As a layperson, I find this an approachable way to get an overview of a topic. However, I am only interested in a few select topics, and I was not able to find a way to subscribe to specific ones, such as #insulin-resistance (topic request ;-)
Another thing I really value in science YouTubers (e.g., youtube.com/@Physionic) is the deep dives they offer into the research—highlighting conflicting results, paying special attention to meta-analyses, etc. That would be amazing, although I realize it may be too much to ask.
Adding personalised highlights of topics of interest into the monthly emails is my top priority for 2026. Won't be too complicated, just have to set up the right API back-and-forths etc.
Good point about deep dives. I could do that for the topics I'm an expert in, however, by definition, I think this requires domain expertise and is therefore not scalable nor to be automated easily. I'm not sure if this fits the scope of the current project as I want people to find the papers of interest, read a quick summarization and then encourage them to read a selection of papers themselves.
Location: Europe
Remote: Yes
Willing to relocate: No (travel up to 2x/month)
Technologies: Jira, CSS, JavaScript, TypeScript, Java, PostgreSQL, MongoDB, OpenAI/Anthropic/Google LLM APIs, WEKA, Deeplearning4j, Figma, Shopify
Résumé/CV: https://drive.google.com/file/d/1WXOCz93eXYKDjAuVp0clZ3T8dyMp7OoV/view?usp=sharing
Email: hi@fact.ai
Work hours: CET to EST/CST
What I do:
- Turn ambiguous SME workflows into shipped AI/LLM products (0-1 -> scale)
- Double conversion (+100%), launch to millions of users, and lead cross-functional teams (6–12; coord. ~50)
- Build pragmatic delivery: discovery with real users, thin slices, instrumented loops
About me:
I’m empathetic with users and exacting about results: I’ll sit with customers to find what they need (not just what they ask), then drive an executable roadmap and delivery process. However I’m not strong at outbound prospecting. My strengths are inbound discovery, executive alignment, and durable partnerships. Roles I fit into: AI Product Manager / Product Lead / Head of Product (AI/LLM)
Could someone please help me understand how a multi-modal RAG does not already solve this issue?[1]
What am I missing?
Flash 2.5, Sonnet 3.7, etc. always provided me with very satisfactory image analysis. And, I might be making this up, but to me it feels like some models provide better responses when I give them the text as an image, instead of feeding "just" the text.
Multimodal RAG is exactly what we argue for. In their original state, though, multivectors (that form the basis for multi-modal RAG) are very unwieldy - computing the similarity scores is very expensive and so scaling them up in this state is hard.
You need to apply things like quantization, single-vector conversions (using fixed dimensional encodings), and better indexing to ensure that multimodal RAG works at scale.
Hey, sorry cannot be an intelectual sparing partner to you in any way. However, I think your sleepless days/nights were very well spent, thank you!
I am out of depth when it comes to the graph theory. But I certainly like the way you are approaching this topic.
Oh, one of my favorite researches is Michael Levin I found his work to be incredibly inspiring https://www.youtube.com/watch?v=uFMLpZkkH_8 He puts consciousness on a spectrum as well.
I'm sure you're as capable of intellectual discussion on the topic as anyone. I'm no expert and have zero formal training in any field. Like many of us, what I do have is the motivation to perpetually learn.
Michael Levin is great! Thanks for sharing all these great resources.
When I find some time, I want to put together a Python model/visualization that simulates some examples. Either way though, the part that makes this "fun" is just exploring the bounds of the universe and consciousness from the confines of my own mind. I have a feeling such activities are on some level what the universe 'wants itself' (us) to do. ¯\_(ツ)_/¯
Well, you seem crazy enough to, possibly, find some sort of enjoyment from this book: The Romance of Reality by Bobby Azarian The author explores an philosophical framework to explain the universe through the lens of entropy. I loved it.
Also, it is my firm believe that in the moment you have emergence at play in any shape or form, simulation is the only viable path for in depth research. In my mind any form of neural network is a deterministic chaotic system[1]. If this would be true, simulation is the way how you test and discover truths. So I am a big fan of your intent to start programming.
Another book which your text reminded me of was Scale by Geoffrey West. Again not something pertinent to your effort of Modeling Consciousness, but it informed and delighted me in my attempts to make sense of reality.
Here is the perspective of a serial founder, exploring fields which I might be able to disrupt:
- The regulatory moat is immediately intimidating.
- The data moat, often, is quite surmountable as long as LLMs can generate high-quality synthetic data (e.g., user preferences). On this I disagree with the author, to some extend.
- The "distribution moat" is another significant barrier. Even if I have a superior product, if the marketing and sales demands are so high that neither I nor an army of bots can manage it alone, the business becomes nonviable (e.g., enterprise sales).
- "Switching costs" form the next moat. The higher these costs, the greater the value per dollar I must offer over the incumbents (e.g., software for dentists).
- Another key barrier is the “business rules” moat. Achieving 80% of the required features may be easy, but as customers demand 90% or 95%, the complexity and cost of reverse engineering grow exponentially. The more mature the market, the higher these demands (e.g., Jira).
With the power of LLMs at my disposal, I have reaffirmed two core beliefs:
1. I must focus on a niche small enough, so that I am the only provider. (e.g., accounting software for gym owners in the north of France)
2. I must offer a value proposition different from that of the incumbents, where competing with me, would harm their business. (e.g., image editing app where you pay per hour used)
You're likely tossing out a random example on #1, but if that were a real idea, you'd need a good answer for: why can't gym owners in the north of France just use quickbooks or xero.
