Andres Andrade (@axandrade) header
Andres Andrade (@axandrade) profile picture

´He who jumps into the void owes no explanation to those who stand and watch´

Sao Paulo, Brazil
Joined February 2009
Photo by Andres Andrade (@axandrade)
Andres Andrade retweeted
Ontologies for AI in practice: Do LLMs need ontologies, or are text definitions good enough? A three-part experiment on a small, deliberately ambiguous genealogy dataset (fictional people) tested what changes when an LLM is given progressively more semantics.  The task: count a person’s ancestors. The evidence is anecdotal by design, but practical. 🔎 Part I: adding semantics step by step Data alone: many errors, and a correct count only by chance Cleaner dates: no change Plain-text definitions: a substantial improvement RDF knowledge graph: the same answer, but it exposed hidden assumptions, including an imprecise question (“how many ancestors in this dataset?”) Curated graph: more precise and consistent answers Ontology: the correct count 🧪 Part II: model size and stability Small local models (4B to 8B) could not use added semantics Impact started to appear at 32B parameters Text definitions produced high variability across runs Knowledge graphs with ontologies matched or beat definitions and were more consistent Removing the text definitions from context improved graph-based results 🧩 Part III: how to provide semantics OWL axioms only, OWL with definitions, OWL with operational directives, Apache Ossie (OSI) and Google’s OKF were compared. Takeaways: - Axioms alone were not enough, and the natural-language definitions inside ontologies did most of the work - With a capable model, the format mattered little. Opus reached perfect results across representations, while DeepSeek-R1:32B used none of them - Ontology plus knowledge graph was the most effective and consistent option - Among approaches that annotate data directly, OKF ranked highest Definitions alone are not enough. Results were strongest when context-specific meaning is connected to the data elements themselves, as in a knowledge graph. By Andrea Splendiani Ontologies for AI in practice / part I sbee.link/gkx3nd6rec Ontologies for AI in practice / part II sbee.link/erhmudfwc6 Ontologies for AI in practice / part III sbee.link/3d7xhwbnm4 #LLM #FAIRData #OWL #SemanticLayer #DataGovernance --      Missed a talk at #CDL26? Or missed the whole thing?      All talk recordings will be available in the next few days.      With this many ideas and speakers, nobody could catch them all live      ✅ Were you there? Replays included in your ticket   🎟 Couldn’t make it? Get a remote ticket for permanent on-demand access sbee.link/pqwxhda3n4
6
19
1
209
11,301
Andres Andrade retweeted
The new enterprise sales playbook: - Services as the Trojan horse: companies will let a few smart humans in long before they'll buy software they don't understand. - Ground-truth data capture: capture workflows nobody wrote down. The weird exceptions and workarounds are what generic AI gets wrong. - Expand margins through software & hardware: every customer should need fewer of your people than the last. Palantir got called a consulting firm for years for doing this.
21
14
8
290
25,079
Andres Andrade retweeted
Computer, visualize earth across 400M years
Computer built an interactive globe of Earth's 4.5 billion years from 14 open datasets, with its own engine moving 258 plates across 109 paleo-elevation maps. Drag the timeline, or type a city to see where it sat 400 million years ago. deeptime.pplx.app/
62
39
4
544
46,979
Andres Andrade retweeted
Whenever I feel demotivated i rewatch this... Elon Musk in a living room in 2008, 12 days from bankruptcy, pitching a $109,000 car to people...
1,134
12,166
740
98,345
10,948,238
Andres Andrade retweeted
4
2
1
52
4,148
Andres Andrade retweeted
Whenever I feel lost, I watch this.
