Tasks

• Groceries – done
• Break down Ritchey Tomorrow
• Logistics with SP – done
• Vote? Tomorrow

SBIRs

• My new Linux OS had no network adaptor support. Struggled for about 2 hours before figuring out how to roll back to the earlier version using GRUB on the boot screen
• Now I can start of adding clustering in. Running!
• Clustering optomization

The Puzzling Success of Overparameterization: Lottery Tickets or Escape Dimensions?

• Lotteries and tickets are often used as a didactical analogy to explain the success of overparameterized neural networks: “larger networks succeed because they more likely contain a well-initialized subnetwork that can learn the task in isolation, much like buying more tickets increases the chances of winning a lottery.” This explanation is intuitive but misleading: it suggests that subnetworks can be treated in isolation from the rest of the network. Following this reasoning leads to interpreting learning in wide networks as a multi-start optimization process, where gradient descent simply conducts a parallel search over subnetworks. We argue that this view is flawed since, among other reasons, winning tickets can be made to fail by perturbing the rest of the network. We put forward a more accurate intuitive picture for the success of overparameterization based on the geometry of loss landscapes: increasing width expands the set of available dimensions for optimization, making it easier to escape bad local minima. Moreover, as width grows, bad minima become increasingly rare relative to good minima. As the field grows mature, it is important to refine the analogies we use to explain foundational phenomena, such as the apparent redundancy of large networks, reconciling practitioners’ intuitions with modern theoretical insights.

Tasks

• Bills – done
• Chores – started
• Dishes – done
• Work on the GL/R post. Start adding in the end of R, where words transition to deeds and why that matter
• Library run – done
• 3:00 Fidelity – done

SBIRs

• Finish the embedding run and verify that everything worked. Then power down and do clustering Monday – done

Predictive Data Debugging: Reveal and Shape What Your Model Learns, Before You Train

• We introduce predictive data debugging: given a preference dataset, we can accurately predict which behaviors RL will amplify or suppress before you train, trace them back to the responsible data, and reshape the dataset and/or training process to prevent undesired effects.

Tasks

• Finish the GL/R post
• Good day to go up Hamburg – hot!
• Did the 2026 Form 1 thing!

SBIRs

• 9:00 standup – done
• Code up the full dimension reducer/clusterer – done and running!
• 4:00 ADS Meeting – cancelled

Looks like a rainy day. It was, but with a break in the middle!

Tasks

• More on the G/R post – progress, and I seem to have published the draft, which means I really need to finish it
• Library? Done. And another book showed up after
• Installed mirrors on the other road bikes

SBIRs

• The reducer code had some kind of memory leak and killed the IDE. Restarting where it left off. Done
• I’ll also need to re-run the entire code with a smaller table (100k) to be able to see which reducers to use – done. Interesting results, too!

• Code up the full dimension reducer/clusterer

Motor insurance frausters abusing AI to exaggerate claims

Tasks

• More G/L post
• Returned to sender

SBIRs

• Updated the doc and sent back to Tivern
• 9:00 Standup
• SEG lift stop work – done?
• 1:00 – 2:00 Discussion on WG – done
• 5:00 WG meeting? Done. Meh

My Moltbook post got published in the June edition of CACM!

Tasks

• Post office – need to try again
• Library – done
• Working on the Gaslight/Rashomon post

SBIRs

• Travel expenses – done
• Abstract got approved. Time to start thinking about slides

5 hours of driving for a one-hour meeting. I am so glad to be out of the car!

State media control influences large language models | Nature

• Millions of people around the world query large language models (LLMs) for
  information. Although several studies have compellingly documented the persuasive
  potential of these models, there is limited evidence of who or what influences the
  models themselves, leading to a flurry of concerns about which companies and
  governments build and regulate the models. Here we show through six studies that
  government control of the media across the world already influences the output of
  LLMs via their training data. We use a cross-national audit to show that LLMs exhibit
  a stronger pro-government valence in the languages of countries with lower media
  freedom than in those with higher media freedom. This result is correlational, so to
  triangulate the specific mechanism of how state media control can influence LLMs,
  we develop a multi-part case study on China’s media. We demonstrate that media
  scripted and curated by the Chinese state appears in LLM training datasets. To evaluate
  the plausible effect of this inclusion, we use an open-weight model to show that
  additional pretraining on Chinese state-coordinated media generates more positive
  answers to prompts about Chinese political institutions and leaders. We link this
  phenomenon to commercial models through two audit studies demonstrating that
  prompting models in Chinese generates more positive responses about China’s
  institutions and leaders than do the same queries in English. The combination of
  influence and persuasive potential across languages suggests the troubling conclusion
  that states and powerful institutions have increased strategic incentives to leverage
  media control in the hopes of shaping LLM output.

Tasks

• Chores – done
• Dishes – done
• Bills
• Ride up to Mike’s – done
• Suz at 2:15 – done

SBIRs

• Finish Q5 report – done
• 3:00 tagup meeting – cancelled
• 4:00 ADS – done

I got something published in CACM! Minimally Acceptable Systems: Tolerable at the Lowest Cost Possible – Communications of the ACM

Gemma 4 12B: The Developer Guide – Google Developers Blog

• 1.Native MacOS Apps: The mobile Google AI Edge Gallery is officially expanding to desktop platforms, running Gemma 4 12B offline, natively on Apple Silicon GPUs. It comes with a secure sandboxed Python execution loop to write, execute, and plot scientific charts inside the chat bubble. In parallel, the Google AI Edge Eloquent app on Mac launches support for Gemma 12B to power Voice Edit conversational inputs.

