s-macke

Yes, that is the first "bad" ending. After that follow just the one relevant context clue and look under the bed. That might be already enough.

I chose this game, because the game just helps you, at the every step, what you have to do next. Not much to try out. Just the narrative changes. One time, you have to go to work and one time you have to flee.

Other text adventures are even more problematic. I saw GPT-4 trying for dozens of steps in the "The Hitchhiker's Guide to the Galaxy" adventure just to turn on the lights. And this just the first command you have to get right in the game.

The costs have dropped significantly months after I created the cost image. Now I use GPT-4 Turbo. This GPT-4 model understand how text adventures work and there is no need to give him known commands.

Of course you try even more sophisticated techniques than mine. I tried the ReAct pattern and virtual discussions. So far, he always stumbles at the same place in a critical understanding of the text. And I tried exactly this critical step dozens of times.

You will understand the issue yourself, once you play the game yourself. It just takes 20 minutes and is very easy:

https://adamcadre.ac/if/905.html

GPT-3 doesn't even manage the first few steps of the tested text adventure. And GPT-4 is not good at playing these adventures either.

However, my code run a newer version of the Z-machine. So Zork and many other text adventures will work. I have not tried many other games though.

I didn't make text-adventures with LLMs. I try to solve them [0].

So, far, none of the 7 tested models were able to win even one of the easiest text adventures. I tried many prompting techniques. But only GPT-4 was able to play through the first half of the game.

[0] https://github.com/s-macke/AdventureAI

Here is my approach to reverse-engineer a DOS game: Starflight [1]. Actually, all the tools failed to disassemble the code and I had to write all tools myself.

What I find most fascinating is that reverse engineering these kind of games is very satisfying and even more fun than playing them.

[1] https://github.com/s-macke/starflight-reverse

Here is my list of my favorite channels:

The best channel about AI

- AI Explained: https://www.youtube.com/@aiexplained-official

Papers and news about AI and computer graphics

- Two Minute Papers: https://www.youtube.com/@TwoMinutePapers

Awesome channel about science.

- Veritasium: https://www.youtube.com/@veritasium

Very good and long interviews

- Lex Fridman: https://www.youtube.com/@lexfridman

The rest of the list is also for fun and science

- Mentour Pilot: https://www.youtube.com/@MentourPilot

- Tom Scott: https://www.youtube.com/@TomScottGo

- Kurzgesagt – In a Nutshell: https://www.youtube.com/@kurzgesagt

- Sabine Hossenfelder: https://www.youtube.com/@SabineHossenfelder

Well, technically yes.

Both use the same technique: Smoothed Particle Hydrodynamics.

But the PS3 was able to fill the whole screen with these particles. Hundreds of them. The game Fluidity seems to have approx. 20.

Yes, we can since a long time. They were already a thing 14 years ago on the PS3 in the game "Pixeljunk shooter" [0].

But real time simulations often use massive simplifications. They aim to look real, not to match exact solutions of the Navier-Stokes equations.

[0] https://www.youtube.com/watch?v=qQkvlxLV6sI

In fact, the performance differences between the models are so significant that even a micro benchmark demonstrates their capabilities.

For example, consider my analysis [0] based on observing the progression of Large Language Models (LLMs) in a single text adventure.

[0] https://github.com/s-macke/AdventureAI#evaluation-of-other-m...

Then you might be interested in my Go implementation of Forth [1]. No assembler required, but bootstrapped and with some rudimentary internal type checking. The heap is just an `[]any` type array.

It runs in the browser via WebAssembly.

[1] https://github.com/s-macke/Forthly

I have experimented with the new WebGPU API a lot and it seems this API is predistined for node graph type of programming. I am hoping that someone will build a website such as Shadertoy [1] and compute.toys [2], but with a node graph editor.

[1] https://www.shadertoy.com/ [2] https://compute.toys/

My first racing game. I was about 8 years old and it is still one of the racing games I play from time to time.

One of the best demonstrations of the power of the C64: real physics, good AI opponents and a 3D graphics engine that is still unmatched on the C64.

Geoff Crammond is a great game developer.

A web 2D fluid simulation with surface tension, viscosity and air bubbles is this one:

https://cselab.github.io/aphros/wasm/electrochem.html

Here is the list of the 100k GPT-4 tokens as text file.

https://gist.github.com/s-macke/ae83f6afb89794350f8d9a1ad8a0...