347
4,578
238
24,921
873,034
Andres Andrade retweeted
Agents are eating discovery, so what happens to marketplaces? I see two futures: 1/ Marketplaces build their own agents, and each one stays the destination for search 2/ Agent-to-agent infra is built out and my agent searches across marketplaces and suppliers via API Future 1: Each marketplace has an agent Resy has an agent. I text it in iMessage, use it in the app, or have my personal agent talk to it. Resy is still the destination for reservations in my mind. > Likeliest where brand and taste matter or supply is exclusive: restaurants, events, fashion Future 2: Marketplaces become suppliers behind my agent I tell my agent I need a doctor. It checks my insurance, location, and medical history, queries across marketplaces/practices, verifies credentials, and brings back three options. I approve one, and booking happens between agents. I never see the exchange. My agent is my entrypoint, and the marketplace is a supplier behind the scenes that I never see. > Likeliest where supply is fragmented and fit matters more than brand: care, in-home help, moving, home repair IMO these aren’t mutually exclusive. Both can exist, but they compete for the entrypoint. Under 1, the marketplace still owns the relationship and brand in my mind. Under 2, it’s one source - supply becomes headless to me. Today, marketplaces do four main jobs: discovery, trust, transaction, and operations. With discovery melting away, the value moves to what sits underneath the marketplace: > Trust and verification layers agents can query (credentials, identity, insurance, outcomes) > Supply-side agents that manage availability, pricing, and listings across channels > Structured-data registries for fragmented categories (care, home services, specialty health) > Agent-to-agent transaction rails: payments, dispute resolution
28
13
3
151
20,555
Andres Andrade retweeted
Payment for agentic order flow is the new AI biz model. AI is just front running .
1
4
447
Andres Andrade retweeted
That‘s pretty much it. The rest is noise.
Woman who lost her battle with cancer left behind a note on what matters most (2017) [via/ FB Holly Butcher]
32
1,340
8
20,898
1,258,671
Andres Andrade retweeted
Technical moats are mostly over at this point. If you’re building something that is meant to be durable, the only moats now are: • Proprietary data • Network effects • A physical component
1,144
2,078
907
25,710
3,343,815
Andres Andrade retweeted
It is the beginnings of the end of software AI is starting to become good enough at reverse engineering and improving almost anything The latest models were a step-change in capabilities that flew under the radar
157
96
72
2,337
495,520
Andres Andrade retweeted
i actually think there's a huge opportunity here we're so caught up in our indie hacker bubble that we forget most people have no clue how any of this works the average person doesn't want to build their own software, they just want something that works i think who's buying software is shifting more than software itself
Software is dead It's now the era of personal software and everyone just builds what they need whenever they need it I think
72
14
6
453
316,806
Andres Andrade retweeted
I’m in love with this sentence: “A person's capacity for growth is directly linked to how much truth they can handle about themselves without running away.”
83
3,152
111
16,201
299,041
Andres Andrade retweeted
Solving a Rubik's Cube with graph theory.
476
10,786
1,014
73,122
12,038,080
Andres Andrade retweeted
One of the greatest viruses of the modern man is looking at Instagram Stories. You will melt your brain looking at a barrage of outcomes—like vacations, parties, cars—with no visibility of the process that got them there.