Hyperfascism’s sole imperative is to produce one horrible, transfixing image after another; the images themselves may be hyperreal, but the cost of creating them is all too real.

Tasks

• https://www.chaircaningbaltimore.com/contact-us/

SBIRs

• Q5 report – first draft is done
• Travel – done
• Dis something(?) for the Dahlgren trip

Source: Why AI Agents Fail in Production – by Paolo Perrone

Tasks

• 10:20 Dentist – done
• Send Sean updated book screenshots – done. Need to repackage the project

SBIRs

• 9:30 Meeting – technical problems. Became a 3:00 meeting
• Work on Q5 report. Need to finish COB Monday
• Travel

Tasks

• Some laundry – done
• Book edits – done

SBIRs

• Password? Can’t do anything. Filed a ticket. Fixed
• Travel
• 9:00 Sprint Review – done
• Stories – done and thrown out. we are on a fire drill for some reason
• 3:00 Sprint Planning

Sigh

Tasks

• Bills – done
• Dishes – done
• Chores – done
• Laundry
• MD 2026 Form 1 – mentioned this to Aaron. We will chat about it tomorrow
• Book stuff – not today

SBIRs

• 9:30 Meeting with Ron – done. More followup next Thursday. Also chatted about derivatives

Is 4chan running WH Comms?

Give your ideas some legs: the positive effect of walking on creative thinking.

• Four experiments demonstrate that walking boosts creative ideation in real time and shortly after. In Experiment 1, while seated and then when walking on a treadmill, adults completed Guilford’s alternate uses (GAU) test of creative divergent thinking and the compound remote associates (CRA) test of convergent thinking. Walking increased 81% of participants’ creativity on the GAU, but only increased 23% of participants’ scores for the CRA. In Experiment 2, participants completed the GAU when seated and then walking, when walking and then seated, or when seated twice. Again, walking led to higher GAU scores. Moreover, when seated after walking, participants exhibited a residual creative boost. Experiment 3 generalized the prior effects to outdoor walking. Experiment 4 tested the effect of walking on creative analogy generation. Participants sat inside, walked on a treadmill inside, walked outside, or were rolled outside in a wheelchair. Walking outside produced the most novel and highest quality analogies. The effects of outdoor stimulation and walking were separable. Walking opens up the free flow of ideas, and it is a simple and robust solution to the goals of increasing creativity and increasing physical activity. (PsycInfo Database Record (c) 2025 APA, all rights reserved)

Tasks

• Registration – done
• Library card – done

SBIRS

• 9:00 standup – done
• 2:00 SEG – done
• 4:00 ADS – done
• Working on UMAP/clustering – good progress

[2605.22391] Epicure: Navigating the Emergent Geometry of Food Ingredient Embeddings

• We present Epicure, a family of three sibling skip-gram ingredient embeddings retrained from scratch on a multilingual recipe corpus. We aggregate 4.14M recipes from 11 sources spanning seven languages, English, Chinese, Russian, Vietnamese, Spanish, Turkish, Indonesian, German, and Indian-English, and normalise the raw ingredient strings to 1,790 canonical entries via an LLM-augmented pipeline. A 203,508-edge ingredient-ingredient NPMI graph and an 80,019-edge typed FlavorDB ingredient-compound graph, 2,247 typed compound nodes across 15 categories, seed three Metapath2Vec variants that share architecture and hyperparameters and differ only in the random-walk schema: Cooc walks the co-occurrence graph only, Chem walks the typed compound metapaths only, and Core blends both via injected ingredient-ingredient walks at controlled mixing, placing each model at a distinct point on the chemistry-vs-recipe-context spectrum.

[2601.06060] Why Slop Matters

• AI-generated “slop” is often seen as digital pollution. We argue that this dismissal of the topic risks missing important aspects of AI Slop that deserve rigorous study. AI Slop serves a social function: it offers a supply-side solution to a variety of problems in cultural and economic demand – that, collectively, people want more content than humans can supply. We also argue that AI Slop is not mere digital detritus but has its own aesthetic value. Like other “low” cultural forms initially dismissed by critics, it nonetheless offers a legitimate means of collective sense-making, with the potential to express meaning and identity. We identify three key features of family resemblance for prototypical AI Slop: superficial competence (its veneer of quality is belied by a deeper lack of substance), asymmetry effort (it takes vastly less effort to generate than would be the case without AI), and mass producibility (it is part of a digital ecosystem of widespread generation and consumption). While AI Slop is heterogeneous and depends crucially on its medium, it tends to vary across three dimensions: instrumental utility, personalization, and surrealism. AI Slop will be an increasingly prolific and impactful part of our creative, information, and cultural economies; we should take it seriously as an object of study in its own right.

Tasks

• 11:00 Coffee w/suz – done
• 2:00 Fidelity
• 3:00 Alden meeting

SBIRs

• Got the results of the run converted to a spreadsheet. The results are interesting

• Started a new run that’s going to explore the space in the 10-ish rows,

Tasks

• Vehicle registration

SBIRs

• Chain the UMAP to the HDBSCAN and do a parameter space exploration that determines optimal UMAP parameters per dimension. Done and running