Wow

Tasks

• Bills – done
• Dishes – done
• Chores – done
• Truck – done
• Library – done
• Groceries – done
• Maybe this? Nope. Couldn’t get out there before 4:30, and traffic on Fridays is bad enough to make me question my life’s choices
• FCC
• Download papers to review – done

SBIRs

• 3:00 MDS
• Maybe some BS/WG?

Steve Bannon predicts Trump will issue national security declaration on elections in late August

• The third step is for Trump to issue an executive order that could attempt to force voters to prove their citizenship or create a national voter ID law, both of which are aspects of the SAVE America Act, which has stalled in Congress. The scheme’s supporters have also suggested that the executive order could give Trump “king-like powers” to ban the use of voting machines or scale back access to early voting and mail-in or absentee ballots ahead of the midterms. 

Tasks

• Need to update the article so that it says that Surveilence Capitalism has grown beyond the original descriptions to include X and Y,. And when viewed through…. Done

SBIRs

• 9:00 Standup? Done
• 12:00 Demos – done
• 2:30 BD – done
• 3:00 Sprint planning – done
• 3:30 Tech summit stuff – done
• 4:00 MDA – Friday

The Heresy of Vector Embeddings: Early Modern Manuscript Verse and LLMs between Meaning, Form, and History

• The European Research Council-funded project “STEMMA: Systems of Transmitting Early Modern Manuscript Verse, 1475 to 1700” computationally models the manuscript circulation of early modern poetry to gain macro-level insights into the overlapping contexts in which it was written, read, and copied. To facilitate STEMMA’s ultimate goal of modeling the circulation of manuscript poetry as a large-scale network, the project began by combining and reconciling five of the most authoritative sources of relevant data using staged LLM-assisted pipelines. These pipelines were effective on a large proportion of the data, but they struggled in relation to particular features of early modern poetics, including imitation, translation, and unique works. This paper considers how general models’ shortcomings in relation to early modern poetry can be considered a feature rather than a bug, pointing up new insights into poetics by defamiliarizing them. It considers productive intersections between literary theory and criticism and recent scholarship on LLMs and suggests that two seemingly very different fields of study address what are, at core, rather similar fundamental questions.

Tasks

• Bank stuff for Sande – done?
• Truck + library? Not yet
• FCC – look for classes? They start Aug 22
• Interactions article – Reading Capitalism’s New Clothes | Evgeny Morozov
• 3:30 UMBC – that was fun. Went everywhere

SBIRs

• 2:00 EL – maybe progress?
• Slides for sprint review – done. Close stories – done. Write stories
• Create a new project for the BS/WG app? Maybe just a folder in the current for now. Made a tar file for the template, which works just fine!

Agentic AI’s OODA Loop Problem

• The Source of the Problem The fundamental problem is that AI must compress reality into model-legible forms. In this setting, adversaries can exploit the compression. They don’t have to attack the territory; they can attack the map. Models lack local contextual knowledge. They process symbols, not meaning. A human sees a suspicious URL; an AI sees valid syntax. And that semantic gap becomes a security gap.

Tasks

• Truck?
• FCC – done
• Library, if it’s not raining – RAIN
• Interactions article – Reading Capitalism’s New Clothes | Evgeny Morozov

SBIRs

• 9:00 standup
• Reviewed EL’s work, which missed a lot, and the slides were not terribly helpful. We’ll try again tomorrow
• PySide6 – walk through the running QUI code. Seems to be nice and clean

• Flailed around a lot at getting Qt Creator to work. Close, but not enough
• Stripped out as much as I could to get a qml-template framework to use for starting projects

Tasks

• Truck – dropped off
• FCC
• Interactions article -worked on it a bit. Reading Capitalism’s New Clothes | Evgeny Morozov

SBIRs

• Password? Done
• PySide6 = progress
• Good meeting with Aaron on WG/BS

Europol flags 4,340 ‘horrific’ URLs linked to The Com

• Europol and its partners’ investigators flagged 4,340 “horrific” URLs for removal over several weeks in June and July as part of an ongoing crackdown on The Com (short for community), a loosely knit network of online groups whose young members participate in a range of illicit activities. These range from hacking, swatting, and digital extortion to real-life shootings, stabbings, and other physical violence.

