Anyone else notice that a fair amount of Justin’s original video descriptions have been replaced with LLM summaries? Here’s one example, the Blackbird song page as of 2022 from Internet Archive:
To me, this new LLM-generated page copy sounds as vapid and inauthentic as a lot of other LLM output. I don’t have a strong opinion about AI in general, it’s just that AI copy sounds awful and Justin’s was wonderfully real. I don’t know how many other pages were replaced by LLM output. Here’s the new LLM page:
How to play Blackbird by The Beatles on the acoustic guitar
Want to learn Blackbird by The Beatles on guitar? You’re in the right place! In this detailed lesson, we learn how to master Blackbird using an authentic and surprisingly simple technique that makes this legendary song much more accessible, even for beginner and intermediate guitarists.
We start by learning Paul McCartney’s unique picking style, in which the thumb and first finger work together to create that beautiful, flowing sound. You’ll see how a small adjustment can make the entire song feel more natural.
Here’s what we learn together:
Essential Fingerpicking Pattern: Thumb and first-finger strumming to create the song’s iconic rhythm.
Fretting Hand Techniques: How to move smoothly through slides, chromatic climbs, and chord transitions.
How to Stay in Rhythm: Why Keeping Groove Matters More Than Playing Every Chord Perfectly.
Bridge and Chorus Walkthrough: Step-by-step practice of the climb-up and climb-down sections.
Bonus Ending (Optional): An optional instrumental ending you can add to make the song your own!
Whether you want to sing along or play Blackbird as a beautiful solo guitar piece, this lesson gives us the skills, confidence, and musicality to bring it to life.
Thinking about this more, I think it stands out so much because Justin’s teaching style and voice is why we’re all here. Even if the new LLM-generated text was great (and it’s not!), it’s not him, so it seems extra strange to replace the thing that we’re all here for.
I assume that JG staff had reasons to do this, like maybe it ranks better in search engines or, sigh, it makes LLMs refer/link to the page more often. It just sucks for those of us who appreciate Justin’s style.
I suspect they’ve done it in part for consistency from page to page and also because it’s faster to generate for each page. This page is big enough that I’m sure Justin doesn’t really have the time to be so directly involved with every aspect of the website anymore.
I agree that I prefer more “personal” sounding copy.
That could be. To me, the inconsistency is a major feature, not a bug. Justin adapts because there’s not a formulaic way to explain everything. He points out the things that he thinks are noteworthy and will help, and that will vary by lesson.
Justin’s approach is also what differentiates this site from anything else, including sites that have gone “full slop” or asking an LLM myself.
If it’s no longer possible for him to write new pages, maybe leave existing pages as-is and have someone familiar with his style write new pages. Ideally, a real person who has watched the video, plays guitar, and learned from Justin.
Videodescripotions are only meta data, descriptive and with the aim to become as “findable” as possible for those who search. The content library is MASSIVE so what’s wrong with some AI generate meta data to label a bunch of youtube descriptions? Doing that all manually is less algoritm efficient and has a higher personnel cost. a cost now rather invested in helping out people in 1-to-1 contact like @FannyJustinGuitar does a lot etc.
The descriptions might be AI but that’s him IN the video’s and it will always be him.**
His lesson, his style, his accents, his priorities…
I’m a bit allergic to lots of words without meaning. Whereas the original description contained some to the point information about playing the song (limited though it was), the AI generated text is just generic, and could almost be copy pasted for any other song (so I’m a bit doubtful about the metadata thing). Reading the original, I feel like I have picked up some information. Reading the AI generated text, I feel more like I wasted my time.
search optimization was killing “personal” language and inflections before LLMs and I didn’t like that, either. and honestly, with search engines (ahemgoogleahem) changing the way their search categorization works in an attempt to cut off the link farm slop (but always being a bit behind it), frankly I don’t think it’s worth it.
search optimization ruined recipe websites and forced everybody to write a stupid (most likely made up) novel before telling you how to make the stupid food. search optimization (especially with LLMs now) has ruined your ability to search for real product reviews. and it’s ruining DIY content. I hate starting and stopping youtube videos as I DIY a repair…I’d much rather use a website with text and photos and maybe short clips after a written description. but youtube videos are mostly the only place to find the DIY repair info I’m after anymore. it’s just not a good thing for web content intended to be consumed by people (isn’t this what the entire JG website is supposed to be for?). It might be useful for website owners to drive traffic. But what’s that really worth now that the Dead Internet has come to fruition?
To me, LLM generated YouTube metadata for search engines is an efficient use of time. But I think the OP is talking about the detailed lesson notes on the JG website.