I like your train of thoughts. I think you're missing the network effect. It is often an overplayed classic, but I do think that it matters in an AI world.
As being part of a software development team working on RPS (runway performance) I can tell you, it is stories like this which keep us up at night.
The amount of business logic involved is staggering, but miraculously there are experts who truly seem to understand almost all of it.
Understanding “almost all of it” is not enough, so there are several different experts who in their entirety hopefully capture 99.99…% But you don’t know what you don’t know…
One of the “calming” facts is, that in the tens of thousands of test cases we compared our results to existing solutions, we did discover errors in their calculations. The “calming” aspect is, due to the many safe-guards these errors seemingly never surfaced.
Coming from a machine-learning background I quickly understood that there is little appetite for AI solutions, as any optimization which would reduce fuel consumption, is easily offset by all the safety margins.
In my mind the goal of the software is to make everything explainable and proof that you truly did think of every eventuality.
I am very curious to learn what major f*-up lead to planes being grounded.
My absolutely new favorite heuristic for decision making comes from Mr. Bezos:
Is the decision costly to reverse (costly as in time / money)?
No → Make a decision fast.
Yes → Take your time, consider alternatives, talk to people, think some more.
I read "Hour of Our Delight: Cosmic Evolution, Order, and Complexity" by Hubert Reeves as a teenager which I can also recommend. That's how I learned about the low entropy sunlight converted to higher entropy IR radiation on earth, allowing complex life to exist without breaking the 2nd law of thermodynamics. So I was happy to see it talked about.
Now, life seen as an entropy-increasing system was a really interesting take on the topic! Until now my understanding was that it's the universe expansion that actually sorta decreased the local "entropy density" and allowed to radiate low entropy IR in the ever colder empty space in the first place.
> The author argues in that life is very likely because it is the most efficient way to increase entropy in the universe.
In the universe? You could argue life is the most efficient way to increase entropy within a local system ( like ants on a forest floor dissembling an insect carcass ), but I find it hard to believe life is the most efficient way to increase entropy in the entire universe. The assumed expansion of the universe itself increases entropy in a manner no life could ever hope to achieve.
I'll thank this recommendation by recommending The Gramatical Man: Information, Entropy, Language and Life by Jeremy Campbell. More or less same topics.
A life form is definitely something I'd consider to be highly organized, i.e. low entropy, but that order is maintained via a greater relative increase of entropy by destroying and consuming other "instances" of low entropy (plants, animals, etc) for nutrition, which are they themselves sustained by disproportionate increases of entropy (consuming of other organisms, the sun doing its thing, etc).
So, as a life form tends to fight for its own survival -- survival being "just" a process of increasing entropy every else but the life form itself -- it is indeed an extremely efficient way of increasing entropy in the universe, in that is requires only a single point of decrease in order to cause a theoretically infinite instances of increase.
I suddenly feel an awful lot like a cow on my way to the slaughter indeed, and have the urge to go play Dark Souls.
Computers made of water, wind and wood... may be bubbling, sighing or quietly growing without our suspicion that such activities are tantamount to a turmoil of computation whose best description is itself. -- A. K. Dewdney, Computer Recreations [0]
I don’t think that’s what they’re saying. They’re saying both a bathtub’s whirlpool and an organism’s energy consumption are examples of methods to increase entropy and both of them may emerge specifically because of the universe’s general tendency toward higher entropy.
For anyone interested in this topic, I can wholeheartedly recommend the book: "The WEIRDest People in the World" by Joseph Henrich.
The book provieds a beautifully constructed framework to understand culture and its effect on psychological development, as well as economic development.
1. In the book, the author argues that innovation rates of countries correlate positively with the psychological dimension of "individualism". (FIGURE 13.5)
2. The the dissolution of strong kinship ties correlates significantly with high economic growth. An argument the author repeats in his paper from 2022 [1]
My personal opinion is: gaining wealth is not a zero-sum-game, but a — the more participants the bigger the pie gets — kind of game.
I truly believe, if cultures and institutions change, everyone in the world can be wealthy. With "wealthy" I refer to the living-standard North-American or Europeans enjoy.
> My personal opinion is: gaining wealth is not a zero-sum-game, but a — the more participants the bigger the pie gets — kind of game. I truly believe, if cultures and institutions change, everyone in the world can be wealthy. With "wealthy" I refer to the living-standard North-American or Europeans enjoy.
I'm less convinced. A lot of the improvement to living standards in the west for the past 50 years or so have been a result of shifting labor to countries with lower standards of living. That being said living standards in post-war US is not a bad target and I don't see a physics problem with attaining that.
This is why I love HN, you gave me a surprising, but after evaluating your links trustworthy explanation.
Thank you!
If anyone else is wondering Exodus Cry is an organization which originated out of a weekly prayer group founded 2007 [0], they apparently lobbied Mastercard to only accept providers which verify the identity of all performers & review content before any upload [1]. Which is almost impossible for the tubes / onlyfans => no more payment
As a layperson, I find this an approachable way to get an overview of a topic. However, I am only interested in a few select topics, and I was not able to find a way to subscribe to specific ones, such as #insulin-resistance (topic request ;-)
Another thing I really value in science YouTubers (e.g., youtube.com/@Physionic) is the deep dives they offer into the research—highlighting conflicting results, paying special attention to meta-analyses, etc. That would be amazing, although I realize it may be too much to ask.