Alexander Mattick
Guest on Machine Learning Street Talk (MLST). 568 claims across 1 episode, Sep 2026.
Positions over time
Every pair of statements that was compared. Only a verified change counts as a change of mind; rewording, added nuance and different questions do not.
No two statements by this speaker have been compared yet.
Predictions, in their words
As stated, each with its moment. A date appears only where the speaker gave one.
“We don't need to research all these other complicated things, but instead if we just become really, really good at maximizing reward, then everything else becomes obsolete.”
Expects: The speaker says Sutton's view implies that becoming very good at reward maximization would make research into other complicated approaches obsolete. · no date given · not yet resolved
“That one of the big challenges is having doing effectively a sensing part in a reasonable way which also gives you relatively strong guarantees because if you, first of all, you don't get strong guarantees, you might crush someone.”
Expects: A robotic system without strong guarantees might crush someone. · no date given · not yet resolved · In a shared prediction of 3 →
“I'd rather spend that money to kind of fine tune my C_ and C_ G_ code for a couple of seconds than uh because it would give more effective speed.”
Expects: An industrial customer might prefer investing in C/C++ code optimization if it yields greater effective speed than buying an unreliable robot. · no date given · not yet resolved
“I'd rather spend that money to kind of fine tune my C_ and C_ G_ code for a couple of seconds than uh because it would give more effective speed.”
Expects: The company might prefer to spend the money fine-tuning its C and C++ code, which could provide more effective speed. · no date given · not yet resolved
“because ultimately you would expect something like this for a general model.”
Expects: General models will probably exhibit a similar pattern in which coarse structure emerges before finer details. · no date given · not yet resolved
“And of course, it might be that it's a one architecture just scales more nicely.”
Expects: Alexander Mattick allows that one model architecture might scale more efficiently than another. · no date given · not yet resolved
“Well, arguably we already have plenty of them. Uh the the the problem is more that to have something like physics have I don't think the entire field is set up to do that. And I think that it will not be set up to do that and Up until the point where things start slowing down.”
Expects: The machine-learning field will remain unprepared to produce a unified theory comparable to physics until the pace of progress slows. · no date given · not yet resolved
“But I have a feeling that we are going to hit some really, really hard limitations, not necessarily by the fundamental idea, but basically by the nature of how the real world works and kind of how diverse tasks can become. and how complex tasks can become.”
Expects: The speaker expects world models to face severe limitations due to real-world conditions and the diversity and complexity of tasks, while not necessarily being fundamentally flawed as an idea. · no date given · not yet resolved
“But I have a feeling that we are going to hit some really, really hard limitations, not necessarily by the fundamental idea, but basically by the nature of how the real world works and kind of how diverse tasks can become. and how complex tasks can become.”
Expects: Alexander Mattick suspects that world-model approaches will encounter hard limits due to real-world conditions and the diversity and complexity of tasks, though not necessarily due to the underlying idea. · no date given · not yet resolved
“Of course you still have some flexibility, right, because if you like let's say if every point starts pushing into the mode of the d of the density, at some point you would have a kind of pushing a p uh pushing away effect again. So you have a similar thing as a contrastive model, but now it is”
Expects: A repulsive effect can eventually emerge if all points move toward the density's mode, while the fixed-density approach retains flexibility. · no date given · not yet resolved
All 31 predictions
“uh con especially Bayesian inference gives you a lot more structure a lot more information than what we give it credit for. Like some product just decomposing a problem into conditional probabilities gives you it gets you like ninety nine percent of the way there, but it can already help guide the actual problem. And I think that something like all these agentic things could really benefit from actually thinking about this as like a Bayesian problem of kind of going out gathering data.”
Expects: Agentic systems could benefit from using a Bayesian approach in which they gather data. · no date given · not yet resolved
“If you don't use a if you use them for instance as a pre training task, I don't see a reason why not. I I don't see a reason why you should but I don't see a reason why it wouldn't work. The the question really is is if we end up looking like let's say like uh like it's a we we jump forward five years from now and LaCour's startup is doing whatever he he did, like they are doing currently research. Whether what they end up doing like they might still call it energy based models, but whether it's actually an energy based model As you and I would be thinking about this right now, I highly doubt that.”
