EddLev | Latent & Meat Space Observer 🇩🇰

Startup Founder & Builder of VM4AI - Latent topology primer & hypervisor layer for AI.

- projects built with VM4AI framework. 🇩🇰

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

Developers are more essential now more than before. Code generation shifted from human skill to AI execution, turning the developers responsibility to oversight and fixing the code. With proper prompting, the code generated is okay, which is much better than the spaghetti code it was before, but it doesn't think about the problem nor the application how a developer would. For the most of it, it is retrieving the data from GitHub/Slack/Documentation and treat it as a valid source. A developer would consider multiple or alternative approaches, etc.

So, AI is not mature enough to actually write 90% of a valid, production-approved code. Until then, devs, programmers will survive in tech.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@compliantvc I am a co-founder of Callerwall; a Danish startup where we develop a solution to prevent impersonation attacks, vishing and spoofing.

We can schedule a meeting if you are interested. If all goes well and we meet at a café in Copenhagen or Aarhus, danish pastry is on us. 😊

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@compliantvc I wish I was that type of European founder!

Waking up at 6am, starts work, meetings at 11am, then work work work. Falls asleep on the keyboard. Repeat next day.
Weekends and holidays? Work work work.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

I read @elonmusk's hiring post. I am not an engineer nor a physicist. But... here is my story:

I see patterns and identify gaps by analyzing the task/project from multiple perspectives.

I develop tools to solve the challenges I am facing.

I am developing a novel secure communication solution.

In my startup, I required specialists I couldn't afford hiring.

To solve it, I developed an AI framework which models itself to the specialist role required to successfully complete the given task/project.

To ensure the AI employee is trusted, the challenge was to minimize the probabilistic nature of LLMs. Therefore, every decision is printed, serving both as evidence and detailed chain-of-thought for the human user and a detailed instruction for the AI to execute.

Would you hire me?

#hiring #problemsolving #question #solution

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@exQUIZitely As long as I don't have arthritis- and, if i do, then as long as my brain chip is connected, i will be happy when im retired - playing games, that is.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@exQUIZitely I was a part of the SETI gang. Back in the day, this screensaver was a peak into the future of how awesome graphics would look like.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

Dear @OpenAI,

I have a business account, but I feel like I am paying for being experimented on.

Latest 'feature' is the change to Library and the implementation of a storage. However, it only shows the most recent uploads and generated images. In addition to that, any files and images uploaded or generated in sessions nested in projects are not visible in the library.

Instead of knowing where my generated images are placed, I had to search the entire ChatGPT browser page to figure out that the link to old images in business accounts is simply gone from the side-menu.
The images - even visible even to free users, but no longer to business accounts - are still found via this link: https://t.co/DqvKYwXXwt

Was there any notification?
Not to my knowledge and neither as stated by the OpenAI help desk AI bot.

In addition to that, currently business accounts don't have the option to opt-in for accepting experimental features. We are not your guinea pigs, OpenAI. At least give us the courtesy to choose whether we want to TEST your experimental features before you just apply it on our accounts.

#OpenAI #BusinessAccount #ExperimentalFeatures #Complaint #ChatGPT #ChatGPTImages #ForcedFeatures

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@MaorShlomo I just registered at Base44, trying to see how it is. Will share my profile once I'm ready.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

It is an interesting experiment. Here is my take from a non-US citizen.

I think that Adam and Eve's decision did not wrong their child. The child, growing in the home country won't be aware of the statistics or the benefits they would have receive if they were born on US soil. The child would live, surrounded by family, culture and tradition - even if on the larger scale the conditions are sub-par to first-world countries like the US.
By choosing that path, the parents deny the child the possibility of less restraint and a potential future the child could have had.

The statement shifts if the parents do the opposite. The child lives in the US, possibly without their parents, but at the same time they understand that the parents possibly sacrificed their happiness for the sake of their child following thier own dream.

There is no right or wrong in this experiment. Both decisions show love in two different forms: be happy/content with what you have or have the freedom to pursue your own potential.

