iiinova — where your digital self keeps evolving

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 11/10/2025

@enosarris.bsky.social will Riley Greene need to go through Cody Bellinger transformation?

120 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 31/08/2025

A fine summer night for some minor league baseball …

010 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 19/07/2025

How can Chiefs keep getting away with this?

020 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 20/06/2025

Look at who is on tv while I am just having brunch outside of Wrigley!? @enosarris.bsky.social

010 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 27/12/2024

Before you go…Puka or Chase? (Assuming salary fits) Is ChatGPT right?

000 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 25/12/2024

How can Chiefs keep getting away with…this?

020 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 28/11/2024

8/ Key takeaway: Size isn’t everything. Alignment > Scaling. By fine-tuning with human feedback, InstructGPT shows we can get better, safer AI without endlessly chasing bigger models. Linked to Paper 👉 buff.ly/3Z2e0v3

120 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 28/11/2024

1/ Imagine if AI could actually do what you asked, instead of hallucinating random facts or being toxic. That’s the idea behind InstructGPT—OpenAI’s game-changing 2022 paper. It’s not just smarter; it listens. 🧵👇

100 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 27/11/2024

1/ In 2018, OpenAI released a bombshell paper: "Improving Language Understanding by Generative Pre-Training". It didn’t just change AI—it gave us the roadmap for GPT models we see today. 💡✨ What’s it about? Let’s break it down in 10 tweets that are smarter than a fine-tuned Transformer. 🧵👇

100 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 26/11/2024

Travel is tough during holidays. If you can take advantage of Clear, I’d be happy to share guest passes. 3 months should cover my friend to 2025z Hit me up via DM or reply!

Guest passes for CLEAR plus for 3 months

010 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 25/11/2024

Want to geek out more about LLM training costs, Transformers, or the future of AI? Let me know! Meanwhile, think of Transformers as the Netflix of AI: efficient, addictive, and totally changing the landscape. 🍿🤖

000 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 25/11/2024

Look at the chart 📊: - 2017 (Transformers): ~$800. - 2019 (BERT): ~$1,344. - 2020 (GPT-3): ~$153,600. - 2023 (GPT-4): ~$1.5M+. Transformers scaled up but kept costs **lower per token** thanks to parallelism & GPU/TPU advancements.

100 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 23/11/2024

TL;DR: Reed Hastings wants Netflix to "win" your attention. 🎥🛌 In AI, attention means helping models "win" by focusing on the most important data. 🧠✨ "Attention is All You Need" didn’t just reshape AI—it reshaped how we think about focus.

100 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 22/11/2024

Not all GenAI = LLM! 🤯 LLMs: Generate text (ChatGPT 📝). Image GenAI: Create images (MidJourney 🎨). MidJourney is GenAI but NOT an LLM. LLMs are a subset of GenAI, and GenAI is a subset of DL. It’s a hierarchy! (picture from Sebastian Raschka's Build A Large Language Model from scratch book.)

100 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 18/11/2024

(While PyTorch is not a prerequisite, the book did come with Appendix A to help you learn PyTorch, and if you have no idea what PyTorch is, you should work through Appendix A to get an understanding)

110 0

👟Jim “🈶️” Liu👟 @ founder mode @itsj.im · 17/11/2024

Three stages of coding a large language model (LLM) are: - implementing the LLM architecture and data preparation process - pre-training an LLM to create a foundation model - fine tuning the foundation model to become a personal assistant or text classifier