1,396
4,343
702
39,839
2,240,925
The Price of a 100 Bagger Everyone wants to find a 100 bagger. We look at $AMZN, $AAPL and $NFLX and imagine investing $10,000 and watching it become $1 million, but we rarely think about the thousands of difficult days between those two numbers. In hindsight, extraordinary wealth looks almost effortless because history removes the uncertainty that made earning it so difficult. A 50% crash 20 years ago looks like a tiny wrinkle on a chart today, but when half your money is disappearing in real time, you have no idea whether you are looking at an opportunity or the beginning of the end. Great investments disappoint, become expensive, attract competitors and go nowhere for years, repeatedly giving intelligent investors intelligent reasons to sell. The price of a 100 bagger isn’t what you pay for the stock, it is everything you have to endure while owning it. Ironically, losing money isn’t the only thing that separates investors from extraordinary businesses, because making money can be equally dangerous. When $25,000 becomes $250,000, selling doesn’t feel fearful, it feels responsible, creating one of investing’s strangest paradoxes: we spend years searching for an exceptional business and then allow its exceptional success to become the reason we no longer own it. This is why finding a 100 bagger is easier than owning one. Plenty of investors understood that $AMZN was extraordinary long before it became one of the world’s largest companies, but recognizing greatness and allowing greatness enough time to matter require entirely different skills. Finding greatness requires intelligence, while holding greatness requires temperament. Eventually success creates another problem because the investment becomes large enough to frighten you. A $20,000 position feels like an opportunity, while the same position after becoming $1 million suddenly feels like a risk, even though nothing about the business necessarily changed. Sometimes the investment doesn’t become more dangerous, the amount of money simply makes us more afraid. Then there is envy, perhaps the most underestimated destroyer of long term wealth. Your wonderful business might be compounding at 20% while someone else’s stock triples in 6 months, and suddenly an outcome that should make you wealthy begins to feel inadequate. The market doesn’t always need to make your company fail to separate you from it, sometimes it only needs to show you something more exciting. A mature investor eventually learns to watch someone else make a fortune without feeling required to participate. There will always be a stock outperforming yours and an opportunity you somehow missed, but you do not need to own everything that goes up. You only need a few things you understand well enough to stay with when staying becomes difficult. Of course, this doesn’t mean blindly holding forever, because businesses deteriorate and moats disappear. The real skill is distinguishing between a broken stock and a broken business, because a stock can break many times on its journey to becoming a 100 bagger while the business only needs to break once. Patience means surviving changes in price, while stubbornness means ignoring changes in truth. Perhaps that is why 100 baggers are so common on historical stock screens and so rare in actual portfolios. The difficulty isn’t simply identifying an extraordinary company, it is surviving the countless moments when selling it appears perfectly reasonable. Compounding asks us to be right about something important and then resist the endless temptation to interfere with it. So when searching for the next 100 bagger, perhaps the most important question isn’t whether you are smart enough to find one, but whether you have the temperament to experience one. Buying a great business can take an afternoon, while owning it long enough to change your life can take half your life. 🌹
48
72
22
691
72,360
Andres Andrade retweeted
true value creation comes from one of two things: creating desire or deleting dread. a ton of founders often confuse which one of these their business actually does. even worse, it often doesn’t do either.
35
20
2
410
20,360
Andres Andrade retweeted
I have been out on a remote construction site talking to the people doing the real work, moving the earth, bending the metal, and building their community. It is abundantly clear that AI is enabling them to spend more time doing their craft. More time building, less time buried in paperwork, timecards, invoices, and receiving. It is a force multiplier for the real humans shaping the world we all live in. Much of the AI narrative from SV feels completely detached from this ground truth. A young foreman pulled up a chair next to me because he was interested in AI and had heard I was from Palantir. He was absolutely full of brilliant ideas about how he could use AI to build more, do more, waste less, and keep his coworkers and community safer. The irony was not lost on me as he told me about the real-world issues with groundwork: you don’t know what you might uncover until you start digging. True ground truth. He was not thinking about how AI might replace his craft. He was thinking about how it could make him better at it. Sometimes you have to go sit down with humility and LISTEN to the people at the edge. Get dirty and sweaty. Be curious. The signal is incredibly strong for those willing to go looking for it. Everyone I talked to shared a desire to pass their craft on to the next generation. They loved working with their hands, and they took enormous pride in what they built. Their faces lit up as they showed me homes they had constructed, homes where their friends and family now live and raise families of their own. That is what made sovereignty feel tangible to me. Sovereignty is their choice and ability to own this culture and heritage that are so important to their craft. It is their ability to use AI on their terms, to strengthen what they have built, not outsource it to some faceless PE firm or let fear tactics scare them and their community out of what they know is right. Sovereignty is their right to own their destiny and their culture, and to reject one being pushed upon them. Institutional sovereignty stands on the shoulders of the people and cultures that make up the institution. We must do everything we can to protect and enable the individual cultures that make us more different, not more similar.
