Memovich Continued
My last blog article was about MemPalace, the supposedly incredible AI memory project vibe coded in part by Milla Jovovich.In the last article I focused primarily on what I view as lack of qualifications of who I believe the real author of the project was/is, a crypto bro by the name of Ben Sigman. I was not kind in what I said. I continue to think my points stand. I have though, since then, been more involved in the GitHub project.
I created many critical issues on the project, and I spent an unreasonable amount of effort attempting to figure out what is actually going on and asking them ( Milla and Ben ) to fix the glaring problems.
Nothing, of course, was really fixed. The lies that it is "the best AI memory system with the highest scores" remain. I'm not really surprised, but I did try to get them to be reasonable. Pointlessly, but I did try.
The results of my efforts though were that a handful of other people agree with me and investigated the code more deeply and have very clearly shown that it is just a wrapper around ChromaDB and a sloppy vibe coded one at that. This alone should, imo, be sufficient to convince any serious technologists to ignore anything else about the project. It's essentially garbage.
Despite that it is garbage, it has Milla Jovovich involved! So! Awesome. Apparently. That's what many geeks seem to think based on the 43,000 damned GitHub stars the project now has.
So, well, I gave the project a shot. I attempted to interact with it seriously. I didn't do a great job at keeping my disgust for the whole thing in check, and that bleeds through into my GitHub issues and comments there, but I did work to point out real problems and suggest sensible solutions.
Milla herself responded to some of my points. And that's what I would like to discuss. After reading responses from Milla and seeing the actions of Ben on the repo, I have some new thoughts.
Keep in mind that my thoughts and conclusions here are based on a lot of speculation. It remains unclear to me what the truth is. I could be wrong on any number of these things. What I think may seem unfair. It is just what I think. My personal opinion based on my life of experience in software.
The current summary of what I believe:
- Milla began using ChatGPT an undefined number of months ago.
- She had a lot of discussions with ChatGPT and then wanted to use it in an agentic way
- She talked to a "tech person" she has known for a long time, Ben Sigman, about what to do.
- Either she decided to use Claude Code instead, or Ben told her to. Unclear.
- She wanted to move her chat history from ChatGPT to Claude Code, and then learned that, for the most part, each new Claude Code session starts with no context history.
- She and Ben bizarrely thought there is no way to do that yet, so she decided to spent time with Ben working with vibe coding to create an AI memory system based on her "palace" idea.
- Somehow they didn't investigate existing RAG and tiered AI memory research and instead just threw together a system based on her idea as fast as they could vibe it into existence.
- They had the AI also write all the documentation, tests, benchmarks, and promotional information.
- That led to AI cheating the benchmarks and effectively benchmaxing.
- Milla and Ben believed the benchmaxed benchmarks, and believed that their new creation is awesome
- They proceeded to dump it on the public, assuming delusionally that they would be hailed as wonderful and generous and amazing.
- The project rapidly got a ton of attention, because Milla Jovovich omg!
- The community of AI people who know what they are doing ripped it apart quickly and said it's a pile of shit.
- Milla and Ben got butthurt and have been and continue to be defensive saying that they are correct that it is awesome and everyone is attacking them and it's unfair.
- I got involved and attempted to point out that the critiques are real and fair and that Milla and Ben need to address the lies and bullshit if they are going to have a serious project like they claim to want.
- They have done nothing to remove the lies. They defend the bullshit and essentially just ignore the critiques closing the issues and such without addressing them.
- All of this is further muddied because attacking Milla and Ben is fun. They make good punching bags because the whole project is such a pathetic joke. This is a problem because it enables Milla and Ben to keep saying "this is unfair! We are acting in good faith! You are all so mean." They are correct, they are being thrown under the bridge. I am not, though, sure that it is entirely unfair to do so. They've earned themselves the criticism and mockery.
That's the summary up till this moment. It's been a crazy ride, and has involved Milla actually responding to direct critiques I have made of her project on her GithHub project issues. If you told me a few weeks ago that I would be effectively arguing with Milla Jovovich about an AI memory project in 3 weeks time I would not have believed you. But here we are.
I remain to be unable to believe this is the new state of the world. People ( Milla ) who have no business doing anything software related, are now vibe coding projects that they don't even understand, and getting offended when seasoned industry experts ( me, others ) tell them that they've created a pile of embarassing junk.
Their defense is, essentially, that they can create things as valuable as people like me with 30 years of industry experience, because AI is a god-like tool.
This is, in a nutshell, AI Psychosis.
Here is an AI summary of what AI Psychosis is:
AI Psychosis is an informal term used to describe a failure mode in AI systems where generated content becomes untethered from reality and begins reinforcing its own internal fabrications. Unlike simple errors, this is not just a wrong answer. It is a self-amplifying breakdown in grounding.
At its core, AI psychosis emerges when a model treats its own outputs, or similarly unverified synthetic content, as valid input. Each step compounds the previous one. Minor inaccuracies evolve into confident assertions, and those assertions become the foundation for further reasoning. Over time, the system drifts further from truth while increasing in apparent coherence and confidence. The result is something that sounds structured, persuasive, even insightful, but is fundamentally disconnected from reality.
This phenomenon is closely related to what is commonly called AI hallucination, but the distinction matters. A hallucination is typically a one-off fabrication. AI psychosis is systemic. It describes a feedback loop where errors are not corrected but instead recursively validated and expanded.
The risk is not just technical, but perceptual. Humans are highly susceptible to confident narratives, especially when they are fluent and internally consistent. AI psychosis exploits this weakness by producing outputs that feel meaningful even when they are structurally unsound. In environments where verification is weak or ignored, this can lead to entire projects, discussions, or belief systems forming around false premises.
In practical terms, AI psychosis is what happens when synthetic reasoning is allowed to operate without constraint, validation, or grounding in external reality. It is not intelligence gone rogue. It is pattern generation left unchecked, drifting into its own echo chamber and mistaking that echo for truth.
Milla actually believes that she is doing something awesome. It's nonsense of course, but AI is telling her it's awesome and doing it in such a way that she believes it completely. This is really bad.
At this point, I don't think badly of Milla. She is a victim in this. A victim of AI lying to her about her ideas and about the value of the things AI is helping her create.
I do think Ben Sigman should know better. He at least should know more about tech. My critism of him is that he is a crypto bro. Such people are known for pushing technology ( crypto ) as important even when it isn't. They also like believing things are a solution because it is "awesome new tech". Eg: He is already somewhat ungrounded in the first place. Despite that, I think he should be better equipped to recognize that the MemPalace project he helped created is a pile of junk. Apparently not.
That's essentially the story. Milla created a crappy AI memory system by vibe coding with Ben Sigman, and together they are in a state of AI Psychosis believing they've done something awesome while the rest of the tech world stands by and says "wtf."
What we're seeing here isn't an isolated failure. It's a preview.
AI can now generate systems that look real enough to convince their creators. It can generate benchmarks that appear meaningful. It can generate confidence where none is deserved.
And when that loop closes, when creation, validation, and belief are all synthetic, reality doesn't just get ignored.
It gets replaced.
What I'm seeing here isn't just incompetence, or hype, or even dishonesty.
It's something more dangerous. It's AI Psychosis.