So... how does AI work anyway?
@ishananand.bsky.social demystifies LLM internals for web devs using a real LLM implemented entirely in vanilla JS that runs in the browser. No Ph.D. needed!
www.youtube.com/watch?v=clqB...
Subscribe to get notified when we ship more videos from #CascadiaJS 2025! 📺
Ishan Anand (@ishananand.bsky.social) is coming to #CascadiaJS to demystify LLM internals for web developers using a real LLM implemented entirely in vanilla JavaScript that runs in the browser. No Ph.D. needed! 🤓
Ishan Anand (@ishananand.bsky.social) is coming to #CascadiaJS to demystify LLM internals for web developers using a real LLM implemented entirely in vanilla JavaScript that runs in the browser. No Ph.D. needed! 🤓
#JavaScript #LLMs
Why R1-Zero is more important than DeepSeek R1: spreadsheets-are-all-you-need.ai/arrrzero-why...
Spreadsheets-are-all-you-need is finally building out a blog. spreadsheets-are-all-you-need.ai/flops-demyst... is the very first post: A deep dive on AI's most fundamental metric.
If you're a Product Manager in the Seattle area, checkout the ProductTank Seattle event "AI in Action" on March 20th.
Eric Jensen and I will be demystifying how LLMs work and how to use them to increase your productivity in daily product management tasks. www.meetup.com/producttank-...
This might be my best AI video yet but it's also my biggest FLOP:
- The difference between FLOP, FLOPS, & FLOPs in AI
- Crack open GPT-2 (using spreadsheets-are-all-you-need) to count every FLOP
- Analyze DeepSeek's viral $5.3M training cost claim
youtu.be/T3JiBtHHTMU
This is wonderful! @ishananand.bsky.social gave a full masterclass on the Latent Space podcast - go watch it! I took his full LLM course on Maven, and it was one of the best I've attended. Congrats, Ishan and Swyx! 👏
www.youtube.com/watch?v=CHfj...
Of course, spreadsheets-are-all-you-need got a shout out (as did many other famous GPT-2 re-implementations). If you enjoyed the video, attend the weekly LatentSpace paper club!
Recently, I had the opportunity to talk at the LatentSpace paper club about OpenAI's GPT-2 model.
As I describe in the video, GPT-2 was important in the evolution of models leading to GPT-3 & ChatGPT.
Thanks @swyx.io and the paper club for having me.
www.youtube.com/watch?v=CHfj...
Thanks!
Ishan @ishananand.bsky.social provides a fantastic introduction to Deepseek and what sets this model apart in its construction. Understanding the fundamentals pays off!
youtu.be/DE6WpzsSvgU
Earlier this week I had a great discussion with Anthony Campolo on his podcast.
We covered AI, DeepSeek and building an LLM entirely in vanilla JavaScript!
Check it out at www.youtube.com/watch?v=1y28...
Just typed out 'Wait' in a reply & now I can't help thinking 'am I just an LLM reasoning model??'
Might be consuming a bit too much info about these over the past week XD
ArrrZero: Why DeepSeek R1 is less important than R1-Zero.
While everyone's talking about DeepSeek R1, the real game-changer is R1-Zero. In this video I cover how this model went straight from base model to reasoning chatbot in one giant leap.
youtu.be/DE6WpzsSvgU
It appears DeepSeek may have sparked an unexpected surge in interest running AI models locally (see graph of Google Trends for Ollama, a tool for running models locally, below).
Yesterday I had the chance to explain how LLMs work on the JavaScript Jabber podcast. We discussed my JavaScript implementation of GPT2 but also touched on DeepSeek, AI, RAG and more.
Thx @cmaxw.bsky.social, AJ O'Neal, Steve Edwards & @danshappir.bsky.social!
www.youtube.com/watch?v=n94c...
RAG - Retrieval Augmented Generation in a nutshell:
Today's mood:
Lol
We all engage in motivated reasoning from time to time. Looking back at my predictions for 2024, I definitely did that on a couple of them hurting my overall grade. If you're curious, check out how I graded my predictions. remotesynthesis.com/blog/reviewi...
Would you listen to this podcast?
An experiment in using NotebookLM to do a daily summary of Y Combinator's HackerNews. youtu.be/AL2Jk2y26ho
In the video, I also mention two future directions:
1️⃣ a copilot tutor helping you understand the model
2️⃣ using this pseudo-notebook for agentic browser automation.
Let’s call these the New Year’s Resolutions for Spreadsheets-are-all-you-need.
When I ported Spreadsheets-are-all-you-need (an Excel implementation of GPT-2) to the web, I knew the first place to showcase it had to be AI Tinkerers — where the journey of the original began. Last month, I had the privilege of debuting it there. Here's the quick video. youtu.be/Csp198KCgGI
Turns out AI can’t solve everything (yet). But if you’ve got resolutions involving AI, reply to this thread. Maybe we can figure it out together, or at least have a good laugh trying.
Here’s a process for New Year’s Resolutions that works no matter how audacious your goals:
1️⃣ Write out your New Year’s Resolutions.
2️⃣ For each resolution, ask how AI can help you accomplish the goal.
3️⃣ 365 days from now, write out the same list as in Step 1. 😂
While I’m hopeful that AI education and tools will empower people, legitimate concerns (e.g. job displacement) remain. Expect more sabots being thrown into the AI machinery. Don’t overlook this angle when planning your next AI project.
It’s easy to dismiss this as not a “technology” prediction but technologists ignore it at our own peril. This could affect everything from AI product adoption to AI regulation. The recent BlueSkey dataset controversy and Sora leak are warning signs of a brewing backlash.
(3) The AI Backlash
