Victor's Out-of-Band CompSci Laboratory

My public journal about what I am doing.

[vobcsl]

Regarding this Blog

For people of similar interests in computer science. I focus on programming language design. Also interested in headless programming where applications sit on a home server and are accessed on clients within the home network.

I have an Odyssee branch of VOBCSL which I store videos.

https://odysee.com/@dudeofx:1

[vobcsl]

🙄 I am actually disappointed, I was thinking it was smarter then this, I thought they had trained it to respond with: "I don't know what you are taking about"

Meta AI failed my touring test. Its true that the most likely response to that answer is yes or no. But that is not the case here.

The average human being who does not know how LLMs work would think Meta AI is role playing. But its not, not at all.

I am thinking they could have fixed this with reinforcement learning BUT I think people prefer it to behave like its role playing.

---

I tested GPT5 via DuckDuckGo, and they fixed it there.

---

Apparently Gemini also passed the test.

----

Follow up questions. I like the Gemini response. GPT5 gave me a defensive vive.

----

For us humans answering that question requires us to recall past behavior (which AIs did not have). Also we automatically remember experiences of what happens to overcooked food and dangers of fire. AIs would have this information. Hence explains Meta AI's response.

🤔 Question: In the case of AI agents who did have access to the stove but could not check if the stove was on or off. Like they could turn it on or off but they could not check the status of the stove. How would they respond?

The memory system of AI agents would be something like the guy in the movie Memento. And a little bit more complicated then that because they have to store context also. Little notes plus hallucination could spell disaster as in the case of Memento.

[vobcsl]

Communicating concepts we don't have words for

Its actually common to run into concept we do not have words for.

Did you remember to turn off the stove?

That particular question provokes a concept we don't really have a word for but its fairly common. The concept is a high provability of danger. Overcooked food, maybe even starting a fire. The probability of not having a home to return to.

We don't have a specific label or word to describe that situation. But we communicate it efficiently using that question.

Trapishka!!!!!!

What if we did assign a word to that situation. "Trapishka" It be very efficient to be driving and your spouse suddenly screams "TRAPISHKA!!" and you would know exactly what that meant, and therefore turn around and head back home to handle it.

Yes it be very efficient to communicate that way BUT we can't do that with every existing concept because then we would have too many words to remember.

[vobcsl]

The sea of information as sand pebbles.

In LLMs information is mapped onto N-Space. This space can be described as a terrain. Language can be said to be a vehicle to navigate this terrain. Words are just landmarks or labels to be used for navigation. Similar to a map.

Looking at the image above, there is a picture of sand up-close. Lets say we are looking at the terrain of information up-close. All them pebbles look somewhat the same but each one exists separate from the other. In order to navigate this terrain of information we pick a location. We give it a label and that is a word. All the surrounding pebbles are actually a little bit different from each other but we don't care because to us they are all similar and therefore we associate them all with the same label.

Some cultures may find a specific region more important and therefore are more careful with their labels. That's why some languages have as many as 300 different words related to snow and ice.

[vobcsl]

hehe - small joke referring to the inability of the Purist to contain the GOTO statement. The Purist being strict followers of structured programming.

[vobcsl]

Using AndroidSDK Command Line Tools

Creating an APK file using the command line tools was a bit of a doozie but its still doable. Old guides are a bit outdated. The following is the one that worked. I was able to compile and run an app.

https://www.hanshq.net/command-line-android.html

It seems Google is being extra picky about privacy. Labeling apps as unsafe for not meeting their specs. 🤔 Why don't they label themselves as unsafe for privacy. They and Meta are the biggest privacy offenders out-there.

[vobcsl]

Getting back to Android Development

Its been a decade since I developed for the Android platform. I do feel rusty and I do *NOT* like what google has done with Android Studio. Its literally the most resource heavy app I have ever come across with. I will not get a faster machine just to entertain their bloat. Its just going to get fatter. They probably add a bunch of AI stuff that will raise the specs even further. No thanks.

From what I gather we can still develop apps using the command line tools only. *I am greatfull*

I left Android because Google doesn't play nice. They always make things harder for 3rd parties and developers. The way I saw it was WebAssembly and ever smaller hardware will eventually obsolete Android. Hasn't happened yet but it on its way. At one point the desktop PC will be as small as a cellphone. We are almost there. When we do get there, people will be able to install Windows or Linux on these devices. Hence no need for Android.

I came back to help the F-Droid community and some of the Android developer communities. Get some exposure for SavageGardens.