Expects: The speaker highly doubts that, five years from the episode date, the startup's work will qualify as an energy-based model under the current understanding, even if it is labeled one. · 21 Sep 2031 – 21 Sep 2031 · not yet resolved
“The actual latents don't hold any information, but of course in practice if you enforce the fact that the change from let's say latent number one to latent number two should be really small, then you would expect something like this, where effectively like the big structure up here first, because on the big s because if we don't have the big structure it's kind of the small changes later on cannot reconstruct this. So”
Expects: Large-scale structure will emerge before smaller details when successive latent changes are constrained to be small. · no date given · not yet resolved
“Uh I would imagine that most that they say in models would structure it like this. This doesn't mean that you couldn't build a diffusion model which doesn't do that. I could imagine going and kind of enforcing effectively a constraint that says well you have to do the the small things first. I don't think it would be a good model. Uh you might need like let's say thousands of times more parameters to actually uh fit this. You you still would be uh you still would be able to”
Expects: A diffusion model constrained to produce small details before large-scale structure might require thousands of times more parameters to fit the data. · no date given · not yet resolved
“Uh I would imagine that most that they say in models would structure it like this. This doesn't mean that you couldn't build a diffusion model which doesn't do that. I could imagine going and kind of enforcing effectively a constraint that says well you have to do the the small things first. I don't think it would be a good model. Uh you might need like let's say thousands of times more parameters to actually uh fit this. You you still would be uh you still would be able to”
Expects: The speaker expects most diffusion models to produce large-scale structure before details, while acknowledging that a model could be designed to produce small details first. · no date given · not yet resolved
“So you would expect also that the higher information things to appear first, because have they they are the free wins. I don't think it has to be like that, but it sh but I would expect it to be like that. The same thing like a rectified flow. In a rectified flow I don't think it should be that surprising that if you basically do a linear interpolation of big things appear first.”
Expects: Higher-information features will probably appear first in diffusion models because they are easier gains, though Alexander Mattick does not consider this necessary. · no date given · not yet resolved
“uh explains this idea of overparameterisation while the parameteric uh with the parametric kind of says, well actual actually this overparameterisation might be bad, but we can exactly explain kind of how the the local architecture is. I would im however, I would imagine that in the long term, the functional analysis people they are going to be more useful just because they explain a specific architecture but they explain architectures in general. It can know a tangent kernel,”
Expects: The speaker expects function-space analysis to be more useful than parameterized analysis in the long term because it can apply across architectures, while parameterized analysis explains specific architectures. · no date given · not yet resolved
“Every proof works for literally anything but the idea of looking at this from a function space point of view works even if you have like let's say we all go back and do like MLP mixers now. We no longer do transform, we all do MLP mixers. It still would work from a theory point of view.”
Expects: Function-space analysis would remain applicable if architectures shifted from transformers to MLP mixers. · no date given · not yet resolved
“Uh just because kind of this have this loop of I make a prediction, then I go, I make an experiment. It I think is much more uh much more useful. And I think that the the big thing is that theory will never be able to explain literally anything, because uh it might be able to explain things on like a pure functional point of view. Uh let's say like the mean field people, they can completely explain if I have let's say uh a density composed of let's uh let's say a Gaussian mixture model.”
Expects: The speaker believes that theory will never explain literally everything. · no date given · not yet resolved
“What can always run into trouble with reinforced reinforcement learning is above there is a section where they have like killing an enemy hero actually gives a negative reward and the reason is and they have an explanation next to it is that killing an enemy hero gives you a lot of gold and there's already like a bonus for getting gold so if you also had like a zero positive reward for killing enemy hero you would sit into the middle of the map and just kill the enemy hero over and over and over again without making any progress.”
