Post

@hasantoxr

More from @hasantoxr

Hasan Toor

@hasantoxr

Jun 16

I built a RAG system on my own laptop that never sends a single byte to OpenAI.

100% offline. 100% open source.

Here's the entire 6-step build in plain English:

Firstly, what is RAG?

It means giving an AI access to your own files.

Instead of asking ChatGPT and hoping it knows the answer, you point it at your own PDFs, notes, and emails. It reads them and then answers based on YOUR stuff.

Same idea as "chat with your PDF" but it runs on your laptop and no data leaves your machine.

Step 1: Install Ollama.

This is the engine that runs AI models on your laptop.

You download it once. Works on Mac, Windows, and Linux. Free. No account needed.

It's called Ollama. Think of it as Spotify for AI models. You pick a model and it just runs.

Read 10 tweets

Hasan Toor

@hasantoxr

Jun 9

I use Claude every day without exhausting my token limit.

Claude doesn't count how many messages you send. It counts how much it has to read and write. Some chats use 10x faster than others.

If you want to use Claude without running out of limits, use these 10 tricks:

1. Start new chats aggressively

Claude re-reads your entire conversation every single time you hit send.

By message 25, a simple question costs 10x what it did at message 2.

New topic means new chat. Every time.

2. Batch your questions into one message

Three separate messages means Claude loads the full conversation three separate times.

→ Bad: "Fix this" then "Also explain it" then "Make it shorter"
→ Good: All three in one message

Same answers. One third of the cost.

Read 12 tweets

Hasan Toor

@hasantoxr

Jun 7

Satya Nadella took over a company that was losing and turned it into the most valuable on Earth.

He wrote the playbook in "Hit Refresh." Most people quoted the growth mindset part and missed everything underneath it.

Here are 10 things this book actually argues about technology, power, and what it costs to lead through transformation.

1) Culture is the only variable that determines whether technology investments return anything.

Nadella inherited a Microsoft where internal teams competed harder against each other than against external rivals. Engineers sabotaged launches. Bonuses were structured to punish collaboration. The technology was world-class. The culture made it irrelevant.

His first move was not a product decision. It was a cultural one. Everything after that was downstream of that choice.

2) The fixed mindset inside Microsoft was the company's biggest competitive threat, not Google or Apple.

Nadella uses Carol Dweck's framework explicitly and means it structurally. A company that rewards knowing the answer punishes asking new questions. A company that punishes asking new questions stops finding new answers. At scale, that is not a culture problem. It is a survival problem.

Microsoft had spent a decade rewarding people who were right. It needed a decade of rewarding people who were learning.

Read 13 tweets

Hasan Toor

@hasantoxr

May 15

I'm canceling my Adobe subscription after testing this.

Skywork just stacked GPT-Image-2 + Nano Banana 2 into one design workspace.

No Photoshop. No Illustrator. No designer.

Posters, logos, and full brand kits in seconds.

Here's how it works ↓

It's called Skywork Images.

→ Powered by GPT-Image-2 (99% text accuracy)
→ Nano Banana 2 (4K renders in under 10 sec)
→ Fully editable canvas — not a one-shot generator
→ Exports straight to PDF, print-ready

This isn't a prompt box. It's a workspace.

Try it free: skywork.ai/p/Ht3XoB

The killer move: "Open to Edit"

Skip the blank page entirely.

Browse the showcase library → click any pro design → it loads straight into the canvas as a new project.

Zero prompting. Just remix.

This is how non-designers ship work-ready posters.

Read 9 tweets

Hasan Toor

@hasantoxr

May 12

I'm replacing every memory layer I've ever built into an agent with this.

SureThing dropped SOTA on LongMemEval.

88.0% overall. 91.0% knowledge update. 76.7% single-session preference.

Number one across every category that actually matters.

Then their own AI walked up to the screen and started explaining the whole thing itself.

Nobody asked it to.

Every memory system I've built before this worked the same way.

Store something. Retrieve it later. Hope the retrieval actually finds the right thing.

Two separate systems pmretending to be one.

@getsurething threw that model out completely.

The memory IS the computation. Fully fused. One architecture, not two bolted together.

That's the difference. That's why the numbers look the way they do.

The benchmark breakdown:

88.0% overall on LongMemEval
91.0% on knowledge update
76.7% on single-session preference

Top of every single category.

They didn't optimize for the benchmark.

The benchmark just revealed what the architecture was already doing.

Read 7 tweets

Hasan Toor

@hasantoxr

May 9

A Chinese lab just dropped a 1 TRILLION parameter thinking model.

For free.

It's called Ring-2.6-1T from InclusionAI and it just made every $200/month "agent" subscription look like a scam.

Here's why this changes everything ↓

The numbers are absurd:

→ 1 Trillion total parameters
→ 63B active (MoE architecture)
→ 262,144 token context window
→ 65,536 max output tokens
→ $0 input. $0 output.

This isn't a stripped-down demo. This is the full model.

It's a "thinking" model built specifically for agent workflows.

Not chat. Not Q&A.

Real autonomous execution, coding agents, tool use, long-horizon tasks where the model has to stay coherent across hours of work.

The kind of thing OpenAI charges $200/mo for.

Read 10 tweets

Share this page!

Enter URL or ID to Unroll

Hasan Toor

Try unrolling a thread yourself!

More from @hasantoxr

Hasan Toor

Hasan Toor

Hasan Toor

Hasan Toor

Hasan Toor

Hasan Toor

Did Thread Reader help you today?

Don't want to be a Premium member but still want to support us?

Send Email!