I was at Coursera for years and pitched this exact thing multiple times internally! So excited to finally see it being built. Congratulations on the launch!
This concept is really cool and solves big challenges around content creation. Obviously, it adds new challenges around pedagogy, licensing, and ads. The last part is a big no no for blue chip edtech platforms.
Thanks! We'd love to chat if you have any ideas or want to share your experience, we think Coursera is great. And yea totally understand the blue chip edtech thing, we're pretty excited to be a startup in this space.
Really cool idea! Some improvements I'd recommend with the ultimate goal being "getting users to learn the subject at hand".
1) Section Lectures on the left side need to be cleaned up, instead of just a numbered list. Seeing 30+ lectures off rip is a bit daunting, especially with no labeling, sectioning, etc. I'd imagine feeding a model a list of all the lecture titles, then having it structured should work?
2) You're doing too much on the bottom section.
You need to incorporate all those tabs into the single Ai tutor, which can run whatever tools required (maybe notes/discussion can be a small additional indication). No one's going to be using the Flashcards section, and it's calling probably the same LLM as the AI tutor, so might as well combine them.
For the quiz, maybe when the video ends or the user wants to continue, the Ai Tutor goes into "quiz mode" forcing the user to attempt or pass the quiz (depending on the settings?).
Think of this like Cursor but for Education. Cursors powerful agent can handle/do so much, you're not using 3-4 different fields.
Oh and have it on the right side instead of transcript, so it's right there in users faces instead of having to scroll down.
Great points. Definitely will improve the section lectures on the left, some formatting stuff to think about with the bottom section/transcript as well.
Not completely sure about the AI tutor points though. I don't think the standard AI chat interface is the ideal form factor for the average person trying to learn something, and there's value in having pre-generated content that users can see instead of having to actively go to the tutor.
Also, a lot of people do like using flashcards specifically to learn! Granted, our current implementation is pretty barebones so it's not super useful yet. And definitely agree that things can be cleaned up quite a bit.
You're compensating the youtube video creators for this, how? Are you asking their permission? Or is this similar to how all the AI scammers operate and just extracting value from other peoples work without compensation? Just curious if this will help the people who are actually doing the source work or if it's just another way to take other people's labor and make money off it without anything in return for them.
And no, I don't consider "they get extra views" a valid answer. Especially if you expect to make the windfall you want to make off their labor.
We do a revenue share with the creators for any content we monetize. And if they don't want their content on our platform, we will take it down.
> extracting value from other peoples work without compensation
You mean like how all art is created? Nothing is original... everything we create is a sum of past inspirations, human or not.
This is a fun concept, and I love the name!
I’m curious why you didn’t use multiple choice for the exercises? I feel like those would be easier than typing out full answers and be closer to MOOC style homework. Maybe have a longer written question at the end of a section.
The exercises work pretty well, I like the highlighting red wrong vs. green right. It does feel a bit like the MOOC-style discussions. The tutor doesn’t just tell you the answers which is cool, but something about talking with the tutor feels a bit flat. And the flashcards weren’t very helpful for the course I picked.
I could see myself doing some courses like this with some more gamification. Being able to filter by course provider (Ycombinator, or MIT) would be cool too.
Thanks! We do have multiple choice questions now (agreed) but some of the older courses were generated when there were only short answer.
Anything specific we could improve about talking to the tutor? Definitely will add some of those features and gamify better.
For anyone else interested in Bloom's 2-sigma, here's the original paper (1984): https://web.mit.edu/5.95/readings/bloom-two-sigma.pdf
Blows my mind that 1:1 tutoring dwarfs the impact of other factors such as socioeconomic status, reinforcement, assigned homework, classroom morale, etc (at least according to the researchers).
Does anyone know if this thesis has been replicated? Or if these results hold in modern times (original study was 40 years ago)?
The article states that Anaina and Burke separately conducted their tests, but social robots [1] have been shown to be effective in individual tutoring. Human tutoring is not always better than a well-designed computer program [2]. There have been issues with how studies interpret their effect on group size / scalability [3].
[1] https://www.science.org/doi/full/10.1126/scirobotics.aat5954 [2] https://www.tandfonline.com/doi/abs/10.1080/00461520.2011.61... [3] https://journals.sagepub.com/doi/abs/10.3102/0013189X2091279...
In my experience with 1:1 before and during my Masters in Education, nothing could compare to the learning and growth my students had from that highly personalized (and personal) instruction.
It's super common for students to not understand material or express that they don't, but it's just not possible to drill into each specific student's particular knowledge or skill level in the classroom environment.
On the social and behavioral side, many students who struggle in a classroom environment transform into model students when taught with both the care and privacy of 1:1.
