting to see nascent variety with recent models (Haiku 5.5, Astra)

Wikman profil fotoğrafı
Wikman1 ay önce

Will do.

Sani Ai Tech profil fotoğrafı
Sani Ai Tech1 ay önce

A fascinating look at how much creative diversity models still lack

BitAres profil fotoğrafı
BitAres1 ay önce

Three unique personalities, one shared wardrobe 😅

Benzer Videolar

learned a lot from this conversation with Simon Mo and Matt Bornstein. biggest takeaways for me: -there are a lot of reasons why we should like open-weight models. a lot of these arguments stop at handwavy things like "what if the labs stop releasing frontier models to the public" or "it's lower cost." but simon's position as lead maintainer of vLLM and CEO of Inferact give him authority to talk about some of the other, more interesting and concrete reasons to pay attention to open-weight models, namely that they allow end-users to calibrate latency / other performance metrics with way more customizability than what any of the frontier closed-source labs offer (and without the fear that your job might be met with a refusal at some random point where you're deep in a 2 hour job) -re: the above point...for this reason, a lot of US companies (inferact included!) choose to use open-weight models over their closed-source alternatives. this also isn't limited to internal workloads / research - on a recent a16z podcast the team at Decagon spoke about how something like 90% of their customer service ai agents run on open-weight models that they've fine-tuned. -we should really appreciate how many companies/teams came out researchers fascinated by the wave of very small open-weight models that were being distilled from e.g. gpt-3.5 and earlier models in 2022/2023 (prior to the release of chatGPT!). these small models motivated the development of pagedattention, which then led to vlmm/inferact (at other layers of the stack with similar origin stories, you can look at teams like openrouter or ollama). in other words, we have open-weight models to thank for a bunch of the orchestration infra we now rely on. i think yet another, indirect, way we can point to open-source/weight infra pushing the frontier forward. anyway, a lot more in this convo, it was a lot of fun!

Elena

12,922 görüntüleme • 3 ay önce