Alex Siegenfeld

MIT Physics PhD studying the application of physics to political, economic, and social systems

[Alex Siegenfeld @AlexSiegenfeld]

Intro to complex systems science: https://t.co/KOh8qhEP7I

-assumes only high school math/science background so that it's accessible to anyone interested in systems and society

[Eitan Borgnia @EBorgnia]

We found compacting agent context on cache misses reduces token costs by over 50%.

The graphic is a real Fable 5 trace from our CTO, where we simulate cost would before/after compacting after periods of over 5 mins of inactivity.

By keeping context within 128k-256k tokens, spend goes from being quadratic to linear in the number of turns.

The only problem is self-summary takes 1-2 minutes on big traces! We trained an absurdly fast compact model to get the cost savings without harming the user experience.

Relace Compact is a modified LLM designed for token-level classification, which means we can run it at 50k tok/s and an order of magnitude lower than cache read cost ($0.20/million input and $0.20/million output tokens).

Try it out in Jacq, or visit our docs to integrate into your own product!

[Alex Siegenfeld @AlexSiegenfeld]

My paper with @yaneerbaryam featured on the cover of Entropy

[Entropy MDPI @Entropy_MDPI]

📢#mdpientropy New Issue Released!
Volume 27, Issue 8 (August 2025) – 109 articles
https://t.co/MDidb0DjRT

The #CoverPaper is "A Formal Definition of Scale-Dependent Complexity and the Multi-Scale Law of Requisite Variety"
https://t.co/Tcnn9Jb2mX

[Alex Siegenfeld @AlexSiegenfeld]

@wolftivy Ashby's law of requisite variety and its multiscale generalization (see eg https://t.co/KOh8qhEhia for a review) seem potentially relevant here

[Daniel Kang @ddkang]

I'm excited to announce our library zkml for trustless machine learning! After months of hard work, we've supercharged its performance & expanded its capabilities. Now, zkml achieves 92% accuracy on ImageNet!

Blog post: https://t.co/NQ88FIXQxt
GitHub: https://t.co/DwtT9uhFQ3

1/

[Alex Siegenfeld @AlexSiegenfeld]

@tomaspueyo Great thread! We analyzed the statistical power of many of the mask studies and came to the same conclusion: "The studies that did not find statistically significant effects prove only that masks cannot offer protection if they are not worn." https://t.co/CwbDQ5QwIt

[Alex Siegenfeld @AlexSiegenfeld]

Our paper uses compartmental models as an example to highlight some of these more general modeling principles and pitfalls

[Alex Siegenfeld @AlexSiegenfeld]

New paper w/ @praty1931 and @yaneerbaryam looks at how many epidemic models focus on details that don't matter while ignoring those that do.

This has led to erroneous conclusions (e.g. that travel restrictions don't work, a widespread belief in early 2020)

[Yaneer Bar-Yam @yaneerbaryam]

New paper: Modeling Complex Systems: A Case Study of Compartmental Models in Epidemiology

"Our purpose is to show how assumptions can constrain model outcomes to a narrow portion of the wide landscape of potential epidemic behaviors.

@AlexSiegenfeld

https://t.co/I4XYmgVLVo

[Alex Siegenfeld @AlexSiegenfeld]

Flawed modeling also led to the initial mistaken conclusions regarding masks: https://t.co/PU1n0Omfvq

[Alex Siegenfeld @AlexSiegenfeld]

Instant-runoff voting (a form of ranked-choice voting) eliminates the need for primaries.

It does so by simulating a series of runoff elections without the need for people to vote more than once

More info: https://t.co/dE3YvmXdb1 @fairvote

[Alex Siegenfeld @AlexSiegenfeld]

To the extent that it works, this problematic strategy takes advantage of negative representation within our system: pushing political opinion to the right can result in the outcome moving to the left.

https://t.co/9yc58b4k0K

[NPR @NPR]

This election cycle, you may have noticed a strange phenomenon of political advertising: Democrats paying for ads that support fringe Republican candidates. https://t.co/M0gkxBv9O5

[Alex Siegenfeld @AlexSiegenfeld]

As shown in our research (https://t.co/srXvqLRpyG), the negative representation is in part a consequence of our primary system

[Sihao Huang @Huang_Sihao]

Preprint with @AlexSiegenfeld and @StatModeling - How Democracies Polarize: A Multilevel Perspective

Almost all democracies employ elections at various scale, but how do the nested elections interact with heterogenous political geographies? 🧵

https://t.co/MbcjsghWvM 1/

[Andrew Gelman et al. @StatModeling]

Statistics and science reform: My conversation with economist Noah Smith https://t.co/DqQmvH7CbZ

[Alex Siegenfeld @AlexSiegenfeld]

@orion7710 @MIT_Physics @ScienceMIT MIT does allow students to wear masks, although it no longer requires it

[Tomas Pueyo @tomaspueyo]

Since the Bucha massacre was made public, Germany has paid ~$1.5B in gas to Russia

Yet 🇩🇪 will still close its nuclear power plants in 2022. Why? I dug up the details. Not pretty.

The only conclusion is that 🇩🇪 would rather kill nuclear than fight Putin & defend Ukraine
1/

[Spencer Greenberg 🔍 @SpencrGreenberg]

The exact wording in polls, surveys and psychology studies matters more than most people seem to realize.

This is a big deal if you want to learn from polls and academic papers, or if you conduct studies yourself.

Here are some dramatic real examples:

[Andrew Gelman et al. @StatModeling]

“Masks block some fraction of viral particles from dispersing from those who are infected and from infecting those who are susceptible . . . studies that did not find masks to be effective had limited statistical power and therefore do not imply that m https://t.co/4W9xuDCGB5

[Hertz Foundation @HertzFoundation]

New research from Hertz Fellow & @MIT PhD student @AlexSiegenfeld explains why the effectiveness of surgical #masks in preventing respiratory infections has been underestimated. https://t.co/jqCQQ4SPP6