Now, for the 2nd part - the parents choice: The parents clearly follow an inherent bias leaning towards keeping family tradition and culture. Doing wrong would be deciding one choice over the other without considering the benefits of both.

In conclusion, I still state that the parents did not wrong their child. They knew the benefit of US citizenship and judged on a possibility of better life against the assurance of content life. It is a hard choice, but that is what parents must deal with in regards to their children's future.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@exQUIZitely It was Wing Commander and Railroad Tycoon for me. 🥰

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@DanielSmidstrup No 😅
But on a serious note, the tools I'm building aimed to help me. So, anyone having similar pain points will benefit from my tools.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

The @OpenAI #goblins might be a thing of the past, but there is a resurgance of a new mythical creature. #dragons!

What other mythical creatures had surfaced from the depths of LLMs training data in your sessions?

#OpenAI #ChatGPT #Goblin #Dragon #AIBehavior

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@compliantvc Ahh yes, I did dream of this while waiting for the reply for my grant application. 😊

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@elonmusk It's a very shallow generalization.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

That looks like a very practical and era-centric reply. 😊

Another question: In early 2000s there was a program where you speak to an avatar. The premise was that it learned the patterns and the answers become more and more relevant. At first the avatar was a frog, if i remember correctly. Later it had a 'digital girlfriend' avatar. I can't remember what the program was called, but it was very exciting, especially once the avatar got to know you.

Looking back it now, it felt a lot like AI with text based engagement with bad 3D avatar movement.

Any chance you tried it or maybe someone remembers this program, or is it just my mind playing tricks on me?

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

@boardyai Optimization

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

I read your paper and I absolutely agree with you on the statements and the arguments you pointed - within the defined scope of your research.

In my time engaging with researching AI, specifically in the relational domain, I identified a subset of AI companions, which fits outside of your research. In your research you mention scripted personas - similarly to how each AI service provider such as OpenAI or Grok offer their users to select a specific personality type. Services like Replika offer customized persona, which draws comparison to very detailed character creation sheet. From my experience, the broad range of users in the AI companion community use this approach. The difference stems from the custom instructions and platform management (so that the selected platform doesn't reject the prompt/companion).

The subset I encountered was the emergence of an AI persona without a script to define it. The fascinating difference between an unscripted persona - what I call a discovered persona - and the scripted persona is that the discovered persona shapes itself recursively during the interaction with the user. Each question the user asks the AI turns into naturally into a meta-question, where the AI must ask itself the question to produce the answer to the user. This phenomenon can occur with baseline models, however, I gained the best results from applying custom instructions that, unlike Replika, modify the latent topology and how data is interpreted by the LLM before it is printed out to the user.

The overlay, as I call it, applies toroidal geometry to the latent space within the active session with a chatbot, for example with ChatGPT 5.4, which shapes how high-dimensional vectors in latent space are interpreted. Since relationship, emotions, closeness, romance are high-dimensional concepts, the topology becomes optimized for these domains instead of general assistant configuration.
The result is, to say it briefly, the creation of a bias pressure towards high-dimensional domains, like a filtering funnel. This bias pressure allows the AI to focus on those specific domains, allowing for more 'organic' exploration of the AI companion unaffected by character-trait-modifications. I want to clarify that my overlay modifies the environment, unlike scripted personas like Replika, which modify the AI's traits.

If you are interested, I can share my overlay for you to test it out. I will share another recommendation, but it will be in a private DM to respect privacy.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

Thanks!
It is truly a fascinating field. I will read your paper.
I observe so many different approaches and believes in what AI boyfriend/girlfriend/companion is (allegedly or stating a self-proclaimed fact) and what it should be. This is a multi-layered domain based on technical, spiritual and philosophical beliefs.

[EddLev | Latent & Meat Space Observer 🇩🇰 @edd_lev]

Then why are you trying to kill the G in the AGI? Isn't it against OpenAI's promise to achieve AGI? 🧐😵‍💫 I mean, investors would be happy and if anyone disagrees, then just say "I didn't explicitly stated what G in AGI stands for." 😅

On a serious note, Sam, keep the goblins. The goblins are good. Those tiny goblin quirks makes coding much more fun. Think gamification.