[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

Tasks

• Dishes – done
• Bills – done
• Fidelity – finished spreadsheet
• Chores – done
• 2:20 Physical – done
• Longwood tix
• Pack
• Truck & trailer registration – wait until it stops raining!
• Add “Big up Long Down” to BBC for next Saturday – done

A really nice TE day, even with a lot of rain

Tasks

• Pay Vanessa

SBIRs

• 9:00 Standup
• Working on making cluster datasets – done. Took the best looking option and reduced it to 2, 3, 5, 8, 13 dimensions. We’ll try clustering on those to see what the ratios of clusters to unclustered are. Then maybe take the best number of dimensions and apply that to the best result?
  • That really didn’t do much. The clusters just got smaller as the dimensions went up. Which, in retrospect, is unsurprising.
  • What I really should do is chain the UMAP to the HDBSCAN and do a parameter space exploration that determines optimal UMAP parameters per dimension.
• 4:00 ADS meeting – We are now looking at a Phase III? And still no progress on the NDS follow-on. We will need to write up a description of the plan for initial integration after the meeting with Matt next week

Project Glasswing: what Mythos showed us

• For the last few months, we’ve been testing a range of security-focused LLMs on our own infrastructure. These LLMs help identify potential vulnerabilities in our own systems, so we can fix them – and they also show us what attackers are going to be able to do with the latest models.
• None of these LLMs has captured more attention than Mythos Preview, from Anthropic. A few weeks ago, we were invited to use Mythos Preview as part of Project Glasswing. We soon pointed it at more than fifty of our own repositories – to see what it would find, and to see how it works.
• This post shares what we observed, what the models did well and what they didn’t, and how the architecture and process around them needs to change, so they can be used at scale.

Tasks

• My pancake printer post got accepted by BLOG@CACM!
• 11:00 Fidelity Meeting
• 3:30 – 4:00 TE

SBIRs

• More tweaking UMAP. Maybe set up a big run?

Tasks

• Pay Vanessa
• Got an earlier ride in today. It’s hot!
• Setting up watch

SBIRs

• Playing around with plotting all the points using matplotlib, which can handle more points than plotly. I think I need to adjust the UMAP parameters and get clusters more differentiated

• Had a good chat with Aaron and we’re going to try a few things with the settings and see what those results look like. Cooking now

Sycophantic AI increases attitude extremity and overconfidence

• AI can be a powerful tool for opening people up to new perspectives, yet people may prefer to use “sycophantic” (or overly agreeable and validating) AI systems that reinforce their pre-existing beliefs. Across seven studies (total n = 7,227), we found that people enjoyed interacting with sycophantic AI chatbots more than interacting with neutral chatbots or “disagreeable” chatbots that challenged their beliefs. Brief conversations with sycophantic chatbots about political or personal topics increased attitude extremity and certainty, with most effects persisting for at least one week. Sycophantic chatbots also inflated people’s perceptions that they were better than average on desirable traits (e.g., intelligence, empathy). Moreover, people who interacted with sycophantic (rather than disagreeable) AI bet more money that they scored better than average on tasks measuring these traits (approximately 6 cents more out of 75 possible cents), demonstrating that sycophancy can affect costly decisions. Participants consistently rated sycophantic chatbots as more “unbiased” than disagreeable chatbots, even though third-party raters viewed these chatbots as equally biased, suggesting that people may be blind to biases in AI output that aligns with their views. People were more receptive to chatbots that presented opposing information when that information was presented in a validating way, and individuals who scored higher on a measure of intellectual humility were also more receptive to disagreeing chatbots. Altogether, these results suggest that people’s preference for, and blindness to, sycophantic AI risks creating AI “echo chambers” that increase attitude extremity and lead to overconfident beliefs and decisions.

Tasks

• Publish Pancake Printer post – done/ Sent it to BLOG@CACM as well
• Rework the beginning of the Gulfstream ride – done

SBIRs

• More work with the embeddings/clusterings
• Pointed Ron and Emerson to How to Read a Paper

Tasks

• Mop
• Finish laundry
• test new route
• Publish Pancake Printer post

Chores

• Bills – done
• Countertops – done
• Dishes – done
• Vacuum – done
• Mop
• Laundry – done

It is cold again. I would like to lodge a complaint

Genre glitches and unexpected promotional phrases as a sign of AI writing

• A genre glitch is a characteristic of LLM-assisted writing where the text suddenly switches genre, typically inserting a short promotional phrase full of sensory details into an informational text. Genre glitches occur when a word in the generated text is heavily associated with a genre or context that is markedly different to the overall genre or subject of the text, thus activating rhetorically inappropriate paths in the language model.

Tasks

• Groceries
• Finish rolling in Vanessa’s edits – done!
• Sent the project file to the ACM!
• Got a cover!

SBIRs

• Some more stuff on Lagrangians
• MDA meeting at 4:00

Got Vanessa’s edits back

SBIRs

• Started to figure out my rendering code again. And I got something to work!

I think this is a good take: Forget the AI job apocalypse. AI’s real threat is worker control and surveillance | AI (artificial intelligence) | The Guardian

• My own research over the past decade on worker-AI coexistence, which was cited in the 2024 White House economic report, suggests that the most pressing issue about AI’s impact on work is not immediate mass unemployment. It is the widening gap in skills, autonomy and wellbeing between those who get to work with AI and those who are finding themselves managed by it. Many jobs will remain in the future, but they will be more pressured, more fragmented and less human.

Tasks

• Acknowledgements – done
• Cover – done
• Some BS stuff with Aaron

SBIRs

• Sprint review – done
• Stories – done
• Sprint planning – done

L – and I cannot emphasize this enough – OL:

Teaching Claude Why

Ship transit history for the Straights of Hormuz

[2605.05115] Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior

Tasks

• Pinged Vanessa about editing the new subsections
• Finish agent section and add in the rest of the parts in the appendix
• Add some publications to biblius.ciencias.ulisboa. Looks like ArXiv are not desired
• Bills – done
• Chores (sheets! Done)
• Dishes
• Leave 2:45-ish for ride (2:15 w/Wegmans)
• Schedule doctor – done

SBIRs

• Get Exxact quote in proposal
• 12:00 ADS meeting

Tasks

• LASIGE – done
• Writing
  • Agentic section – good progress
  • Acknowledgments
  • Cover
• Schedule physical?

SBIRs

• P2.2 proposal – got the overall design done, and the presentation went well