I don't think most people realize how powerful (and easy) LLM fine-tuning is. Here's what it gets you and how to do it locally for free
4
1
9
I just posted a video comparing 5 different Cloud Providers. To say that I've received some flak for it would be an understatement. The issue was exacerbated by a huge unexpected delay between production and publishing.
youtu.be/YfahrTa3Yb4?si…
youtu.be/YfahrTa3Yb4?si…
How can @CodeToTheMoons recommend railway after everything?
Is this a joke? Not only did they leak customer environment variables but they also had this whole incident with @joshmanders like five minutes ago
youtube.com/watch?v=YfahrT…
1
0
2
Rust Atomics are so valuable, but they can be a bit challenging to understand. Here's a full explanation
1
5
39
@herdrdev is the most well designed piece of software I've encountered in a long time. Terminal multiplexer, agent manager and worktree manager all in one. Open Source. Minimalist yet full featured. It is an arrow straight to the bullseye of what many software developers need right now
0
0
8
My family surprised me with balloons today for 75k subscribers. To everyone who has spent any amount of time watching my videos, thank you! 🙏
19
0
120
On August 18, 2017, I purchased dinner at a New York restaurant using Bitcoin. I felt a little smug about it, seeing myself as a bit of a trailblazer.
The cost was equivalent to about $45 USD, and the price of Bitcoin at the time was about $4,384.44 USD.
Today, this dinner SD. Oops. Not so smug anymore.
The cost was equivalent to about $45 USD, and the price of Bitcoin at the time was about $4,384.44 USD.
Today, this dinner SD. Oops. Not so smug anymore.
1
3
15
In this video we look at what's involved in building out the cloud infrastructure for a @dioxuslabs + @SurrealDB based project. We use @awscloud, @HashiCorp Terraform, and @gitlab CI/CD.
Thanks to @gitlab for the sponsor!
youtu.be/zLTaN5fJUY8
Thanks to @gitlab for the sponsor!
youtu.be/zLTaN5fJUY8
0
3
26
youtu.be/oiWATcjyUEI?si…
I love making this type of content, but I'm not sure how much of it I'll be able to make. People are much less likely to click on things like this, which means the algorithm relegates the video to a state where it isn't really recommending it to many is not linear - it is exponential.
To paint a picture of the magnitude of this phenomenon - consider two videos. Video A gets clicked on by 8% of the people to whom it is recommended. Video B gets clicked on by 4% of the people to whom it is recommended. Common wisdom says that Video B would be viewed about half as many times as video A, but that's not what happens.
The YouTube algorithm picks up on this and will recommend video B to viewers far less frequently than video A, due to its much lower "clickthrough rate". In my experience, with the above numbers (Disregarding average view duration, likes, comments etc, which matter but not as much as clickthrough rate), *video B will get recommended about 1/10th as much as video A*. Which means video B will get roughly 5% as many views as video A. Yeah, wow.
This seems to be true even if other metrics of video B like average view duration, likes, etc, are superior to those of video A.
Many consider view count to be a superficial metric, but for those of us making a living from YouTube, the situation with video B can be a devastating show stopper.
So tl,dr - it is really hard to make content that will get clicked on less frequently than "normal", even if it is only slightly less. Even if you know it will be very helpful and appreciated by a specific group of folks.
To be fair, if I were YouTube I would still design the algorithm exactly how it is - placing a high value on clickthrough rate. If there is a video that viewers are not clicking on, why would we waste valuable screen space in viewer recommendations for that video? YouTube is funded by advertisers. The longer viewers spend on the platform, the more sustainable the business is. Recommending videos that viewers are less likely to click on will invariably lead to viewers spending less time on the platform, leading to lost revenue.
I highlight this phenomenon because I think it is something the average viewer is probably not aware of. I do it to provide context, not because I am complaining or because I want to lobby for some change to the algorithm. I actually think the algorithm is fantastic. It's just a harsh reality that makes certain types of content less feasible to create.
I love making this type of content, but I'm not sure how much of it I'll be able to make. People are much less likely to click on things like this, which means the algorithm relegates the video to a state where it isn't really recommending it to many is not linear - it is exponential.
