SBIRs

• Was able to log in today!
• 9:00 standup – nope
• Holiday lunch – fun! Long drive though
• 2:30 IRAD meeting – done. Maybe good?

Hike in the snow yesterday:

Nice post about the disintegration of the useful internet

Tasks

• Showing at 3:15. Get the car washed while waiting. Nope, Wednesday now
• Hotels! Lyle Lovett and John Hiatt – done

SBIRs

• Write a Gutenberg parser that splits the beginning and ends off books and looks for weird formatting (e.g. lists of numbers, a single pix, etc.) The output should go in the ‘processed’ folder. Then use that folder to create csv (pickle?) files of each book with embeddings in them. Nope. Can’t log in because CMMI has killed the minds of IT. It’s a “known problem.” They are “working on it.” Soooooooo angry.
• After waiting for 5 hours, I can now log into my laptop. It was easy, but required Hidden Knowledge. Which didn’t need to be Hidden.
• Put together the white paper proposal for Neural Network Learning Capacity on Parametric Function

Language models are persuasive – and that’s a good thing

• Two new studies were published last week, in two of the most prestigious scientific journals, and sharing an overlapping authorship list. Persuading voters using human–artificial intelligence dialogues (Lin et al, 2025) was published in Nature, and tested the effect of AI dialogues on voters’ preferences in the context of elections in the US, Canada and Poland. The levers of political persuasion with conversational artificial intelligence (Hackenburg et al, 2025) was published in Science, and tested the effect of AI dialogues, when driven by different models and different prompting strategies, on attitudes of UK adults across British political issues.

[2510.11789] Dimension-Free Minimax Rates for Learning Pairwise Interactions in Attention-Style Models – this is lit review for the MDA study

Tasks

• Send off BLOG@CACM – done
• Bills – done
  • Pay Edwins – done
  • Pay Just Landscaping – done
• Chores – done
• Dishes – done
• MVA – trying to get my eye exam info sent over. Otherwise I need to schedule an appointment, which doesn’t seem terrible
• Groceries – done
• Get together with Terry for tix? – Sunday?

Bridging Social Media and Search Engines: Dredge Words and the Detection of Unreliable Domains

• Proactive content moderation requires platforms to rapidly and continuously evaluate the credibility of websites. Leveraging the direct and indirect paths users follow to unreliable websites, we develop a website credibility classification and discovery system that integrates both webgraph and large-scale social media contexts. We additionally introduce the concept of dredge words, terms or phrases for which unreliable domains rank highly on search engines, and provide the first exploration of their usage on social media. Our graph neural networks that combine webgraph and social media contexts generate to state-of-the-art results in website credibility classification and significantly improves the top-k identification of unreliable domains. Additionally, we release a novel dataset of dredge words, highlighting their strong connections to both social media and online commerce platforms.

Egalitarianism is not Equality: Moving from outcome to process in the study of human political organisation

• Many traditional subsistence groups have been described as ‘egalitarian societies’. Definitions of ‘egalitarianism’, especially beyond anthropology, have often emphasised equality in resource access, prestige or rank, alongside generalised preferences for fairness and equality. However, there are no human societies where equality is genuinely realised in all areas of life. Here we demonstrate, empirically, that nominally egalitarian societies are often unequal across seven important interconnected domains: embodied capital, social capital, leadership, gender, age/knowledge, material capital/land tenure, and reproduction. We also highlight evidence that individuals in nominally egalitarian societies do not unfailingly adhere to strong equality preferences. We propose a new operational framework for understanding egalitarianism in traditional subsistence groups, focussing on individual motivations, rather than equality. We redefine “egalitarianism” societies as those where socio-ecological circumstances enable most individuals to successfully secure their own resource access, status, and autonomy. We show how this emphasis on self-interest — particularly status concerns, resource access and autonomy — dispels naive enlightenment notions of the ‘noble savage’, and clarifies the plural processes (demand-sharing, risk-pooling, status-levelling, prosocial reputation-building, consensus-based collective decision-making, and residential mobility) by which relative equality is maintained. We finish with suggestions for better operationalizing egalitarianism in future research.

Portugal is having a general strike today

Tasks

• Set up shared expense spreadsheet
• Mtn bike today?
• LaTex to HTML converter – installed
  • Let’s see how it works? – pretty good!
• Pay Edwins
• Pay Just Landscaping

SBIRs

• 9:00 standup – done
• 4:00 ADS weekly – done. Need to write up a proposal to explore the theoretical spaces of D2A

Tasks

• Got a nice response from Bloomsbury, which might go somewhere
• Work on BLOG@CACM post – finished!
• Lunch with Aaron? Should be warm enough to ride there – mostly, a bit of rain on the return but not too bad.
• Alden meeting? Yup. Jimmy’s a dad!

Self-organized Collapse of Societies

• Why are human societies unstable? Theories based on the observation of recurring patterns in historical data indicate that economic inequality, as well as social factors are key drivers. So far, models of this phenomenon are more macroscopic in nature. However, basic mechanisms at work could be accessible to minimal mathematical models. Here we combine a simple mechanism for economic growth with a mechanism for the spreading of social dissatisfaction. Broad wealth distributions generated by the economic mechanism eventually trigger social unrest and the destruction of wealth, leading to an emerging pattern of boom and bust. We find that the model time scales compare well with empirical data. The model emphasizes the role of broad (power law) wealth distributions for dynamical social phenomena.

