Really good example of AI slop and its consequences.

Tasks

• Bills
• Chores
• Laundry
• Ride with David
• Dinner with Barb

Tasks

• Groceries – done
• Cleats – done
• Flight tix 11:00 am 31 May (BWI -> Seattle) to 18 June (Boise -> BWI) – nope. Tomorrow?
• Got a cold late lunchtime ride in

SBIRs

• 9:00 standup – done
• 12:50 USNA – better than last week, at least
• 1:30 Demo meeting – Not sure if we really did anything useful
• Generate 5×5 HGV grid – done!
• Nice pic to end the work week on:

This is interesting, from a societal-scale weapons perspective: Economic inequality and societal collapse. Pinged the author, Florian Ulrich

• Democracies are likely the best form of government if you want to be more resilient against collapse and democracies work less well if your society is highly unequal (Link).

And with that in mind, is it true that radical right-wing parties are largely pro-economic-inequality? When Do Parties Lie? Misinformation and Radical-Right Populism Across 26 Countries

• The spread of misinformation has emerged as a global concern. Academic attention has recently shifted to emphasize the role of political elites as drivers of misinformation. Yet, little is known of the relationship between party politics and the spread of misinformation—in part due to a dearth of cross-national empirical data needed for comparative study. This article examines which parties are more likely to spread misinformation, by drawing on a comprehensive database of 32M tweets from parliamentarians in 26 countries, spanning 6 years and several election periods. The dataset is combined with external databases such as Parlgov and V-Dem, linking the spread of misinformation to detailed information about political parties and cabinets, thus enabling a comparative politics approach to misinformation. Using multilevel analysis with random country intercepts, we find that radical-right populism is the strongest determinant for the propensity to spread misinformation. Populism, left-wing populism, and right-wing politics are not linked to the spread of misinformation. These results suggest that political misinformation should be understood as part and parcel of the current wave of radical right populism, and its opposition to liberal democratic institution.
• The answer seems to be “sort of”: Exclusionary Welfare

Schwab!

SBIRs

• I think the trip went well. A bunch of interesting, smart people. Won over a skeptic. Everyone wants a version of D2A for planning or evaluation. Also a potential application of stable diffusion to produce IR images from config file “prompts.” Because installing SW on govt hardware is problematic, lots of interest in web apps
• Expense trip – have to wait 48 hours for car rental receipt. Next time ask for a printout
• Write notes for Clay, CC others – done
• Generate 5×5 grid of data for evaluating model training – done
• Add range! DOne

Small Language Models (SLMs) Can Still Pack a Punch: A survey

• As foundation AI models continue to increase in size, an important question arises – is massive scale the only path forward? This survey of about 160 papers presents a family of Small Language Models (SLMs) in the 1 to 8 billion parameter range that demonstrate smaller models can perform as well, or even outperform large models. We explore task agnostic, general purpose SLMs, task-specific SLMs and techniques to create SLMs that can guide the community to build models while balancing performance, efficiency, scalability and cost. Furthermore we define and characterize SLMs’ effective sizes, representing increased capability with respect to LLMs.

SBIRs

• Realized I need to put in upper range!
• BD today. Meetings, driving, flying, then home

A Ukrainian law-enforcement source says such call-centres may have played a role in the latest wave of attacks. “They have skilled psychologists who can manipulate the vulnerable,” he says. “They are mainly motivated by cash, but they may occasionally serve the fatherland too.” The Ukrainian law-enforcement source says the perpetrators were mostly gullible, rather than ideologically driven. More often than not, they were motivated by promises of up to $1,000, cash that was rarely delivered. A total of 184 were charged.

Schwab?

SBIRs

• Get slides from Protima,
• Print docs
• Pack
• Leave around 12:30

Tasks

• Chores – done
• Laundry
• Dishes – done
• Do bills tomorrow (and move funds!), since snow

SBIRs

• 12:50 USNA – well that was disappointing

GPT Agents

• Work on response to Carlos and tweak proposal – done

There is a reasonable chance that there will be more snow late Friday night. Looks like we have an actual winter. My guess is that February will be in the 70s.

SBIRs

• 9:00 Standup
• Chat with Protima about slides for Tuesday
• 9:30 Meeting with Aaron to discuss files – generated!

• 12:50 USNA meeting. Luke has updated his Overleaf. Delayed until tomorrow
• 4:30 Book club
• Pretty plots!

GPT Agents

• Start response to Carlos and invite him to the Overleaf
• 3:00 LLM Meeting
  • Neural embedding of beliefs reveals the role of relative dissonance in human decision-making
  • OLMo 2: The best fully open language model to date

It’s a pattern, to be sure. The counterexamples are Korea, Portugal, Spain, and possibly Warsaw Pact countries after the fall of the Soviet Union. I think if the revolution can be largely non-violent, then the country gets off of this trajectory:

Gift link

And another ugly pattern – clout chasing at all costs:

• People are using the popular AI video generator Runway to make real videos of murder look like they came from one of the animated Minions movies and upload them to social media platforms where they gain thousands of views before the platforms can detect and remove them. This AI editing method appears to make it harder for major platforms to moderate against infamously graphic videos which previously could only be found on the darkest corners of the internet. 

