itsuka

You can see the glass material I was referring to in this video (timestamped): https://developer.apple.com/videos/play/wwdc2023/10076/?time...

(deleted)

Updated resources on June 21st:

- SDK on Xcode 15 beta 2: https://developer.apple.com/download/all/?q=xcode%2015

- Developer documentation - visionOS: https://developer.apple.com/documentation/visionos/

- Plan your visionOS app: https://developer.apple.com/visionos/planning/

- Human Interface Guideline - Designing for visionOS: https://developer.apple.com/design/human-interface-guideline...

- Apple Design Resources - visionOS: https://www.figma.com/community/file/1253443272911187215/

- App icon tool (Parallax Previewer): https://itunespartner.apple.com/assets/downloads/Parallax%20...

- visionOS Beta Release Notes (new features, known issues): https://developer.apple.com/documentation/visionos-release-n...

Demo: https://smart-placement-demo.pages.dev

I am an advocate for knowledge sharing and have previously contributed (a tiny amount) to the community mentioned above, Reactiflux. There, I was able to share my knowledge freely without fear of being penalized or judged through a voting system, or being heavily moderated as is the case with Wikipedia or StackOverflow. I also didn't have to worry about my contributions being eternally indexed on the internet. As a contributor, this is a feature (much less so for the lurker).

On that note, I recently had to request a deletion from Internet Archive because I shared content on my personal website that violates a ToS (it's a Slack archive that I have already anonymized). Unsurprisingly, my request went unanswered.

Are you referring to the text-gecko model? I don't think it's released to the public yet.

> If I remember correctly gecko is small enough to run on device.

Yes, I also remember reading that it's designed to be lightweight enough to run on a mobile device.

textembedding-gecko reached GA on June 7th, together with text-bison.

You can view all the models available to you on Model Garden.

For those who are curious (like me), Human Interface Guidelines and design templates for visionOS will be published later this month alongside the first visionOS developer seed.

(source: WWDC23 Slack)

If I remember correctly, Safari addressed many transform-related bugs two years ago, which resulted in me noticing fewer bugs (quite drastically). Around that time, Safari became my primary development browser, which may have also contributed to this observation.

Based on Web Platform Test [0], it looks like Chromium browsers are not performing any better in this area.

[0] https://wpt.fyi/results/css?label=master&label=experimental&...

But I feel the pain of having to carefully test 2D/3D transform and animation on all browsers across platforms (even Safari on iOS and macOS can have different behaviors).

How about prompt construction? I first heard about it from the maintainer of LangChain, and I also came across it in Azure docs: https://learn.microsoft.com/en-us/azure/cognitive-services/o...

I'm not sure what you're trying to build, but by the end of the second course, you should be able to create a customer service chatbot that is equivalent to what others have built. If you're interested in building/fine-tuning an LLM, that's totally beyond my knowledge.

I have taken Andrew Ng's courses on DeepLearning.AI and I highly recommend them if you are interested in quickly building LLM-based apps.

I suggest taking them in the following order:

- https://www.deeplearning.ai/short-courses/chatgpt-prompt-eng...

- https://www.deeplearning.ai/short-courses/building-systems-w...

- https://www.deeplearning.ai/short-courses/langchain-for-llm-...

Note: although I only have basic Python skills, I am still able to follow these courses

I have never built an agent before, nor am I knowledgeable about the latest studies in this field. So what I am saying below is likely to be nonsensical.

I was thinking that perhaps we have been working with abstractions that are too low-level. Instead of providing a set of tools such as API calls or text splitters, wouldn't it be more reliable to give agents templates or workflows of successful tasks, such as trimming videos or booking restaurants?

These templates would consist of a set of function calls, or a graph of connected components in low-code tools like LangFlow. I believe auto agents already use a similar concept where they cache successful tasks for future reuse. The idea is to populate these caches with the most common use cases, and use retrieval if they become too large, so that we don't experience cache-miss most of the time and work with lower-level abstractions (tools) as the baseline. Templates, like prompts, should be portable (e.g. JSON) to avoid the need for everyone to reinvent the wheel. While this solution may not be as impressive as a full autonomous agent and may not work for a generalized case, it should produce a more predictable outcome, I think.

I maintain a list of OpenAI-powered OSS UIs: https://github.com/itsuka-dev/awesome-chatgpt-ui

This one seems to have gpt-4-0314 model: https://chatgpt.nextweb.fun

For gradient tool, I can recommend https://gradient.style by Adam Argyle

Not only is it easy to use (IMO), but it's also more powerful than what design tools offered today, with new color spaces being one of my favorite features.