Tasks

• ICTAI paper bidding – done
• Interactions article
• Fidelity -done
• Bills – done
• Chores – done
• Monitor – done
• Dishes – done

Tasks

• Interactions article – progress
• Get to Know Hashawha—Beginner MTB Group Ride, Thu, Jul 23, 2026, 5:30 PM | Meetup = done! Tired! Accidentally wound up on the intermediate ride

SBIRs

• 9:00 standup – done
• More PySide6 – done
• 4:00 Tim (rework spreadsheet!) – done, but cancelled

Wednespoop

Tasks

• Alt text for the book – done
• Test for Sande – done
• TE?
• More on the Interactions paper. Some nice text from Paul Krugman here
• The rise, the fall, and the resurrection of Iceland
  • This paper documents how the Icelandic banking system grew from 100 percent of GDP in 1998 to 9 times GDP in 2008 when it failed. We base the analysis on data from the banks that was made public when the Icelandic parliament lifted among others bank secrecy laws to investigate the run up to the financial crisis. We document how the banks were funded, and where the money went with a comprehensive analysis of their lending. The recovery from the crisis is based on policy decisions which in hindsight seem to have worked well. We will analyze some of these policies, including emergency legislation, capital control, alleviation of balance of payment risks and preservation of the financial stability. We also estimate the output costs of the crisis, which was about average relative to the 147 banking crisis documented Laeven and Valencia (2012) and the 100 banking crisis documented by Reinhart and Rogoff (2014). Our computation of the governments direct costs, reveals that the recently concluded negotiation with foreign creditors may even leave the Icelandic government in net surplus as a consequence of the crisis, although there is still some uncertainty about the ultimate cost and our benchmark estimate is a cost corresponding to 5 percent of GDP. We summarize several lessons from the episode.

SBIRs

• Continue PySide6 tutorials – made good progress. I got the designer running and was able to set up the code that called it using Gemini rather than the example, because for some reason the default was a QDialog. But the AI code worked perfectly, which the exception of pointing it at the folder with the generated code:

import sys
from PySide6.QtWidgets import QApplication, QDialog
# Assuming the code you provided is saved in a file named 'hello1_ui.py'
from designer.ui_hello1 import Ui_Dialog


# 1. Create a wrapper class that inherits from QDialog
class MyDialog(QDialog):
    def __init__(self):
        super().__init__()
        # 2. Instantiate the generated UI class
        self.ui = Ui_Dialog()
        # 3. Setup the user interface on this dialog instance
        self.ui.setupUi(self)

        # Optional: Connect your "WriteText" button to a function
        self.ui.WriteText.clicked.connect(self.handle_write_text)

    def handle_write_text(self):
        self.ui.textBrowser.append("Hello World!")


# 4. Main execution block to run the app
if __name__ == "__main__":
    app = QApplication(sys.argv)

    # Instance your dialog class
    dialog = MyDialog()

    # .exec() blocks execution, opens the window, and returns the result code
    result = dialog.exec()

    if result == QDialog.Accepted:
        print("User clicked OK")
    else:
        print("User clicked Cancel")

    sys.exit(0)

• Here’s a nice StackOverflow thread on Snipping (includes cross platform)
• Installed OS and IDE updates and everything still works! Reset display to Wayland as well.

Making Kin with the Machines · Journal of Design and Science

• Kānaka maoli (Hawaiian people) ontologies have much to offer if we are to reconceptualize AI-human relations. Multiplicities are nuanced and varied, certainly more aesthetically pleasurable than singularities. Rather than holding AI separate or beneath, might we consider how we cultivate reciprocal relationships using a kānaka maoli reframing of AI as ʻĀIna. ʻĀIna is a play on the word ʻāina (Hawaiian land) and suggests we should treat these relations as we would all that nourishes and supports us. 