Perhaps it’s just me getting used to the “old” way but I do like the older description on this lesson more than the new one.
In case the team wasn’t aware, you can train the LLM to recognize Justin’s writing style and then have it generate new pages in the same style. It takes a fair bit of iteration, isn’t perfect, but this might be a good compromise between efficiency and writing style.
I agree with @shuffle27 .
Cutting costs by generating content via AI will drive away people (at least some people), while it may attract others who will not have the same user fidelity and will typically bounce on different resources, as the AI search engine guides them.
AI Search optimisation has the ultimate goal of making AI Search the delivery channel, so that users don’t have to browse the original website to get their information, but rather have it summarised by the AI agent on the search experience, thus generating more revenue for the search providers. Websites will, therefore, no longer be optimised for humans, but for machines, mainly. Whereas with traditional SEO this was behind the scenes because it relied mainly on metadata, with GEO (Generative Engine Optimisation) it has to do with content directly.
Most digital content will be generated by AI.
The ramifications of that are many and mostly dark, but I don’t want to delve into that.
I strongly hope websites like JG don’t go down this path.
I never said the automated way is better
So yes, I agree with the quality, the personality, etc.
The team makes choices on priorities. JG is not a multimillion dollar corporation and when running a business, you have to make choices. I’d rather see manually crafted descriptions myself but I understand not all of it can be as the team would really want. It’s like that in nearly every company. I can’t go into any further though, since I do no own nor do I partake in any decision making processes.
The comments are duly noted btw and yes we mods pass on things like these
I am always disappointed when I see anyone use generative AI, especially people in creative or creative-adjacent fields and industries. I understand that it might be an efficient way to optimize for other machines to read and find content (I work in digital—I’m not clueless about why companies are doing this). But content written by machines for machines feels lifeless. The content quoted in the original post may be great for Google or ChatGPT but as a human reading it, it says a whole lot of nothing. It’s not even informative, especially compared to the thoughtful descriptions that used to be available. (An “authentic” technique? As opposed to all the fake techniques out there? Sigh. My language-loving, English major heart is sad.)
There’s a reason that a lot of people were adding “reddit” to their Google search terms. The other results were AI slop, not written for people, and not useful as a result. Search engine visibility, maybe, but valuable content? Not so much.
That’s setting aside the very concerning environmental (as well as social) costs of relying on it. I’m discomfited by how normalized it is all becoming, and that smaller business feel like they have to use to keep up even though they may have had humans producing a high calibre of work for years before generative AI was even an option for them. I’d prefer to support businesses that employ real humans to do these kinds of things. That would be “authentic.”
Thumbs up for the description in Blues Immersion, I often read them to quickly revisit the content instead of watching the 20 minutes video again. It’s really well made.
Hopefully, paid premium courses will continue to have human description like Rock Immersion. As the description has a lot more use than metadata/search engine for the courses. Same for grade 1,2,3 etc.
Paid courses are behind the paywall, so not searchable from public domain and hence, probably less an aim for AI optimisation. Hopefully this is the case.
I agree with prevailing thoughts on AI here. Even when accurate, the generated text is completely without character. I hate to see that in written communication (i.e. websites/emails) more and more companies and individuals are trading off their personality and uniqueness for the same style of machine generated content.
I can understand this path for new lessons/videos, even though I don’t like it.
But I can’t see the rationale for replacing existing descriptions with new ones that are much less informative.
As someone who has used AI quite a lot for different purposes I can honestly say that the use of it for anything truly artistic is something that I don’t like to see. It’s something I see often on YouTube videos about some of my favourite musicians, often the facts given are inaccurate and emotionless - which is to be expected, but it’s not acceptable if the perpetrator is making money by doing it! They’ve done a huge shortcut by giving minimal information and not thoroughly checked it before making it live.
The least thing that needs doing is making sure it’s completely factual and accurate plus check that the presentation is something that they would watch without any hesitation.
If that’s what has been done then I would support the presentation, if not I would avoid it like the plague!
We have had, and still have, several general AI topics.
All have split opinion and erred towards controversial.
This topic is going to be most useful and productive- in terms of JG Community feeback to Justin and team - if the conversation remains tightly focused on the topic initiated by the OP in post #1.
Please do not enter into a general and wider AI debate here.
With Richard’s comment in mind, here’s my opinion (FWIW) …
Justin seems to be using AI to summarise his own work and as such I see no issue in that whatsoever. Would I prefer he spends an hour or two editing and hand writing a description from scratch? Yes, but that would give him less time for other more creative and helpful tasks.
