I just added a humans.txt file.

I just created three browser games in three minutes.

A Doodle Jump like, an R-Type like and a 3d space shooter.

Click here to play them:

• Game 1: Doodle Jump
• Game 2: R-Type Like
• Game 3: 3D Space Shooter

Each was just one prompt with Claude 3.7. No corrections or retry.

And only the first one is playable on mobile, the others need a keyboard.

Edit: Playing with LLMs to create these games reminded me of something Steve Jobs said.

LLMs refuse do do certain things, but they can be modified to be ~100% obedient and uncensored. This (jailbreaking) process is called ablation and there is a cool article on how it works.

If you use ollama then you can easily try “abliterated” models, just search for them here.

Now why are they called “abliterated” and not “ablated” models? Simply: abliteration = ablation + obliteration.

When I have to wait for something I sometimes use AI to create images. This is one of them.

It is fun, but it is very difficult to get precise results.

I installed Pi-hole, the adblocking DNS server, on my Synology NAS machine, and since it wasn’t as straightforward as I had hoped, here’s a mini guide that I want to share.

Step 1: Install Container Manager

Go to your synology console (for me it’s at http://synology.local:5000 in my home network) and install the app “Container Manager”. It allows us to run Docker containers on the NAS.

If you are asked anywhere if it should run as root, say yes.

Step 2: Create two folders

Go to File Station and create those two folders:

/docker/pihole/etc-pihole
/docker/pihole/etc-dnsmasq.d

Step 3: Run Pi-hole in a Docker container

Open Container Manager and create a new project. I have used this YAML-configuration.

services:
  pihole:
    image: pihole/pihole:latest
    network_mode: host
    container_name: PiHole
    restart: unless-stopped
    mem_limit: 1g
    cpu_shares: 90
    environment:
      - PIHOLE_UID=1024
      - PIHOLE_GID=101
      - TZ=Europe/Berlin
      - FTLCONF_webserver_api_password=<SET_YOUR_OWN_PASSWORD_HERE>
      - FTLCONF_webserver_port=8080
      - IPv6=False
      - DNSMASQ_LISTENING=all
      - DNSMASQ_USER=root
    volumes:
      - /volume1/docker/pihole/pihole:/etc/pihole
      - /volume1/docker/pihole/dnsmasq.d:/etc/dnsmasq.d

And then run it.

For me this worked, but it could be that your UID and GID are different. Here’s how I found out what mine are:

Optional Step: Find UID and GID

I connected with ssh into the nas, in my case the command was:

ssh admin@synology.local

and then I just ran

id

And then just adjust the values accordingly.

Final words

By configuring your home router or access point to assign DHCP leases with your Pi-hole DNS server as the primary resolver, any device that requests an IP address (computers, phones, etc.) will automatically begin using your Pi-hole DNS service.

Taken on a sunday in February.

Found this TikTok clone today: wikitok.vercel.app.

What a fun idea. While scrolling Instagram and TikTok is rumored to make us more stupid, this could actually make us more intelligent.

I’ve subscribed to a VPN service and it’s really not fun to use them. The problem is that Google, Youtube, and other big sites, detect it and then they spam you with Captchas. Dubious security is exchanged for obvious inconvenience.

A much better experience for me was running Algo. It sets up a VPN on a Digital Ocean server (other options are included, but I use D.O.), and then provides a simple way to set it up on multiple devices using Wireguard config files and qr codes.

The VPN services have the advantage that switching is faster and easier. With Algo I would have to set up my own VPN server each time. But all things considered it’s the better option for me.

Since I had fun with it yersterday, I decided to create one more tiny web app. This time it’s a map of Germany with some landmarks. I believe it’s useful to learn where things roughly are.

This time the LLM did 100% of the coding. I did not have to correct anything manually and the whole process to get the result you see here took about 25 minutes.

Click here to open the map

(It probably won’t work well on mobile.)

I played around with Google Gemini and chatGPT and wanted to see if I can briefly create a small fun web app.

I could, but when something did not work, it took some light manual tweaking and debugging to tell the LLM exactly what the problem was. But it did about 90% of all the work and the whole thing took about 30 minutes to finish.

Click here to play it

And may the Force be with you.