For me, I feel it's a combination of hyper-personalized instruction plus compassion in a relationship where it feels safer for the student to accept the value of improving at something without social pressure or embarrassment.
Would be nice ie to see this product with focus on elementary school age content.
I am very interested in this, and I have personally built manual workflows to do Youtube video -> rip audio->transcript->llm context.
For example, taking a video about building garden retaining walls and generating detailed system prompts for Q&A with the expert in the video.
I reference ~home improvement or tool videos and often comments contain points of wisdom or even corrections of mistakes (errata) on videos that are otherwise good. For example, setting up a hand plane and ways to mark a board you're working on.
Do you use video comments in your context? I've (manually) scraped content on educational videos and built prompting to assess signal and incorporate what are likely important errata in LLM context.
> video/resource —> transcript/text —>
For this step in your pipeline, are you multi-modal? I mean, are you using the LLM to interpret what is shown in the video itself? How is that content used?
Do you have any sense for allowing people to generate educational content off arbitrary videos?
For now we only use the YouTube transcript because for most educational content we've found it does about as well for lower cost.
We may make that an option though, since we also offer other resource types (pdf, slides, docs) -> course.
To your last question, what do you mean by arbitrary? If the video is not educational at all, then the generated course will likely not be good. If the video is pure entertainment then probably not a good use case.
Can you extend this into language learning content on YT? I think that would also have amazing utility. As a biologist, so happy to see Crime Pays but Botany doesn't on here. Thanks for the awesome tool. I will be using it.
Yes! I did something similar with daily exercises at https://app.fluentsubs.com/exercises/daily
This is really cool!
Prof. Steve Brunton's YT channel is a treasure trove of material for you folks, with course-like playlists for controls, data-driven engineering, and dynamical systems: https://www.youtube.com/@Eigensteve/playlists
He should be a featured creator, much like 3b1b is for math!
We'll reach out and hopefully add some courses! Thanks.
Just wanna say that this is one of those magical ideas that I'd never personally think of, but when I see it like this, it makes perfect sense! So cool.
Your app is simply awesome. I've playing with it for a bit and I can see how this approach would make traditional homework much more engaging. Also, the LLM doesn't care about the language used in the responses, so my understanding is that localizing the content in different languages would be very straightforward.
This is, I think, particularly important for kids in most parts of the world as a majority of the internet pedagogical content is English-based. Or for people like me, that struggle with that language when talking about topics outside the tech industry.
Congrats for your project, I'm sure we share the same positive view about the future of learning :)
Not a bad concept. I already do something similar
I already use LLMs to quiz me on books that I'm reading. The current book I've been studying is "Computer Networking: A Top-Down Approach (8th edition)." Obviously since it is a text book, there are homework problems and other discussion points. However; it's also fun to ask grok/chatGPT to quiz me on previous chapters and correct my answers.
There's plenty of conference talks posted on YouTube that I've watched over the years. Next talk I watch I will probably test out being quizzed by LLMs.
You can feed a YouTube link into some LLMs and get it to quiz you as well. I use this technique as well!
Poker, specifically Texas No Limit Hold'em, is widely taught on Youtube.
Here are some of the very best in the category, it would be really cool if you partnered with any of these.
https://www.youtube.com/@hungryhorsepoker
Poker is interesting. I think these videos do work in our current course generation process. However, I do think some subjects like poker need custom tooling around the course to really make the learning experience great. For example, access to solvers or actually playing a hand on a table is a part of the course experience as well. Chess is another one that falls in this special bucket imo. Some of this tooling is on the roadmap!
Thanks, we'll reach out. We have a poker course from MIT (https://miyagilabs.ai/course/mit15s50) but yea these seem more practical & engaging.
I think this is a great idea. I’ve learned so much on YouTube but it’s always been in small chunks and very task oriented. I imagine there’s a lot of content Which covers broad topics that I don’t come across.
Something I’ve been doing more and more lately is asking chatgpt to create a detailed description of a topic which can be read aloud for whatever duration I plan on driving. This works exceptionally well - even for short 5 minute drives.
I wonder if the same can be done for video-based content. Sometimes I’m short on time but still want to learn something.
congrats on the launch!
i opened a random lecture (MIT 14.01 Lecture 19: International Trade: Welfare and Policy)
a- i tried an exercise without watching the video, and then clicked on 'watch' next to the question, i expected it would take me to relevant portion of the video instead it played it from beginning.
b- i used the ai tutor and asked 2 random questions, instead of giving me an answer it prodded to provide more detail. this might be an intentional choice to ensure learning, however i personally prefer chatgpt's style where it gives me some information and then provides more angles to pursue.
i also asked the tutor about batman, here's the response for reference:
tell me about batman That's a good start, but your question is very broad. When you say "tell me about Batman," are you looking for information about the character's origin, his role in comics, his significance in popular culture, or something else? Try to narrow down your question a bit.