Tasks

• More on the Interactions article – progress!
• Library run – done
• Groceries – tomorrow. Don’t feel like dealing with the rain.

SBIRs

• 9:00 standup – done
• Working PySide6 – progress

[2607.15267] Pretraining Data Can Be Poisoned through Computational Propaganda

• Poisoning pretraining data can introduce harmful behaviors to LMs that are difficult to detect and mitigate. Prior work on poisoning pretraining data has largely exploited established data sources such as Wikipedia, which do not represent the large scale and heterogeneity typical of pretraining corpora, and has ignored the interaction between poisoned data and data curation pipelines. We demonstrate that poisoning attacks on pretraining data are feasible beyond this limited setting through an existing web-scale content injection mechanism: public discussion interfaces. Additionally, to measure whether malicious content is included after web crawling and data curation, we introduce HalfLife, a novel analysis for estimating adversarial content inclusion in web-crawl based LM training data. We use HalfLife to explore the feasibility of poisoning pretraining corpora at web scale through open discussion interfaces. Our analysis demonstrates the importance of estimating whether poison injections are included in pretraining data, and establishes third-party webpage content as a possible vector for attacking language model pretraining.

Tasks

• Work on the Interactions paper some more. Read the Mann papers too.

SBIRs

• PySide6 tutorials, which are here
• 11:00 Orest

Overload 2: Main Draft Report

• Since September 2024, the authors have collected over 700 targeted emails alongside content disseminated via Telegram, X, Bluesky, and, most recently, TikTok. A core tactic of the operation remains the direct targeting of media, fact-checkers, and researchers to overload them with specific content to fact-check. The network creates falsified narratives through manipulated content, and now increasingly AI-generated materials, for the purpose of soliciting engagement with fact-checkers. Alas, some publishers continue to routinely debunk individual falsehoods related to the operation without framing them within the broader context of Operation Overload.

Tasks

• Apply to FCC – done
• MORE ride on Thursday!
• Suz at gallery – done
• Work on Interactions paper – progress!
• Got some good TE time

Tasks

• Bills – done
  • Sande – not yet
• Chores – done
• Dishes – done
• Pedals on the Brompton – done
• Ritchey back together – done
• Library
• Laundry?
• Dump run

Lots of meetings today and still hot. Going to go for an early ride, I think

Tasks

• ULis passwords are working for the VPN and webmail. I guess it just needed time to ripple through
• Groceries – done

SBIRs

• Try to get the next PySide6 tutorial in? Got a little in!
• 9:00 Standup – done
• 11:00 Orest – Monday
• 2:00 Matt – straightforward
• 4:00 Tim – need to put together a spreadsheet based on my rate that will provide a starting point based on the funds available, the loaded rate, and available working days. To begin, Dec 10, and Feb 28

Generative AI Is an Engineering Disaster – The Atlantic

• The problem with generative AI, in the industry’s own jargon, is that it does not scale. The cost of growing from, say, a thousand users to a million is a key factor that venture capitalists examine when they evaluate start-ups. They want to see that the cost of adding each new user decreases over time, so that the company can support millions of users and make increasing profits. This is achieved partly through the careful engineering of computer systems that can efficiently handle more users who want to post photos, hail Ubers, or stream music.

Tasks

• ULisboa password 2nd try. This may be a problem. Hmm got the password reset, but things still don’t work.
• Book
  • Alt-text
  • Anything else?
  • Look at interactions article before:
• 3:30 meeting with Jimmy & Shimei – fun!

SBIRs

• 9:00 algorithm meeting
• PySide6 tutorials – rough start with drivers and hardware, but things are working now

Tasks

• Sande – started
• ULisboa password – first try

SBIRs

• 2:00 Sprint planning -done
• Finish slides – done
• Wrote an RFI response in 90 minutes