Ideas are just that—ideas. What really matters is putting them into action. History is full of examples where different people had the same idea, but only those who acted on them made an impact.
Michael Jackson understood this better than most. He believed creative ideas were out there in the universe, just waiting for someone to grab them. If he didn’t act fast, he worried that someone else—especially Prince—might get to it first.
His longtime collaborator, Dorian Holley, recalled that Jackson would sometimes wake up in the middle of the night with a song idea and immediately call his team to start working on it. He genuinely believed that if he waited too long, Prince might "catch" the idea from the ether and record something similar.
Michael Jackson understood this better than most. He believed creative ideas were out there in the universe, just waiting for someone to grab them. If he didn’t act fast, he worried that someone else—especially Prince—might get to it first.
His longtime collaborator, Dorian Holley, recalled that Jackson would sometimes wake up in the middle of the night with a song idea and immediately call his team to start working on it. He genuinely believed that if he waited too long, Prince might "catch" the idea from the ether and record something similar.
❤1
Knowledge Distillation
Imagine you have a master chef who knows every nuance of cooking. They can adjust heat perfectly, season by intuition, and handle ingredients with expertise built over decades. That's your large, complex AI model. Now you want to teach a junior chef (a smaller model) to cook almost as well, but without all the years of experience. Instead of making the junior chef go through the same decades of training, you have them watch the master chef cook and specifically learn the important parts - not every little detail, but the key techniques that produce good results. That's knowledge distillation - taking a big, complex AI model and teaching a smaller one to mimic its important behaviors without needing all the computational complexity.
Imagine you have a master chef who knows every nuance of cooking. They can adjust heat perfectly, season by intuition, and handle ingredients with expertise built over decades. That's your large, complex AI model. Now you want to teach a junior chef (a smaller model) to cook almost as well, but without all the years of experience. Instead of making the junior chef go through the same decades of training, you have them watch the master chef cook and specifically learn the important parts - not every little detail, but the key techniques that produce good results. That's knowledge distillation - taking a big, complex AI model and teaching a smaller one to mimic its important behaviors without needing all the computational complexity.
Speculative Decoding
Imagine you're exploring a massive maze with two versions of yourself:
• Careful-You: who meticulously maps every turn, measures every angle, and checks every wall's material. Slow but nearly always right.
• Quick-You: who runs ahead making rough sketches, sometimes missing details or making wrong turns, but covers ground much faster.
Here's the clever part: While Careful-You is still measuring and mapping the first chamber, Quick-You has already sprinted ahead and roughed out the next few rooms. When Careful-You finally finishes and confirms Quick-You's first guess was right, something magical happens - you can instantly 'inherit' all the rough work Quick-You did beyond that point. You don't just know one room was right - you get to keep all the preliminary measurements and relationships Quick-You discovered while racing ahead.
The real efficiency isn't from having two explorers - it's that when Quick-You guesses right (which happens surprisingly often, like 90% of the time), Careful-You can skip doing the detailed mapping work from scratch and instead build upon Quick-You's rough sketches. And since Quick-You uses simpler tools and skips many measurements, their work, even when occasionally wrong, costs much less energy than Careful-You having to start over. The few times Quick-You makes a wrong turn are more than made up for by all the times they blaze a correct path forward.
Imagine you're exploring a massive maze with two versions of yourself:
• Careful-You: who meticulously maps every turn, measures every angle, and checks every wall's material. Slow but nearly always right.
• Quick-You: who runs ahead making rough sketches, sometimes missing details or making wrong turns, but covers ground much faster.
Here's the clever part: While Careful-You is still measuring and mapping the first chamber, Quick-You has already sprinted ahead and roughed out the next few rooms. When Careful-You finally finishes and confirms Quick-You's first guess was right, something magical happens - you can instantly 'inherit' all the rough work Quick-You did beyond that point. You don't just know one room was right - you get to keep all the preliminary measurements and relationships Quick-You discovered while racing ahead.
The real efficiency isn't from having two explorers - it's that when Quick-You guesses right (which happens surprisingly often, like 90% of the time), Careful-You can skip doing the detailed mapping work from scratch and instead build upon Quick-You's rough sketches. And since Quick-You uses simpler tools and skips many measurements, their work, even when occasionally wrong, costs much less energy than Careful-You having to start over. The few times Quick-You makes a wrong turn are more than made up for by all the times they blaze a correct path forward.
❤1
Forwarded from Beka (Beka)
Hey guys good news :)
Better Auth has been accepted into Y Combinator's Spring 2025 batch (X25)! 🎉
Myself and @kinfishfarms, will be part of YC's first spring batch. Super excited and thanks everyone here for being part of my journey so far :)) but a lot more to come!
Better Auth has been accepted into Y Combinator's Spring 2025 batch (X25)! 🎉
Myself and @kinfishfarms, will be part of YC's first spring batch. Super excited and thanks everyone here for being part of my journey so far :)) but a lot more to come!
👏2
The Ship of Theseus is a philosophical thought experiment that questions the nature of identity over time. It asks: if all parts of a ship are gradually replaced one by one, at what point does it cease to be the original ship?
mogn
The nerve of these people!
Let's say I update the source code of the software one line at a time. At what point does the project stop belonging to the college? 🤔😂
Forwarded from et/acc
$4.15B invested in open-source generates $8.8T of value for companies (aka $1 invested in open-source = $2,000 of value created)
- Companies would need to spend 3.5 times more on software than they currently do if OSS did not exist
(Harvard research on open-source)
- Companies would need to spend 3.5 times more on software than they currently do if OSS did not exist
(Harvard research on open-source)
The man who doesn't read has no advantage over the man who cannot read.
– Mark Twain
😁1💯1
This media is not supported in your browser
VIEW IN TELEGRAM
I let Claude play with a Rubik's Cube
⚡1
mogn
https://github.blog/ai-and-ml/generative-ai/what-are-ai-agents-and-why-do-they-matter/
Anthropic
Building Effective AI Agents
Discover how Anthropic approaches the development of reliable AI agents. Learn about our research on agent capabilities, safety considerations, and technical framework for building trustworthy AI.
mogn
https://youtu.be/3oCFHE9x0As?feature=shared
Who knew ChatGPT started as a reddit chatbot 😂