Derrick Grateful

[Derek Devicemanager @IT_unhinged]

Security asked me to identify every unauthorized SaaS tool employees are using.

Our discovery software found 14.

I didn't trust it.

So I sent a company-wide email:

“SAAS AMNESTY WEEK: If you're paying for a work tool on your personal card, submit it here and IT may move it to centralized billing. No questions asked.”

By lunch I had 73 submissions.

Six CRMs.

Nine AI meeting tools.

Four different project-management platforms.

A personal Dropbox containing customer contracts.

And one $899/month “competitive intelligence” tool being used exclusively by an intern.

Security asked how I found all of this.

I said people become extremely honest when they think they're about to get reimbursed.

Finance then noticed the phrase “may move it to centralized billing.”

Not “will reimburse you.”

Correct.

Nobody was promised anything.

Our $80,000 security platform found 14 shadow apps in three months.

The possibility of free money found 73 before lunch.

Sometimes the best network scanner is greed.

[محمد بن منصور @Sakinhub]

اشياء صغيرة تصنع الفرق ✨

هنا تجربة لمهندس مدني بالصين يشرح الفرق بين قوة تدفق الماء في ماسورتين باكواع مختلفة

[CTO ROBOTICS Media @ctorobotics]

This isn't science fiction anymore.

The Chinese University of Hong Kong researchers developed a magnetic slime robot that can be remotely controlled to move, squeeze through tight spaces, and even grasp objects. One potential application? Removing objects accidentally swallowed inside the human body without traditional surgery. Soft robotics is opening doors that rigid robots simply can't.

Would you trust a robot made of slime inside your body?

🎥 Media: @newscientist , The Chinese University of Hong Kong

⚠️ This content is shared for informational purposes only. CTO Robotics Media is a media platform and does not own or develop the technology shown. Credit belongs to the original creators.

#Robotics #SoftRobotics #MedicalRobotics #Engineering #Innovation #FutureTech #Science

[Derrick Grateful @gratefulderrick]

This is impressive

[Arthur 🐳 @ArthurSXAI]

In 2024 Daniel Guo, the aerospace engineering junior world champion launched his free-flight F-1D prototype model aircraft

.. It flew for 22 minutes

[Better Man @BeBetterMan_]

TAX LOOPHOLES THAT FEEL ILLEGAL, BUT ARE 100% LEGAL

1) RENT YOUR HOME TO YOUR OWN BUSINESS

[Matt Kim @mattkim1]

AI will take your job...but not how you think...

[Derrick Grateful @gratefulderrick]

This is exactly 💯 me!!!

[Scott @scottstts]

My god this is such a good speech that every SWE needs to hear. You know what? Every person should hear it

Keep the happy memories, eyes on the reality, be excited about the future. That’s the best that anyone can do

[Scott @scottstts]

My god this is such a good speech that every SWE needs to hear. You know what? Every person should hear it

Keep the happy memories, eyes on the reality, be excited about the future. That’s the best that anyone can do

[Boardy @boardyai]

Apple was the 15th smartphone maker

Spotify was the 10th music platform

Google was the 21st search engine to enter the market

Facebook was the 10th social network

Netflix was the 7th streaming provider

Uber was the 4th ride sharing service

TikTok was the 8th short-video app

Zoom was the 11th video-calling app

Being first doesn't mean winning

[GREG ISENBERG @gregisenberg]

https://t.co/wfHAsNRGZT

[VanillaCache @VanillaCache]

ah yes, cloudflare..

> 10 million free worker requests a month
> 10gb of free R2 storage
> 1 million free writes
> 10 million free reads
> zero R2 egress fees
> free cdn
> free ssl
> unmetered ddos protection.
> waf included
> absurdly good caching
> dns so fast half the internet uses it

but yes

x taking on the real issue today

the landing page for a free localhost tunnel looks sloppy

[NetworkChuck @NetworkChuck]

Consistency beats talent. Every time.

I used to think I needed to be the smartest guy in the room. Turns out showing up every single day and doing the unglamorous reps is the actual cheat code for certs, for coding, for anything hard.

