After publishing more than 70 YouTube videos on my channel, today I got my first content violation warning for sharing a review about Play 2.0, a video manager for YouTube that facilitates downloading YouTube content for free. Oopsy.🫥 I did the highly recommended training by Google.
If you are a serious YouTube video consumer, consider using Play 2.0. That is all.🤫1
I wonder if I could make my video available elsewhere…🤔 ↩︎
As a paying subscriber to HEY mail, I find it strange that the newsletter consumption feature has never been updated since its initial release1. A product is rarely perfect on its first try. For example: newsletters are presented in a semi-expanded form, but I wish I could display them as “newsletter titles only”. Also, there is no integration whatsoever with third-party read-later service to easily send a newsletter to Omnivore, for example2. Once expanded, it’s not obvious our to return to the list of newsletters sitting in the bin to read the next one.
Here is a quick recount of my “Ah ah moment” this morning, involving generative AI in a simple workflow. I’m still in awe about it.
If I stumble on a long and chatty YouTube video that might be interesting, I’ll use this workflow to quickly get the gist out of it. First, I use Downie (a Mac utility) to download it on my M2 MacBook Air. Then, I use the Whisper Transcription Mac app to transcribe the spoken discussion into text that I export in a PDF1. Next, I open ChatGPT, upload the PDF file and let ChatGPT ingest the content. Finally, I prompt ChatGPT with things like: what is the subject of this discussion? Are there any apps mentioned in the discussion? Who is talking? Is this a controversial subject? Generate a title for this discussion. Generate a summary in less than 300 words. Etc. It’s very effective. You can use the GPT I created for this2. Oh, and I guess you can use the same workflow for a podcast episode!
This workflow was also documented in “My Micro-Workflows Explained” mini website.
A recently published article on The Verge discusses POSSE and the Fediverse: “Publish (on your) Own Site, Syndicate Elsewhere.” This content strategy emphasizes the importance of owning the content you create by publishing it on your own platform, like a personal blog or website, and then syndicating or sharing that content on other platforms, such as social media or content aggregators.
The main idea behind POSSE is to ensure that creators maintain control over their content. By publishing first on their own platform, creators can establish a primary source for their work that remains under their control. They can then share or syndicate this content to other platforms to reach a wider audience, drive traffic back to their own site, and engage with communities on those platforms.
This strategy is particularly relevant in the digital age, where content creators often face the dilemma of reaching large audiences on popular platforms (like social media networks) while also wanting to maintain ownership and control over their work. POSSE offers a balanced approach, allowing creators to leverage larger platforms’ reach without sacrificing their own site’s autonomy.
I’m practicing POSSE myself; all my online setup is built around it. I depend on two publishing poles: Micro.blog and Ghost1. Some find this setup time-consuming and don’t want to be held responsible for replying or engaging on each branch (Mastodon, Bluesky, etc.). My take on this is yes, it might be time-consuming, but I like to engage on each platform because each brings a different type of community. I find it a bit frustrating to reply to someone who systematically shares content from his blog with Mastodon without any reply or acknowledgment. I understand that some posters are very popular and can’t reply to everyone. You can see if someone is replying from time to time. It’s a good idea to check before judging. The danger here is to act like bots if there is no engagement at all.
Micro.blog is responsible for the cross-posting magic. ↩︎
I don’t know if Apple is working on LLM stuff; they probably do, they probably do work on improving Siri, too, if such a thing is possible in the current incarnation of its fundamentals. But, judging by the rapidity of other companies introducing AI features mainly based on LLM models, I don’t expect it would be so hard for Apple to do the same with Siri. But only if Apple accepts to work with CharGPT-back end for a short-term solution. This could be a transitory path in my mind. Because Apple being Apple, they probably would want to put their twist on this: better privacy protection, for example. They like to control the whole stack. That’s perhaps why they are, apparently, investing massively in their one training infrastructure, which would be they accept the fact that on-device training is too limited, even with powerful Apple silicon. It could prove to be a long journey. I don’t expect too much for next year’s OSes. It will be interesting to see where Apple is headed with this AI thing next year.
Meanwhile, when I’m asking Siri queries today, I cannot help but feel the tech is antiquated compared to what we can do with Whisper and the like today.
Image: Dall-E.
I’ve been working with Apple Shortcuts editor recently1, on my iPhone and on my iPad, and boy, what a shocking experience! This editor is nearly unworkable, thanks to scrolling sluggishness, object ordering bugs, object variables becoming empty, constant freezes, keyboard masking input fields, changes not being saved and plain and simple editor crashes. 😤 What is happening with this, Apple?2 I can’t believe Viticci is able to endure this for all his work around shortcuts.
I’m working on something related to using AI in my creative work. I came up with this. AI is used through these front-end clients within my creative workflows:
- ChatGPT website
- Raycast AI Chat
- MidJourney
- Whisper
- Micro.blog podcast episode transcription
- Audio Hijack Transcription block
- Kagi Summarize Results, Summary Document, Ask Questions
- Craft Assistant
- Grammarly
And I expect this list to grow in the coming months.
Following the episode (#097) of Hemispheric Views, it’s my turn, I guess, with a little twist for the post title. 😜
I recently wrote about software is never really finished1. Here’s an example: Keysmith. I bought this utility in February 2021. I forgot to re-install it on my M2 MacBook Air. I just did. But under macOS Sonoma, it has some glitches which make it useless. The app hasn’t been updated for quite a while, same with the app website. Now what?2
I want to augment my creative capabilities1. I’m considering the best way to pay for Augmented Intelligence (AI) services2. I’m currently subscribing to MidJourney for imagery. ChatGPT with GPT-4 is not cheap at 20 $US a month, but I get Dall.E 3. Raycast enables GPT-4 for 8 $US on top of my 8 $US, totalling 16 $US monthly, but I don’t get Dall.E 3. I prefer MidJourney over Dall.E 3 as it seems to allow more diversity in requests. And there is the ChatGPT client to consider. Raycast is one of them. MacGPT is another one. Even Grammarly and Craft are ChatGPT clients3. I need to continue my analysis. Ideas? Suggestions?