Tasks

• Got a response from the hotel!
• Worked on BLOG@CACM post

SBIRs

• More setup. Done!
• Make sure Overleaf works – doesn’t!
• Also got the Alienware set up again, before I turn it into a Linux box

Tasks

• Call MSC -done
• Sent a note to BLOG@CACM on the inverse surveillance capitalism blog post
• 6:30 showing -done

SBIRs

• Sprint review – done
• Sprint planning – done
• Fix VPN – done, but broke the sw manager
• Change password – done
• Maybe fix the process_data.py script? Seems to have worked, actually! Just no helpful output

Just checked his own social media channel, and our president seems to have forgotten

It strikes me that we can use dynamic time warping to compare trajectories

I think the process_data.py code is hung. Will need to fix that tomorrow

Tasks

• Checked the keybox, and it appears to be broken. Good nibble yesterday though
• Email to hotel
• Bills – done
• Pay Barbara – done
• Chores
• NO YARDWORK BECAUSE IT’S SNOWING
• Leave for Suz’s at 1:00-ish – done! Fun! Yum!

SBIRs

• Struggled to get the VPN working so I can check on the instance. Nope. Put in a ticket as around 8:30. If this goes on, I’m going to have to rework my dev environment and will probably need to put together a story for that. Got a response at 4:09 pm on how to get a new cert installed. Not the best use of a day.

Tasks

• Bennie and Phil’s trip to Terry’s
• Drivers license?
• Ping KP at 8:00

SBIRs

• See if I can get the clustering trajectories to run along the time axis as well
• Monitor Gutenberg download – seems to have broken in the AWS? Restarted.
• The index2vec looks pretty similar to the straight embeddings. Not sure if this is the way I want to go or not:

Index2Vec embeddings look a lot like sentence embeddings, but narrower. Maybe

Tasks

• 10:30 – 11:00 showing

SBIRs

• I’m trying to download the gutenberg corpus, but my instance appears to be broken. First, I couldn’t SSH in at all. After restarting, I can, but all my project directories are missing.
• Because of that, I decided to download locally and figure out how to move the data later. Because rsync isn’t available on Windows, I’m running in WSL. Which works, but not in the company VPN. Since I need the VPN to log into my instance, I can’t run it on my company box. Instead I set it up on my dev laptop, which is happily cooking along. JFC.

LLM stuff

• 2:30 Biweekly meeting – done. Shimei suggested using the 2D embeddings and using the Z-axis for time, which looks fantastic!

Paragraph-level embedding

Sentence-level embedding

Got a lot done on P33 on my cruise!

Tasks

• Showing 5:00 – 6:00

SBIRs

• Pinged Tivern about dates
• Chatted with Aaron about the project
• Need to see if I can pull down the Gutenberg corpora – started! Big!

ZZZzzzzZZZZzzzz

Cycling in Mauritius is perfect for cyclists who love beautiful, tropical islands and warm year-round cycling conditions.

Tasks

• A bit of house cleaning and demo prep – and leaving for hours! Got to see the first episode of Andor though, and worked out some travel plans
• Organize conclusions
• Cycling in Mauritius is perfect for cyclists who love beautiful, tropical islands and warm year-round cycling conditions.

SBIRs

• Figure out where I was!

Happy Tday to those who celebrate!

Early science acceleration experiments with GPT-5

• AI models like GPT-5 are an increasingly valuable tool for scientists, but many remain unaware of the capabilities of frontier AI. We present a collection of short case studies in which GPT-5 produced new, concrete steps in ongoing research across mathematics, physics, astronomy, computer science, biology, and materials science. In these examples, the authors highlight how AI accelerated their work, and where it fell short; where expert time was saved, and where human input was still key. We document the interactions of the human authors with GPT-5, as guiding examples of fruitful collaboration with AI. Of note, this paper includes four new results in mathematics (carefully verified by the human authors), underscoring how GPT-5 can help human mathematicians settle previously unsolved problems. These contributions are modest in scope but profound in implication, given the rate at which frontier AI is progressing.

CIFAR10 hyperlightspeedbench is a neural network implementation of a very speedily-training network that originally started as a painstaking reproduction of David Page’s original ultra-fast CIFAR-10 implementation on a single GPU, but written nearly from the ground-up to be extremely rapid-experimentation-friendly. Part of the benefit of this is that we now hold the world record for single GPU training speeds on CIFAR10, for example.

What we’ve added:

• custom architecture that is somehow even faster
• way too much hyperparameter tuning
• miscellaneous architecture trimmings (see the patch notes)
• memory format changes (and more!) to better use tensor cores/etc
• dirac initializations on non-depth-transitional layers (information passthrough on init)
• and more!

What we’ve removed:

• explicit residual layers. yep.

This code, in comparison to David’s original code, is in a single file and extremely flat, but is not as durable for long-term production-level bug maintenance. You’re meant to check out a fresh repo whenever you have a new idea. It is excellent for rapid idea exploring — almost everywhere in the pipeline is exposed and built to be user-friendly. I truly enjoy personally using this code, and hope you do as well! 😀 Please let me know if you have any feedback. I hope to continue publishing updates to this in the future, so your support is encouraged. Share this repo with someone you know that might like it!