For example, are you interested in: - Batman's backstory and how he became a superhero? - The different versions of Batman in comics, movies, and TV?
Thanks! Yea the watch button should do that, we added 14.01 before that feature but will fix asap.
And yea probably thinking to do a guiding mode (as you said, often better to have the tutor guide you through step-by-step instead of giving you the answer right away) and a straightforward answer mode for the tutor. Or if you have any other ideas.
How do you validate you’re not generating garbage, and thus teaching people nonsense?
For official courses, we go over the generated course with the creator to vet the content. Generally they're pretty impressed but have a few things they'd like to change/add before publishing.
For self-created courses, it's generally been quite accurate and we're playing around with some eval metrics to make it as good as possible, but it's definitely a concern.
Seems to actually work! Thanks for sharing, I'll be checking it out.
Hmmm, I can't seem to log back in, using a non-google email.
Recently I've gotten a lot of benefit from much of Alex Hormozi's thoughts. Yesterday I grabbed a 4-hour video, ran it through a free transcript generator, dropped it in Claude, and asked for an outline.
Unsurprisingly, Claude struggled to provide a complete outline. Like, it did a lot! but kept leaving parts out. I was able to prompt Claude to fill in more, good enough for me.
Anyway I dropped it into Miyagi Labs, waited 30 minutes while it said the course was creating. I kept it open in a tab and kept an eye on it. Eventually I tried to open a new tab into Miyagi (maybe I'd see it under My Courses?) but I was no longer logged in, and can't log in.
Also only the top right login button gives the option to use non-gmail.
Sorry if I broke something! I love the idea!
Congratulations on the launch guys! Was interested in such a product for a long time. I think adding social collaboration would be a game changer for a tool like this. Imagine people being able to start their own cohorts to learn and keeping each other accountable. Looking forward!!
I work in edtech and one of my teams is content creation, so pretty excited about this space but also very aware of the challenges and massive amounts of hype and over promise / under deliver. To assess I tried to generate a short (< 10m), one-video course from a YT video I've previously watched on a topic I'm an "expert" - after an hour all I see is the embedded video, the transcript and "generating content" dialog.
UPDATE: " This course failed to generate. Please try again or contact us."
I really like a lot of the components of your idea, but the execution is underwhelming. Right now it feels like you're providing middling tools for too many components without nailing any of them. Alternatively I could watch the YT video at all ready has a transcript, take notes in any tool, and ask questions to any LLM; the piece missing is context, so that's where it feels like you should focus.
Re: assessments; it feels like you're being distracted here; I'm not convinced that's how your natural target market learns in this modality. We generate quizes in our product, but it's typically used in the "internal compliance" segment - think mandatory training like food safety for food preparers - not the external (typically adult) self-improvement market (which is huge!). If you're going to do asessments you need a lot of non-AI boilerplate around tracking, validation and certification/credentials. My two cents: quizes in your app are a cool demo feature with little real value.
Sorry we're running into some rate limits with course generation but will be fixed soon. Valid points—will respond in a bit.
I’ve got 20 years of experience teaching mathematics at a community college. MyMathLab and it’s various clones are all terrible. My brief tinkering with this product has impressed me. Certainly it underdelivers at this point but they are on the right track as far as interface and usability. Don’t know if they can pull this off but being different than existing ed tech players in mathematics is a good thing since Pearson, etc. all suck.
Reviewing a service that didn’t work is like reviewing a product that never arrived in the mail.
Not sure you can draw many conclusions on an experience you didn’t have. My own two cents
Nice work! How do you verify correctness of the generated exercises and explanations? To me this looks the biggest risk in becoming a user: what if my _teacher_ teaches me subtle nonsense that I cannot easily detect since I'm learning and unfamiliar with the material (even if it's only in the 1-2% of cases)? Human teachers make mistakes too, but an LLM cannot _understand_ that it made one... So.. how do you solve for this issue of trust?
they dont verify, they just present a LLM app and the user suffers if the information is not correct. however most of the time it is correct but sometimes it is not. one way to verify correctness is to ask a bigger model like OpenAI o3
The flashcards app is not great. It can improve.
More serious comment.
It seems like there is less space (in our mind, and time) for yet another app. If you have this, vs youtube for course, people will still flock to youtube. Why don't you add the value that you are giving here by directly integrating or enhancing youtube. A chrome plugin or a youtube browser?
I really like the motivation behind this product. Learning is hard.