28
61
15
458
84,064
Andres Andrade retweeted
A new and possibly controversial perspective: In this video, I explain the sense in which generative AI trained by supervised learning is incapable of making novel discoveries. youtu.be/K5LAFEjTlBA The text of the speech: AI Creativity and Discovery Good day ladies and gentlemen. I regret that I am unable to be with you all today to engage in a back-and-forth discussion, but I am nevertheless pleased to be able to share with you, via this recording, some high-level thoughts about the current and future state of artificial intelligence, and in particular about AI’s relationship to science and mathematics, which is, as I understand it, the central focus of this meeting and of the SAIR Foundation. I would like to start with an old joke; I am sure you have heard it before. It is the one about the researcher whose work is being evaluated, and the review comes back, and says “This work is both novel and good. Unfortunately, the parts that are good are not novel, and the parts that are novel are not good.” My first point about AI is that this assessment applies exactly to large parts of AI as we know it today. Not all of today’s AI, but a large part of it. Pretty much all of what we mean by “Generative AI”---which includes large language models, and the images and video models, and even the new methods for learning world models. All of these AIs take large numbers of examples and produce a “model” which behaves similar to the examples, that is, which generates text like people, or images like artists or nature, and videos like we find on the internet. Don’t get me wrong, Generative AI can be extremely useful. No doubt about that. But the assessment of the joke still applies. These systems can produce output that is both novel and good, but not at the same time. In many ways this is just absolutely not a problem. When we ask an AI for an answer from the internet, or to summarize a document, we don’t want it to be novel. We are happy if the quality of the answer, the goodness, comes from the source material—from the people who wrote the document or the articles on the internet. If the AI’s answer is novel it means it is going beyond the source material, adding something beyond it. This is what we call “hallucinations”. In most cases, we don’t like it when the AI makes something up, when it adds something novel. One exception, of course, is when we are looking not for facts or reality, but for fiction and entertainment. We might ask for a bedtime story for a child, or an image based on existing images on the internet but which is nevertheless different and distinct from them. In these cases, it is never easy for us to know how creative the AI is actually being, as we do not know how close the AI’s story, poem, or image is to the source material. In a real practical sense we can not know this because the internet is too big, the possible sources that the AI may draw upon are too numerous. When we ask for a fiction or novelty, the AI can give it to us because its processing is in part stochastic. Every decision can go multiple ways and will go different ways and produce a different trajectory every time. The trajectory can be random—and thus novel—or it can be based on the training data—and thus “good” because the training data is good, sourced from people or reality. Thus, the trajectory is either novel or good—based on randomness or based on data—but never both at the same time. Really, I think it is okay if the output of Generative AI is never good and novel at the same time. For the researcher in the joke this is a devastating criticism, but for most things it is not, and for Generative AI it is not. Generative AI is meant to be a mimic. This is what supervised learning is for. Generative AI can be extremely useful, even when it just mimics, if it is faster, or cheaper, or smaller, or more customizable, or more copy-able, than the thing being mimicked. It is okay if Generative AI cannot be both novel and good at the same time. It is still a transformative technology. But it is a limitation. And remember we are here to use AI for science and mathematics, and for these areas the assessment of the reviewer in the joke is devastating. For these areas we need true creativity and discovery. Generative AI—or Mimicking AI—will never get where us there. For these we need something more, and indeed we have something more in other parts of AI. We have many AI systems which can give us more. We have AlphaGo with its world-changing move 37, or AlphaZero with its brilliant original chess-playing style. We have GT-Sophy that drives simulated racecars better than any human. We have AlphaFold and AlphaProof and Claude-Code, which have brought true advances in science, mathematics, and programming. We have RL-Lyft which optimizes the assignment of cars to passengers in the ride-hailing business. All these systems have found things that are both novel and good. And, truth be told, some language models have been augmented in ways that make them more than Generative AI based on supervised learning. All these systems have some additional features that make them capable