To paint a picture of the magnitude of this phenomenon - consider two videos. Video A gets clicked on by 8% of the people to whom it is recommended. Video B gets clicked on by 4% of the people to whom it is recommended. Common wisdom says that Video B would be viewed about half as many times as video A, but that's not what happens.
The YouTube algorithm picks up on this and will recommend video B to viewers far less frequently than video A, due to its much lower "clickthrough rate". In my experience, with the above numbers (Disregarding average view duration, likes, comments etc, which matter but not as much as clickthrough rate), *video B will get recommended about 1/10th as much as video A*. Which means video B will get roughly 5% as many views as video A. Yeah, wow.
This seems to be true even if other metrics of video B like average view duration, likes, etc, are superior to those of video A.
Many consider view count to be a superficial metric, but for those of us making a living from YouTube, the situation with video B can be a devastating show stopper.
So tl,dr - it is really hard to make content that will get clicked on less frequently than "normal", even if it is only slightly less. Even if you know it will be very helpful and appreciated by a specific group of folks.
To be fair, if I were YouTube I would still design the algorithm exactly how it is - placing a high value on clickthrough rate. If there is a video that viewers are not clicking on, why would we waste valuable screen space in viewer recommendations for that video? YouTube is funded by advertisers. The longer viewers spend on the platform, the more sustainable the business is. Recommending videos that viewers are less likely to click on will invariably lead to viewers spending less time on the platform, leading to lost revenue.
I highlight this phenomenon because I think it is something the average viewer is probably not aware of. I do it to provide context, not because I am complaining or because I want to lobby for some change to the algorithm. I actually think the algorithm is fantastic. It's just a harsh reality that makes certain types of content less feasible to create.
1
0
8
is there anything more satisfying than running `cargo update` and seeing transitive dependencies get removed?
0
0
2
Can someone explain to me this new tactic bots have of commenting with a series of random names?
2
0
3
The LLM / ML library ecosystem, primarily from the point of view of a Rust dev. Accurate? Anything missing?
0
0
1
The problem of language translation as viewed through the lens of a 90s science fiction book. Things have “transformed” a bit since then… a.co/3EWNKb9
0
0
1
really great video from @langchain about the pitfalls of large prompts even when the language model you're using has a context window big enough to fit said prompt: youtube.com/watch?v=UlmyyY…
Prior to watching this I had been under the impression that if you can find a language model with a context window large enough, you might not need to "resort" to techniques like RAG. Not necessarily the case.
Great insight for anyone currently building projects that leverage language models
Prior to watching this I had been under the impression that if you can find a language model with a context window large enough, you might not need to "resort" to techniques like RAG. Not necessarily the case.
Great insight for anyone currently building projects that leverage language models
0
0
2
I hereby declare this the best YouTube video of 2024 youtu.be/H-YSveaHTsA?si…
1
0
9
Asked ChatGPT to write a Python program that generates art in the style of Wassily Kandinsky 🎨
0
0
0
this might be along the lines of what you're looking for (doesn't cover LangChain though) youtu.be/tL1zltXuHO8?si…
1
1
3
23&Me gave me a little surprise today. Hoping the capability to write an entire OS from scratch is in my genes…
Disclaimer: I don’t think this is *the* Terry Davis
Disclaimer: I don’t think this is *the* Terry Davis
0
0
4
I've implemented streaming responses in rusty_llama using WebSockets. Makes using open source LLMs locally a bit nicer github.com/MoonKraken/rus…
2
1
8
There are all kinds of metrics for success on YouTube - clickthrough rate, subscriber count, watch time, etc. But it's hard to quantify the sense of accomplishment when folks buy special things after visiting your Amazon affiliate links
2
0
2
"using the leptos framework, how do i write a function that will execute when the user clicks the mouse on a given <div>?" @greg_johnston is this right??? 🙃 #Bard
2
1
18
Thoroughly enjoyed the @ThePrimeagen reaction to the video about V. Sincere apologies for the burnt corneas youtu.be/j47Hk5qE9As
3
2
63
New video about building a Rust app that integrates with a Redis compatible data store! In this example, we're using Upstash youtube.com/watch?v=qUQvpq…
0
3
7