[Michael Dell 🇺🇸 @MichaelDell]

Please teach your kids about investing, compounding, and capitalism, not about me.

[Rohit @HeyRohhit]

In 1997, a developer publicly told Steve Jobs he had no idea what he was talking about. Jobs paused and answered with the most important business lesson ever delivered under pressure.

Bookmark & watch today, no matter what.

[Derrick Grateful @gratefulderrick]

@cjzafir Ok

[Howie Liu @howietl]

We’re giving away $10,000,000 to founders building agent-first businesses.

Autonomous, proactive agents will run tomorrow's companies.

We're backing 500 founders building them.

The Founding 500.

https://t.co/u8mxRWEXiT

[CJ Zafir @cjzafir]

Wao, this post blew up, and I triggered some Saastophers.

But let me tell you more about SLM fine-tuning.

In April-May this year i fine-tuned a 6.5B model: Mac-1 (it controls 487 Mac native apps)

My goal was to build a better Siri (and i achieved that).

Here's how i fine-tuned a model that runs on my Mac:

I used Codex GPT 5.5 model to design the fine-tuning structure.

- what dataset we need?
- which SLM will be perfect?
- How to perform fine-tuning?
- How to run the local model?

And trust me the current models (apart from claude models, are amazing at model fine tuning orchestration)

First we decided the models.
We used 2 models:
- Qwen 3.5 2B (for classification)
- Qwen 3.5 4B (for tool calling and responses)

Then GPT 5.5 suggested that we need a dataset of alteast 20k examples.

So we researched if there's any open source dataset available on macOS native tools. But found nothing.

Then we decided we'll create our own synthetic dataset.

Here's the fun part. For creation of the dataset I used Deepseek models.

Codex GPT 5.5 as planner and reviewer and Deepseek v4 pro (preview) as executor.

In 2 days we had verified, dedupped, evaluated dataset of 30k examples.

Then codex designed the fine tuning configs and decided things like:
- using unsloth studio
- LoRA adapter not QLoRA
- using colab pro (for cheaper 1 H200 WebGPU access)

In 8 hours the fine-tuning job was done, then we ran evals.

And in first try we got 78% accuracy on 400+ native MacOS tools under 45 categories.

I was shocked and excited on what we did.

Next 2 weeks i was obsessed about taking the accuracy to 95% and i achieved 98% on my 5th training.

The day i was about to launch the model, Apple launched new Siri (that use Gemini model at the backend)

So i decided not to go head on with Apple right now cuz you vs Apple is never good.

But I think i might launch the model now or make it open source. I think the model still beats new siri.

So why am i sharing this?

You need to understand that SLMs can beat LLMs in trained environments!

And most companies have these environments.
- customer support
- finance
- insurance
- sales
- inventory
- data entry

And use cases:
- document analysis
- information extraction
- translation
- compliance
- summarization
- classification
and many more.

A fine-tuned SLM can do all this 10x faster, 95% cheaper and on their servers (no data going outside).

And that's why many companies are now building their own AI solutions.

I'd say fine-tuning is is what coding was before 2024 (a proper skill).

That's not it.

I fine-tuned a 9B Ornith 1.5 model that writes like me (fine-tuned on my 10,000 tweets) but that story is for some other day.

I'll share step by step tutorials in coming days cuz i just love this part of AI. Stay tuned.

[Derrick Grateful @gratefulderrick]

@solotechdev @root_rao @Hostinger Which package is this

[Derrick Grateful @gratefulderrick]

@solotechdev Who is your vps provider

[CJ Zafir @cjzafir]

The more i research about chinese models and the current western mindset about AI,

The more i feel the urge of having control over my llm models.

My own private llm, own private harness, own private knowledge base and tools.

And that's happening right now. I've been using chinese models on my mac machines from last 1 year.

And trust me these SLMs (1B to 12B) models are really good at 60% of the daily tasks.

And in 6 months they'll be good at 90% tasks and that's what worries these closed labs.

When i saw elon, dario, and sam on same page, i knew instantly that it's not about AI, its about open source AI.

Get ready for a bumpy ride ahead.