Alexander Chemeris

Building Digital Infra Robots @ time-series world models and hardcore telco telemetry. ex-@fairwaves CTO, #hackerspace co-founder.

[Alexander Chemeris @chemeris]

The gap between Time-Series analysis and the rest of AI is so painful that I decided to research in public, publishing experiments, thoughts, and interesting papers to stir things up a bit.
Comment or follow if you believe the future is just the next step in time series.

[Alexander Chemeris @chemeris]

Awesome work!
Just curious, have you tried using regularization like SIGReg/VISReg instead of stop-grads?
In our LeNEPA, it significantly improved downstream task performance (https://t.co/3aEF6U0BYS). The challenge might be that your embeddings form a 2D shape instead of pure 1D (time-wise), as in our LeNEPA, though.

[Alexander Chemeris @chemeris]

@SmartPig_Joe It's impossible to keep a "scientific secret" inside a frontier lab, given the churn and the general pace of the industry. It's 100% compute + data + org capacity

[Randall Balestriero @randall_balestr]

We are two weeks away from the third World Modeling Workshop!
- check https://t.co/7fFxfp6M7q for all the latest info/links
- livestream and recordings will be available for free to everyone, no registration required
- a few surprises will be announced during the opening remarks

[Alexander Chemeris @chemeris]

I wonder when we'll see AI designing communication protocols.
Why? Because it requires imagining all possible failure modes and the consequences of various ways to mitigate them. I.e., it depends on a strong world model (implicit or explicit).

[Alexander Chemeris @chemeris]

A big difference between robotics and infra management tasks is that robotics use specific image as the target, while in infra management there is no single target - you solve a multi-dimensional optimization task.

Small change that renders most benchmarking approaches useless.

[Alexander Chemeris @chemeris]

@giffmana Since Pascal?

[Alexander Chemeris @chemeris]

I thought we're past the "are world models useful" and into the "do we need explicit/modular WMs, or implicit WMs that exist inside LLMs are good enough"?

Explicit WMs are more efficient to train and serve, allows MPC-style control loop harness, more control over the objective. And most importantly - allow you to work with dynamic systems, changing in real time.

LLMs are, well, already out there. Efficient for some tasks, grossly inefficient for others. You can't (explicitly) control the amount of computation/WM rollouts, as it's baked into the model (except for the reasoning effort), but for some tasks it's ok - where you're fine with working on a static object.

With the recent work on controlling model behavior without explicit action inputs, the difference seems to blur even more - see the works of @thomas_fel_ @GoodfireAI and recently https://t.co/HRvMSjOKaZ by @mathemagic1an

PS World = data distribution you're training on. My understanding had been that @ylecun has always been taking about physical intelligence and that's where LLMs are not great at / inefficient?

[Lucas Beyer (bl16) @giffmana]

imma just highlight this part for @GaryMarcus and @ylecun because it's easy to miss: no tools no coding => no symbols, just AR LLM

[Alexander Chemeris @chemeris]

@annlaurin Я не про то, что всех учёных заменит AI, а скорее про то, что сейчас всё вертится в AI и там активно найм идёт кажется везде

[Alexander Chemeris @chemeris]

@ziqiao_ma Keep pushing!

[Alexander Chemeris @chemeris]

@giffmana @cbonnett I do say "orthogonal" irl 👀

[Alexander Chemeris @chemeris]

@MozarellaPesto Just this initial sketch r/n https://t.co/rynYLO7leW
We should have preprints with initial results out in the next month or two, fingers crossed. Happy to chat

[Alexander Chemeris @chemeris]

One of the coolest things to watch in modern AI is how even ninth- and tenth-graders (15-16 years old) can already publish decent conference papers and lead research at real AI companies.

[Alexander Chemeris @chemeris]

@HaiyuWu1 But actions need grounding, as they don't exist in vacuum. So it's either labels, or active world exploration, or at least inductive bias.
Otherwise the problem looks ill-defined, no?

[Alexander Chemeris @chemeris]

Great work on stochastic World Modeling with Flow Matching.

FlowWM paper (https://t.co/aiw2RwxDUl) shows that flow matching can generate multiple plausible future trajectories directly in frozen DINOv3 feature space, rather than collapsing uncertainty into one averaged prediction.

The result is strong on their branching-futures benchmark, F1 mode error falls from 17.1 for a matched deterministic predictor to 4.31. On Waymo, the predicted features also improve object detection and depth.

I've been thinking about replacing JEPA's point predictor with a conditional flow over future latent trajectories, so it's exciting to see evidence for the core mechanism.

What's still missing to make this useful for a wide range of industrial applications is calibration. Best-of-N tells us whether at least one sample matches reality. It does not tell us whether recovery has 60% probability and failure 1%—or whether the model gets those weights backwards.

Those probabilities are what we need to make decisions. They let us estimate expected cost, reason about tail risk, and compare how candidate actions shift the distribution.

[Alexander Chemeris @chemeris]

@ziqiao_ma @arcprize Yeah, I think people underestimate the difficulty starting from scratch, and also the importance of velocity

[Alexander Chemeris @chemeris]

@Delong0_0 That's unfortunate :) Thank you!