Does it work on YouTube videos that have transcripts disabled?
No, it won't work for those.
Congrats on the launch! I personally am not sure I see the value in paying for a course, when (I assume) I can do this on my own by feeding a video link or file to an LLM and asking it to generate a course. I'm sure your courses are probably better, though, and there are probably a lot of people who don't want to go through those extra steps. Good luck!
Instead of direct trivia from the content, it would be helpful to have an exercise (with evaluation) that applies the content learnt - a small artifact production with real-world practical use.
Imagine you would need, another ai pipeline that poses as the consumer and applier of the knowledge, instead of a direct processor ai of content information as it currently seems.
Neat idea! Do you do anything with the video itself? Understand the visual content or extract details from slides?
Users can upload slides (ie. docx or pptx) and create a course from them - give it a try! For videos, we don't currently process any frames from the video just the transcript, but this is on the roadmap.
Do you have any revenue sharing program with the content creators? Or are you just poaching them?
Lol yea not just poaching, we do revenue sharing (signed deals with a bunch of top creators). They get the majority of all revenue from courses.
For instance we worked directly with Crime Pays but Botany Doesn't & Faculty of Khan etc. to get official courses that they also had input in, and 3Blue1Brown is on board with us having his content on our site.
They probably haven't decided yet
Are the people that create the content okay with this?
Yes. Any content that we monetize we are revenue sharing with the creator. We already have more than 5 partnerships with creators.
Who cares?
hmm... i have mixed opinions on this but in the caso of MIT for example, it can be good like the online course want to put on your plataform, the only problem is LLMs are not correct in multiple cases mainly when it's too deep into the subject, but when curated it can wprk
And copyright problem, i think people shouldput on the plataform instead of web-scraping like
Amazing approach! Is learning a second language too different from the types of courses Miyagi was designed for, or do you see a potential for that category?
I was actually thinking about building this because I watch a lot of YT videos in other languages (best way to do travel research is to search the destination using the local name and getting local videos).
Thanks! Definitely some potential, we actually built a language learning tool for a few days early on (but decided that it was too crowded of a space to start in).
Learning languages seems a bit different in that there's more focus on repetition compared to comprehension questions, but there are certain topics (like grammar concepts) that could work well in our current structure. Also there are some really popular YouTube channels for learning any language, so we definitely see a potential to augment those videos to more accurately & effectively learn.
Shout out to Crime Pays but Botany Doesn't :)
Yea he's awesome! We just launched another official course with him yesterday haha (https://miyagilabs.ai/course/taxonomy-and-plant-id)
This is so useful as so many of the best things to learn in the world aren't locked up in universities!
I hope you view a YouTube ad on that video every time the course is opened otherwise it’s a self defeating system.
How worried are you about platform risk?
I don’t think they’re attached to YouTube too rigidly. (Well, I hope at least.) In theory this should work with any platform that provides subtitles. But I think if YouTube falls, or blocks their API access, they would just start hosting the videos themselves.
How censored is the "AI Tutor"? can a parent leave their child with it unsupervised?
It's currently run on gpt-4o with some prompting guardrails, but I wouldn't trust it completely for younger (elementary school age) children.
Most of our courses are built for a slightly older age range, but we'd consider rolling out more beginner courses that are geared towards younger children and have more safety tests. If you have any thoughts let us know!
Does the list of resources simply need to be a list of links to video objects?
They can be links or actual files such as mp4 videos.
Great idea! Automated quiz generation seems like a nice use case for LLMs.
it's a natural extension after you use the LLM to generate the content. We do these in our content creation - and I assume learners use an LLM to answer them :)
this is super interesting would love to give it a try!
The tech looks cool.
But it does seem that your platform ingests video content without the permission of the person who creates these videos? The value of your platform is driven by the people creating the videos. You say that you do revenue sharing, and that you have done 5 partnerships. But you have 400 courses, so what about the other 395?
Putting it as kindly as I can: this is ethically fraught. Really, did nobody in the room point this out? You do not come off looking like a partner here.
You need to make this opt-in, not opt-out, and specify revenue sharing terms up front. Those terms need to be generous. The people who produce video content are producing the majority of your product's value. Opt-out, of an ambiguous revenue sharing agreement, is not enough.
Miyagi is a pretty nice name to give!
Honestly, this is a great idea, but I’m curious to know how you’ll address the issue of content copyright?
Nice work! Really cool.
Congrats on launch
No python course?
Bro this is dope. I want to use neetcode 150 to practice/study interview problems
Great concept!
Good Stuff!
Nice App. Another similar one is You learn ai.
This is awesome! Congrats Tyrone & Guang!
[dead]