Yes, a lot of tokens are just for code.

Edit: Here as raw link for the poor mobile devices:

https://gist.githubusercontent.com/s-macke/ae83f6afb89794350...

Before you think, that I have developed it. No. The protean clouds is just a port of https://www.shadertoy.com/view/3l23Rh to WebGPU.

The 3D is real and should work in VR.

Great technology demos. Looks like this is the first step to a WebGPU based Shadertoy [1] alternative. I will definitely compare the demos with my experiments [2].

I have a good feeling about WebGPU. Not only inside the browser, but also outside. A modern standardised interface for all platforms. And it is relatively easy to use, considering the complexity that the interface has to represent.

[1] https://www.shadertoy.com/

[2] https://github.com/s-macke/WebGPU-Lab

This is a better calculator, because it also supports GPT-4:

https://tiktokenizer.vercel.app/

Not sure, why they don't support GPT-4 on their own website.

Extracted via the GitHub Repo tiktoken [1]

After you try to decode a string the list on my computer shows up in /tmp/data-gym-cache/9b5ad71b2ce5302211f9c61530b329a4922fc6a4

[1] https://github.com/openai/tiktoken

If you are using a gas rather than a liquid, it is even simpler. Just set a high pressure at the beginning and a low pressure at the end. The gas will automatically follow the steepest part of the pressure gradient.

The static solution can be calculated by solving the Poisson equation. In [1] you can see a small implementation of the idea.

[1] https://simulationcorner.net/maze/

QAware | Frontend, Backend | München (Munich) / Mainz, Germany | Full-time, Part-time, REMOTE, INTERNS, ONSITE

QAware is a project house with ~180 employees, creating value for enterprise customers like BMW, Deutsche Telekom, Allianz and more. QAware was founded in 2005 and is very well funded.

We are continuously ranked among the top employers in Germany ([1], [2]) - in 2021 we are ranked 2nd in whole Germany and 1st in the IT sector of whole Germany. For more perks, see [3]. I'm already working there for close to 7 years, as my managers are treating me very well and the projects are (most of the time) interesting. We have a great WFH policy (even before COVID) and we don't have open plan offices!

Technologies include (modern) Java, JavaScript, Typescript, Angular, React, Kubernetes, Terraform, Spring Boot, JavaEE, Quarkus (of course not all in one project - the technology choices differ by project). At the moment I'm working extensively with Kubernetes, Terraform, JavaEE and Go.

Office in Munich is close to public transport "Giesing". Office in Mainz is near public transport "Römisches Theather". Office language is German.

We are currently looking for IT Recruiters, Cloud Platform Engineers, Software Engineers, Senior Software Engineers, Software Architects, Project Managers and IT Consultants. If you are interested, take a look at our jobs page: https://www.qaware.de/karriere/#jobs. If you need more convincing, take a look at our recruiting landing page: https://info.qaware.de/senior-software-engineer For questions, just contact me at sebastian (dot) macke (at) qaware (dot) de, I'm happy to answer them. Please don't send CVs to me, I'm just a mere software guy.

[1] https://www.kununu.com/de/qaware

[2] https://www.qaware.de/news/great-place-to-work-triple-fuer-q...

[3] https://www.qaware.de/karriere/offene-stellen/

I have read in a Deepmind Blog post, that language model AIs gives more reliable and correct answers, if it is forced into some Chain-of-thought prompt. Especially for questions, which involve math.

Something like "Think about it first, before you give the answer". Could it be that ChatGPT is being forced in this direction?

The process of writing an optimized compiler involves utilizing hundreds of algorithms and strategies, one of which is tail call optimization.

It seems that this particular optimization technique is frequently discussed and written about, leading me to wonder if it holds a unique level of importance among the other strategies.

Is my perception incorrect, or is there more to the significance of tail call optimization that I am unaware of?

QAware | Frontend, Backend | München (Munich) / Mainz, Germany | Full-time, Part-time, INTERNS, REMOTE, ONSITE

QAware is a project house with ~180 employees, creating value for enterprise customers like BMW, Deutsche Telekom, Allianz and more. QAware was founded in 2005 and is very well funded.

