Hello, dear readers! 👋
In this issue, among other things:
Can AI pick up font pairs
How chasing the number of users can kill a product
The concept of a safe triangle for nested menus
Simple and functional modular grid calculator
Free atomic design system for SaaS products
Large selection of SaaS explainers
Selection of the best logos of the 1960s
Huge library of references for website and application design
Enjoy reading!
🗞 News and articles
Is AI better at picking and pairing fonts than you?
Matej Latin tried to figure out whether it is possible to select good font pairs and individual fonts using AI. He tested a couple of specialized AI tools, and also tried to pick up fonts using ChatGPT.
Unlike special generators, ChatGPT not only picked up working options, but also explained its choice. Nevertheless, Matej came to the conclusion that at the moment AI will not be able to replace the designer in this case.
Main thoughts from the article:
Pair generators often offer non-suitable options, while the ones that work do not differ in originality.
Generators take into account only basic font parameters, such as weight or tilt. This is not enough
ChatGPT selects fonts better and can give meaningful recommendations based on the context of font usage. Nevertheless, his proposals are also unoriginal
ChatGPT is able to select fonts based on suggestions in the form of adjectives. For example, he can choose a "kind" or "modern" font
ChatGPT can be used as a starting point for font selection. This may be useful for non-professionals
ChatGPT contains data only up to 2021 and does not know about fresh fonts
When selecting a font, the designer moves from basic parameters to more subtle ones that AI is not able to analyze. For example, the designer takes into account the language in which the text will be typed, the content of the text itself, or the originality of the combination of two fonts
The choice of font is based on visual and semantic aspects. Generators take into account only the visual part, and ChatGPT is more inclined to analyze meanings + relies on font combination statistics
The Tyranny of the Marginal User
Ivan Vendrov spoke about the "tyranny of the marginal user" — an effect that leads to a strong simplification and corruption of popular digital applications.
He writes that teams of large products focus on metrics like DAU and MAU — active users per day/ month, which creates incentives to attract more and more users. The company cares more about the "marginal", that is, the next user, than about the one who already has and loves the product.
The interest of each next new user in the product is lower and lower compared to the "oldies". A casual user is used to scrolling quickly, rather than delving into complex functionality and customizing the application for himself. As a result, the focus on such users leads to the fact that the mechanics inside many products are simplified and they become unsuitable for old and advanced users.
⚡️ Briefly
Better Context Menus With Safe Triangles. Kosta Alexsoglou spoke about the concept of a safe triangle, which solves the problem of accidentally closing nested menus, and also showed how it can be implemented using SVG and CSS.
New technologies
Google has released a big update to its AI chatbot Bard. What's new:
Added the ability to use information from other Google apps (similar to ChatGPT plugins), as well as work with private information from Gmail, Docs and Drive
Now the Bard can independently request data from Google Maps, YouTube, Hotels and Flights
The Bard's answers can be further checked for inaccuracies using the Google It v2 function. The system will additionally Google the answers and highlight inaccuracies
Now the dialogues with the Bard can be continued by the link. You can also continue the dialog started by another user. For example, it will be convenient to share promts this way
Added the ability to upload images and include them in promt, as well as ask questions about them
OpenAI announced DALL-E 3. The new version of the model will be integrated with ChatGPT, which will avoid painful industrial engineering. It will be enough to describe the request in ChatGPT in simple words, after which the neural network itself will select promts.
The developers paid great attention to working out the details and taught the model to better understand the context of the image described in the script. Also, the model was severely limited by filters of unwanted content. For example, she will refuse to generate images with famous people or create art in the style of specific artists.
On the neural network website, you can already see examples of images that it creates. In October 2023, the developers promise to open access for ChatGPT Plus and Chat Enterprise users.
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.