Expects: Removing the negative reward for killing an enemy hero would lead the agent to repeatedly kill that hero in the middle of the map rather than make progress. · no date given · not yet resolved
“Again it would imagine that if we these models generalize then yes we could imagine things like that especially if we do like planning in if we do planning a sufficiently large world model and assuming we can actually do as assuming Assuming compute does not play a role and kind of latency does not play a role and if we have all the data for the specific type of task,”
Expects: If models generalize, planning with a sufficiently large world model could enable the task under discussion, provided compute and latency are not limiting and the relevant task-specific data is available. · no date given · not yet resolved
“Again it would imagine that if we these models generalize then yes we could imagine things like that especially if we do like planning in if we do planning a sufficiently large world model and assuming we can actually do as assuming Assuming compute does not play a role and kind of latency does not play a role and if we have all the data for the specific type of task,”
Expects: If models generalize, planning with a sufficiently large world model could enable robots to perform tasks, provided compute and latency are not limiting and relevant task data is available. · no date given · not yet resolved
“I can imagine NMPC is ultimately a solution.”
Expects: Nonlinear model predictive control (NMPC) will ultimately provide a solution; Alexander Mattick considers this possible. · no date given · not yet resolved
“Because if it hasn't have this specific form, it doesn't fit into like a uh linear programme or whatever. And suddenly basically the the opinion starts to dim, the more you use it. And I have a feeling that the same thing is gonna be true for like the the world models. I think there's going to be a lot of like really really easy things like the low-hanging fruit which would are just going to be like grasped, and it might be that for many tasks this low-hanging fruit actually is already enough.”
Expects: Alexander Mattick suspects that world models will face limitations analogous to those of optimization-based control, though they may solve enough easy tasks to suffice for many applications. · no date given · not yet resolved
“Because if it hasn't have this specific form, it doesn't fit into like a uh linear programme or whatever. And suddenly basically the the opinion starts to dim, the more you use it. And I have a feeling that the same thing is gonna be true for like the the world models. I think there's going to be a lot of like really really easy things like the low-hanging fruit which would are just going to be like grasped, and it might be that for many tasks this low-hanging fruit actually is already enough.”
Expects: The speaker expects world models to handle many relatively easy tasks, and thinks that capability may already suffice for many tasks. · no date given · not yet resolved
“things like moving around balls on a table, like yeah, I can imagine that working. I can also imagine the same thing working with like an NMPC controller. I could say it gets stuck and use the other hand to move it around. An NMPC controller can come up with this, right? It would start planning and would figure out, okay, I need to do something like this, assuming that the model is accurate enough.”
Expects: Alexander Mattick thinks an NMPC controller could plan a simple manipulation task, such as using the other hand to move an object when stuck, if its model is sufficiently accurate. · no date given · not yet resolved
“I can also imagine the same thing working with like an NMPC controller. For instance, like the example, I could say it gets stuck and use the other hand to move it around. An NMPC controller can come up with this, right? It would start planning and would figure out, okay, I need to do something like this, assuming that the model is accurate enough.”
Expects: An NMPC controller could plan to use the other hand to move a stuck ball, provided its model is accurate enough. · no date given · not yet resolved
“Because deploying this in production production could mean kind of at home I don't think this is where we will see these robots at least not at first, at least not in the like short to medium term and probably also long-term I don't know but”
Expects: Alexander Mattick does not expect home robots to be deployed initially or in the short-to-medium term, and is uncertain about whether they will be deployed in the long term. · no date given · not yet resolved
“Because deploying this in production production could mean kind of at home I don't think this is where we will see these robots at least not at first, at least not in the like short to medium term and probably also long-term I don't know but”
Expects: The speaker does not expect home robots to be deployed initially or in the short to medium term, and is uncertain whether they will be deployed in the long term. · no date given · not yet resolved
“like if you, if you go to, let's say, BMW and you say, well, I have a robot, it does the right thing ninety percent of the time, the other ten percent of the time it does something else, they will, like they will say, no, I will not pay you that money for a robot,”
Expects: A company such as BMW would reject a robot that performs correctly only 90 percent of the time and fails or does something else 10 percent of the time. · no date given · not yet resolved
“like if you, if you go to, let's say, BMW and you say, well, I have a robot, it does the right thing ninety percent of the time, the other ten percent of the time it does something else, they will, like they will say, no, I will not pay you that money for a robot,”
Expects: Alexander Mattick argues that an industrial customer such as BMW would not pay for a robot that performs correctly only 90 percent of the time. · no date given · not yet resolved
Appearances
Claims per episode
- 21 Sep 2026Why Scaling Prediction Cannot Create Intelligence - Alexander MattickMachine Learning Street Talk (MLST) · guest