We are continuously ranked among the top employers in Germany ([1], [2]) - in 2021 we are ranked 2nd in whole Germany and 1st in the IT sector of whole Germany. For more perks, see [3]. I'm already working there for close to 7 years, as my managers are treating me very well and the projects are (most of the time) interesting. We have a great WFH policy (even before COVID) and we don't have open plan offices!

Technologies include (modern) Java, JavaScript, Typescript, Angular, React, Kubernetes, Terraform, Spring Boot, JavaEE, Quarkus (of course not all in one project - the technology choices differ by project). At the moment I'm working extensively with Kubernetes, Terraform, JavaEE and Go.

Office in Munich is close to public transport "Giesing". Office in Mainz is near public transport "Römisches Theather". Office language is German.

We are currently looking for IT Recruiters, Cloud Platform Engineers, Software Engineers, Senior Software Engineers, Software Architects, Project Managers and IT Consultants. If you are interested, take a look at our jobs page: https://www.qaware.de/karriere/#jobs. If you need more convincing, take a look at our recruiting landing page: https://info.qaware.de/senior-software-engineer For questions, just contact me at sebastian (dot) macke (at) qaware (dot) de, I'm happy to answer them. Please don't send CVs to me, I'm just a mere software guy.

[1] https://www.kununu.com/de/qaware

[2] https://www.qaware.de/news/great-place-to-work-triple-fuer-q...

[3] https://www.qaware.de/karriere/offene-stellen/

QAware | Frontend, Backend | München (Munich) / Mainz, Germany | Full-time, Part-time, INTERNS, ONSITE

QAware is a project house with ~180 employees, creating value for enterprise customers like BMW, Deutsche Telekom, Allianz and more. QAware was founded in 2005 and is very well funded.

We are continuously ranked among the top employers in Germany ([1], [2]) - in 2021 we are ranked 2nd in whole Germany and 1st in the IT sector of whole Germany. For more perks, see [3]. I'm already working there for close to 6 years, as my managers are treating me very well and the projects are (most of the time) interesting. We have a great WFH policy (even before COVID) and we don't have open plan offices!

Technologies include (modern) Java, JavaScript, Typescript, Angular, React, Kubernetes, Terraform, Spring Boot, JavaEE, Quarkus (of course not all in one project - the technology choices differ by project). At the moment I'm working extensively with Kubernetes, Terraform, JavaEE and Go.

Office in Munich is close to public transport "Giesing". Office in Mainz is near public transport "Römisches Theather". Office language is German.

We are currently looking for IT Recruiters, Cloud Platform Engineers, Software Engineers, Senior Software Engineers, Software Architects, Project Managers and IT Consultants. If you are interested, take a look at our jobs page: https://www.qaware.de/karriere/#jobs. If you need more convincing, take a look at our recruiting landing page: https://info.qaware.de/senior-software-engineer For questions, just contact me at sebastian (dot) macke (at) qaware (dot) de, I'm happy to answer them. Please don't send CVs to me, I'm just a mere software guy.

[1] https://www.kununu.com/de/qaware

[2] https://www.qaware.de/news/great-place-to-work-triple-fuer-q...

[3] https://www.qaware.de/karriere/offene-stellen/

The adventure game "Day of the Tentacle" had the previous version of "Maniac Mansion" included. According to Wikipedia [1] it was the first time you could play the predecessor game in a newer version.

  The whole original Maniac Mansion game can be played on a 
  computer resembling a Commodore 64 inside the Day of the Tentacle 
  game; this practice has since been repeated by other game 
  developers, but at the time of Day of the Tentacle's release, it 
  was unprecedented.

[1] https://en.wikipedia.org/wiki/Day_of_the_Tentacle

Thanks, I have mentioned this fact at the bottom.

I am grateful to the developer for doing this. This is a kind of hidden communication between fans I enjoy.

Exactly, I mean more "strongly typed". The opposite is JavaScript, which is weakly typed. When speed matters and every bit is important, it is a bad idea to write an emulator in a weakly typed language.

I have corrected the text.

The games were already playable online on archive.org [1]. My project is just focussed on the old flight simulators.

Especially I wanted to have a very light-weight emulator which starts within milliseconds. The major part of the emulator downloads with just 24kB compressed. I suppose I have reached the goal.

[1] https://archive.org/details/softwarelibrary_msdos_games

Yes, I was not informed about it. But I think that's fair. I didn't ask them either before I published the project. Please, don't make it a license issue. I developed it for fun.