Hello, dear readers! 👋
In this issue, among other things:
Promising properties for working with text in CSS
Seven rules for creating impressive apps
How to make an application using AI from a layout in Figma
Important skills for developing a career as a designer and lead
How to rethink your skills in the age of AI
40,000 original posters for old and new films
…and much more!
Enjoy reading!
🗞 News and articles
Why I’m excited about text-box-trim as a designer
Jason Bradberry wrote about promising CSS properties for working with text, such as text-box-trim. They allow you to align text along the borders of characters, rather than along the interlining, and are designed to simplify the work with typography on the web.
At the time of publication, the described properties are supported only in Chrome and Safari latest versions.
Pablo Sánchez’s 7 Rules for Designing the Unexpected
At the Figma Config 2024 conference, Pablo Sanchez talked about seven rules that will help you create an impressive application.
He explained how to bring a fresh perspective to the design and stay true to his vision, and as an example, he looked at a project he was working on himself, the Ableton Note music note app.
Seven tips:
Create something that inspires you. Great design is born out of personal beliefs, not out of dry objective analysis.
Don't watch the references during the brainstorming session.
When looking for ideas, don't limit yourself to the design plane. Rely on emotions, memories, scientific discoveries, or nature.
Try to create a future design that doesn't just solve a marketing problem, but creates a discussion.
Don't underestimate the users and oversimplify the product. Create a design that will encourage the user to discover and explore
Look at design as a remix of ideas, creating something new from existing concepts.
Don't stop at one idea. Develop as many concepts as possible at the same time and choose the best one.
⚡️ Briefly
Felix Lee shared a list of necessary skills for career development both as an individual specialist and as a leader. In a separate table, he compared the importance of skills for each path.
Redefine Your Design Skills to Prepare for AI. NNGroup provided five strategic tips on how to redefine your skills in the era of AI development.
Steve Sewell showed how to make a full-fledged application from a layout in Figma using Visual Copilot and Lovable AI services.
New technologies
The new Chinese DeepSeek-R1 model has made a lot of noise over the past week.
Details:
According to the company, training the model turned out to be 20 times cheaper than Western flagships, and the number of required GPUs was reduced from 100,000 to two thousand. Such bold statements aroused suspicion.
In math problems, code writing, and reasoning, it is comparable to o1 OpenAI and better than Sonnet.
There is censorship on sensitive topics for the Chinese government
The model code is open. It can be used for free in the browser and locally. API access costs significantly less: Compared to o1, it is 27 times more profitable.
The breakthrough was achieved thanks to a new approach based on Meta's Llama open source model.*
Against the background of the release of the model, Nvidia's capitalization fell by 17% in the first day, losing about $ 500 billion. Following it, shares of other American technology companies sank significantly.
Meta shares, on the contrary, have increased in value
Meta plans to use the approach of Chinese companies in its own projects in order to save resources through their research
Immediately after the hype, DeepSeek released Janus-Pro-7B, another open—source model, but this time for generating realistic images. It surpasses OpenAI, DALL-E and Stable Diffusion in GenEval and DPG-Bench tests
An AI assistant has been added to the DeepSeek application, which can process files, search for information on the Internet, help with planning, and much more. It is already available in iOS and Android apps.
Chinese developers from Alibaba have released Qwen2.5, new open—source models with web search, as well as image and video generation.
The Qwen2.5-1M model supports one million context tokens. For comparison, ChatGPT has about 200,000 tokens, while the acclaimed DeepSeek has 64,000.
Qwen2.5-Max turned out to be more powerful than DeepSeek and ChatGPT. She beat DeepSeek V3 in all reputable benchmarks such as Arena Hard, Live Beach, LiveCodeBench and GTA-Diamond.
Several more variations of the models are available on the website.
OpenAI has added the Operator function to ChatGPT. It allows you to create AI agents that can independently perform tasks in the browser or in partner applications. For example, to book tickets. Currently available only in the US.
Keep reading with a 7-day free trial
Subscribe to bezier.design to keep reading this post and get 7 days of free access to the full post archives.