ai and machine learning notes

when i was small I used to talk to my computer chess machine. I thought if it could beat me at chess, of course it would understand. Now the machines are talking back.

[lucymlx]

Worth clicking the link for the gorgeous animated hero image of a sleepy, interactive owl (via Submit a Static Website Form with Cloudflare Workers | Matt Ferderer)

[lucymlx]

Manus

Manus is a general AI agent that turns your thoughts into actions. It excels at various tasks in work and life, getting everything done whil

Meet Manus or, er M'anus the latest Chinese agent with its own computer. It builds websites, writes reports, and runs research tasks, even while you sleep. Enough free credits on signup for a website or two. Have fun!

[lucymlx]

Monorepo of related subdomains

Loving Cloudflare. Got one repo with multiple websites? Here's how to share the /common/ code on build, and make each one share the repo.

Step 1: Create New Cloudflare Pages Projects

Go to Cloudflare Pages

Click "Create a project"

Connect the same GitHub repo: username/repo

Name it: a-site

Repeat for: b-site

Step 2: Configure Each New Project

For a-site: - Build command: npm run build:a-site - Root directory: /a-site - Build output directory: /

For b-site: - Build command: npm run build:b-site - Root directory: /b-site - Build output directory: /

Step 3: Set Up Custom Domains (Optional)

In each project's Custom domains settings: - trades-site: a-site.yourdomain.com - accounts-site: b-site.yourdomain.com

Step 4: Deploy

Push any small change to trigger builds, or click "Retry deployment" in each project.

Your package.json already has the build scripts ready:

{ "scripts": { "build:a-site": "cp -r common a-site/common", "build:b-site": "cp -r common b-site/common", "build:c-site": "cp -r common c-site/common" } }

Each site will deploy independently but share the same repo and common files!

[lucymlx]

Join Claude! | Claude

Join Claude for a chance to win 4 months of Claude Max!

Cloude 4 Opus, is a total star for coding. Attached my signup link.

[lucymlx]

With VERT you can quickly convert any image, video and audio file. No ads, no tracking, open source, and all processing (other than video) is done on your device.

[lucymlx]

Midnight shutdown. A reminder

[lucymlx]

Cloudflared to tunnel on the Doze Server

Sshd installed in WSL. Cloudflared installed in WSL.

Maps services to domain. so edit

~/.cloudflared/config.yml

to accommodate all the services you want to forward and

sudo systemctl (status|stop}start|restart) cloudflared

USEFUL: Cloudflared tunnel list (will check the yml config if any problems)

USEFUL: cloudflared tunnel run long-string-of-tunnel-name

USEFUL: Need this bit in ssh config, for your clients.

link

NOTE: Cloudflare want you to use zero trust - whatever that is, but I couldn't make it work. The above will likely be deprecated at some point.

[lucymlx]

tips 4 wsl

Get bash when you ssh into the server, In admin powershell give the following command

New-ItemProperty -Path "HKLM:\SOFTWARE\OpenSSH" -Name DefaultShell -Value "C:\WINDOWS\System32\bash.exe" -PropertyType String -Force

REM: On client if you need to reboot you'll need to powershell.exe and run a \~\~\~shutdown /r

[lucymlx]

Gemini 2.5 pro topping the Chatbot arena. It's no longer a preview now, and is paid.

Prompts less than 200,00 tokens: $1.25/million tokens for input, $10/million for output

Prompts more than 200,000 tokens (up to the 1,048,576 max): $2.50/million for input, $15/million for output

[lucymlx]

llm command line with openrouter keys

Quite enjoying this command line tool. (llm)[https://github.com/simonw/llm]

pip install llm

llm install llm-openrouter

llm keys set openrouter

Enter key:

Then to set your fave model, see what your options are

llm models list

And (this week) i like to use Gemini Flash

llm models default openrouter/google/gemini-2.0-flash-001

Then you can have crazy thoughts and just ask on the command line

llm 'what is new wave music?'

[lucymlx]

Google AI Studio | OpenRouter

Browse models provided by Google AI Studio

So, you can add your own key for Gemini to OpenRouter and it will use your (free for a number of tokens) Google AI Studio key instead of it's own. Costs 5% of a model usage, but worth it for the all-in-one facility. Gemini flash 2 is getting some love at the moment in the stats.

Add pre-existing keys to Profile -> Settings -> Integrations.

[lucymlx]

Using OBS without video

Setting up my Mac to record work calls for transcription and summarization with AI.

To record only audio in OBS Studio, go to File -> Settings -> Output, make sure Advanced mode is selected. Open Recording tab and choose Custom Output (FFmpeg) in dropdown. Choose e.g. mp3 as container, disable Video encoder and edit the rest of settings to suit your needs.

This will save the output to .mp3 directly and you can use a separate profile for that in case you're also using OBS for normal video recording.

[lucymlx]

What Have Language Models Learned?

Interesting interactive page on BERT a very old LLM that helps show how probability works in the model.

[lucymlx]

Follow this step-by-step guide to get your first tunnel up and running using Zero Trust.

Making a tunnel is relatively easy. A tunnel forwards services (websites or other things) outiside of your home network via Cloudflare, which hides your IP, and adds SSL.

I have my website running on my raspberry pi, and my PC runs OpenwebUI for LLMs too.

No hosting fees too.

[lucymlx]

Things I'm working on. Custom requests via AppScript and Google Spreadsheets.

[lucymlx]

Perlexity gets DeepSeek with some bolded hyperbole, plus it'll answer all your 1989 Tiananmen Square questions.

This model is hosted on servers based in the US and Europe, meaning that your data is not shared with the model provider r or with China. Furthermore, we have eliminated all censorship on answers.

[lucymlx]

Machine learning, explained | MIT Sloan

In case you wondered. Great article