Intro: https://twitter.com/argyleink/status/1649124742463623169

Thank you for taking the time to explain things so clearly to me. Considering the reuse, having values in a separate page makes sense now.

To celebrate React 10th anniversary, Next.js interviewed Andrew and Sebastian from React core team: https://www.youtube.com/watch?v=g5BGoLyGjY0

You're right. I am not as familiar with MDN's organization and CSS as you are (I only recently discovered that CSS is a typed language). However, I find it confusing that some pages have their own page for values, while others don't. As a reference page, I expect the valid syntax for all values to be provided. But I agree that specific examples can be linked to other resources.

I wish they'd update their docs with better examples before creating content in other areas, such as tutorials and blogs.

As an example, when comparing these two pages:

- Blog: https://developer.mozilla.org/en-US/blog/css-color-module-le...

- Reference: https://developer.mozilla.org/en-US/docs/Web/CSS/color

It is apparent that the reference page lacks any mention of features from level 4 such as oklch, display-p3, and color function. Also, a search for "display-p3" in the search bar returns no result on that page.

I personally prefer sizes that are multiple of 8 because it works well with grid-based layouts and popular CSS libraries and design tools like Tailwind and Figma also adopt this system, although not strictly. However, I agree with your general sentiment that the exact number does not matter much. It will matter less even more when Container Queries become a common practice.

I'm currently going through https://practical-accessibility.today course by Sara Soueidan. She is really knowledgeable and, most importantly IMO, she practices what she teaches. Her Twitter feed also has some nice tips on accessibility.

Microsoft Build session: https://www.youtube.com/watch?v=VqJSxoQ6BXM

I've been using Firefox as my main browser to support its decreasing market share, alongside Safari for development and Ungoogled Chromium when things are unexpectedly broken in Firefox (due to strict privacy settings most of the time). I'm considering dropping Firefox because I don't want to deal with updating about:config anymore to remove ads, telemetry, etc. The other two browsers just work for me.

Edit: this is not completely true - I enabled Private Relay (Safari only) and installed uMatrix and uBlock Origin on Chromium, but that's about it.

I tried using Linear before switching to Obsidian Kanban. I found Linear's UX and site structure to be overly complex for my personal needs. Despite following the onboarding process, I found myself clicking too many times to figure out where things are (I acknowledge that this may have been an issue on my side). As a result, I switched to Kanban because I was already using it to store documentation for my projects. For me, it meets the spec as a project management tool and is an improvement over a text-only kanban. I can definitely recommend it.

Demo: https://twitter.com/rupert_parry/status/1653780093712633859

Does anyone know if this affects open source and non-commercial projects as well? I can name over 50 projects that have "GPT" in them.

If I'm not mistaken, @ChatGPTBot on Twitter received some kind of sponsorship from OpenAPI. However, the project on GitHub makes it clear that it is not affiliated with OpenAI. So, some projects may be approved to use "GPT" on a case-by-case basis?

You're welcome :)

> someone actually did a tree interface for ChatGPT!

In case anyone is interested, UCSD Creativity Lab has a couple DAG UIs that offer richer interactions and visualizations:

- Sensecape: https://twitter.com/HaijunXia/status/1646919380704559104

- Graphologue: https://twitter.com/HaijunXia/status/1646917869115166720

According to the discussion on the thread, they are in the process of open sourcing these two projects.

I maintain a list of open source UIs built on top of OpenAI API to explore unique features and integrations that others have developed: https://github.com/itsuka-dev/awesome-chatgpt-ui. Web version, which has more information: https://atorie.xyz/crafts/plurality

In addition to the usual speech synthesis/recognition and embedding/vector search features, there are also:

- Node layout

- Multiple LLMs and parallel output

- 3D avatar

- Selection + custom context menu (for extensions)

- Native app integration such as Siri and Calendar (for Shortcut in Apple ecosystem)

- etc. (I'm sure there are many other interesting UIs and features that I missed.)

This blog post explains why developing accessible components (that work well across devices) is not an easy task: https://react-spectrum.adobe.com/blog/building-a-button-part...

There are so many UX pitfalls that I prefer to delegate the task of fixing them to a library, while still retaining control over how I want to style them.

Someone said that using a chat-only interface is like using CLI tools. It's even worse, I would add, because there is no autocomplete or man command to help you out. Most of us here probably get it, but my dad had a hard time getting a good answer from GPT when he first tried it, even with GPT-4 model.