of true creativity and true discovery. It is important for us to recognize what this is—and that it is not present in ordinary, garden-variety Generative AI. It is something that can not come from just supervised learning, from learning from examples. What is it? Well, it is a simple thing, a commonsense thing. It is not new. We have many names for it, but unfortunately none of them are very good names. I will call it Discovery. Basically, Discovery is just the idea of trying many things and seeing which of them work, then keeping those that worked the best. Evolution by natural selection works this way. The scientific method works this way. And just ordinary life and learning works this way. We try things and remember what works. What could be more obvious? In this behavioral case, psychology has two names for it— “instrumental learning” and “operant conditioning”—and in machine learning it is what we mean by “reinforcement learning”. We also see the idea of Discovery in planning and combinatorial search—anything that involves the idea of “generate and test”. The essence of Discovery is to combine three steps: 1. Variation, 2. Evaluation, and 3. Selective retention. Of course, I am not the first to say this. I am not the first to point out that this combination of steps is key to science, to evolution by natural selection, and to animal behavior. I think particularly of papers by Donald Campbell, by Daniel Dennett, and by Gary Cziko. What is new in my remarks is to directly relate the idea of Discovery to modern AI to help us see that it is not present in supervised learning or Generative AI—in particular, that Discovery is not present in backpropagation or gradient descent. Let me say explicitly what is missing from Generative AI. As we have remarked, these systems do have a stochastic aspect, so they do generate a variety of trajectories and behavior. What is missing is the Evaluation step. The generator was pre-trained by supervised learning, leaving no way at runtime to Evaluate what it generates. And of course without Evaluation there can be no Selective retention, and thus no Discovery. The variation can bring novelty, but without evaluation there is no Discovery, and arguably, no creativity. That is, I would say that creativity requires that the new things generated be Evaluated. Without evaluation, and retention of the best, there is nothing created. The novelty flickers into existence but, if its value is unrecognized, it flickers away and is lost. In many cases, Evaluation is done by people to make a discovery. As when we have Generative AI make many pictures for us, and then we pick the one that we like the best. The human+AI system completes the discovery. In many other cases, the Evaluation comes from a clear objective. Some moves lead to checkmate, some steps lead to a proof, some actions result in high reward, some genotypes make more copies, some theories explain the data better. Some prefer the Variation step to be called Blind variation, where “blind” here means that it is uninformed, a shot in the dark. It does not need to be completely uninformed; a good scientist does not select theories to test at random. But neither can it be completely informed and determined. There must be some uncertainty about where the answer lies in order for there to be a discovery. In practice, the variation is partly informed and partly blind, but it is the blind part that corresponds to the discovery. Now let us briefly go all the way to modern deep learning, to the backpropagation algorithm. At first it might seem that backpropagation is incapable of discovery because it is deterministic and thus incapable of variation. But this is not correct. The weight updates of backprop are deterministic, but the weights are initialized to small random values. The random initialization is often downplayed, but in fact it is a necessary form of variation; it must be done properly to get good performance. In backprop this Variation is done once, at network initialization, so its effect is temporary, and later the network may lose its ability to learn. This is the weakness of deep learning that is alleviated with a new algorithm that my group presented in Nature a couple of years ago. Our “continual backpropagation” made one small change: every so often a less-used neuron would be re-initialized to small random weights. This allows the variation to continue and plasticity to be retained. Although there is much more to be said about Creativity and Discovery, this is the key point: they are more than supervised learning, more than pattern recognition, more than prediction, and more than world modeling. Those things are important, but they alone will not bring us to discovery. Discovery requires Evaluation from a person or from an explicit goal, and only in the latter case will we attain full autonomy. So that is my call to arms. If we want the full power of AI scientists, then we should share the goals with them so they can create, evaluate, discover, and in these ways fully participate in achieving the goals. Let’s be bold! Let’s fully automate Creativity and Discovery!
103
282
104
1,679
697,812