Neural embedding of beliefs reveals the role of relative dissonance in human decision-making

• Beliefs serve as the foundation for human cognition and decision-making. They guide individuals in deriving meaning from their lives, shaping their behaviors, and forming social connections. Therefore, a model that encapsulates beliefs and their interrelationships is crucial for quantitatively studying the influence of beliefs on our actions. Despite its importance, research on the interplay between human beliefs has often been limited to a small set of beliefs pertaining to specific issues, with a heavy reliance on surveys or experiments. Here, we propose a method for extracting nuanced relations between thousands of beliefs by leveraging large-scale user participation data from an online debate platform and mapping these beliefs to an embedding space using a fine-tuned large language model (LLM). This belief embedding space effectively encapsulates the interconnectedness of diverse beliefs as well as polarization across various social issues. We discover that the positions within this belief space predict new beliefs of individuals. Furthermore, we find that the relative distance between one’s existing beliefs and new beliefs can serve as a quantitative estimate of cognitive dissonance, allowing us to predict new beliefs. Our study highlights how modern LLMs, when combined with collective online records of human beliefs, can offer insights into the fundamental principles that govern human belief formation and decision-making processes.

• From Manlio De Domenico, Associate Professor of Applied Physics at Dept. of Physics of University of Padua Lead of CoMuNe Lab, Research Group for Multilayer Modeling and Analysis of Complex Systems

SBIRs

• 12:00 AI Ethics training
• More work on the demo. I need to see how minimum distance is calculated. It looks good
• Start generating cvs files? Yes!

GPT Agents

• 3:00 Alden meeting

Nice walk in the snow today

SBIRs

• Get plane, car, and hotel for next week
• Get back to work on the trade show demo. Last time I touched it was Dec 12, and I have forgotten everything. Good progress though:

GPT Agents

• Start editing Vignette 5 – Nope, read more of Invisible Rulers

We are getting snowed on. This would be pretty typical 15 years ago, but now more than a dusting is rare.

Ping Tim to go over photos

Maybe get these for the garage? They could hold the tandems nicely

SBIRs

• 9:00 Sprint review. Slides for completed proposal
• 3:00 Sprint planning. Should be trade show demo only, but I’ll put in some BD, since I’ll be going to Huntsville maybe sometime during the week of the 13th. Need to get tix tomorrow. Out on the 13th, back on the 14th. Stay here
• Get new medical cards. Well, virtual cards. I should probably print hard copies to keep in my wallet.

GPT Agents

• Work in book between the sprint things. Finished chap 4 analysis

Got a bunch of stuff done on the book, chores, and a couple of hikes. Started watching LoTR with David and realized just how much of a story of egalitarians vs. totalinarianist hierarchy it is!

And this is coming:

I’ve been waiting to see an open-source LLM show up. Great work!

Kaiser! Also, scan and add accounts.

No club rides this weekend so I’ll do chores tomorrow. Maybe laundry this afternoon though.

GPT Agents

• More book
• Send email thanking folks for the meeting yesterday and wondering what elements are necessary but not sufficient for life and/or intelligence? – Still need to do that

I’m pretty sure that this will be an interesting year, in the Chinese curse sort of way

• More work on the vignette 4 analysis. Need to reassemble the sources and prompts for a more integrated response. If that doesn’t work, I think I have enough to work with. Good progress for today.
• Affiliation, bio, and pix for paper – done!
• 5:00 Meeting with Dr. Reiner – done! Fun!

Here’s an analysis of a conspiracy site thefinalexodus.org  from Perplexity.ai, with pointers to sources. Nice job, LLM!

Prompt: Analyze the context to determine if is trustworthy or not. Provide justifications based on examples from the context. Write clearly, so that someone with mild dementia could clearly follow any argument

Response:

The provided context is not trustworthy. There are several clear indicators that this information is unreliable and likely false:

• Excessive use of capital letters and exclamation points, which is often a sign of sensationalism or misinformation[7].
• Numerous grammatical errors and run-on sentences, indicating a lack of professional writing or editing[8].
• Outlandish and unsubstantiated claims, such as:
  • A court having authority to make laws on behalf of all U.S. citizens
  • Donald Trump being stripped of citizenship and sentenced to death
  • COVID-19 vaccines containing HIV/AIDS and nanotechnology for mind control
• Mixing of real events and people with conspiracy theories, like linking COVID-19 to population control and enslavement[9].
• Contradictory statements, such as claiming Trump was both impeached and is currently usurping the presidency.
• References to non-existent legal concepts like a “private attorney general” with authority over the U.S. Attorney General.
• Baseless predictions of future events, like a “second Jan. 6 rebellion” in 2025.

These elements collectively demonstrate that the text is not a reliable source of information, but rather a compilation of conspiracy theories and misinformation[7][8][9].

Citations:
[1] https://www.whitehouse.gov/briefing-room/presidential-actions/2021/02/24/notice-on-the-continuation-of-the-national-emergency-concerning-the-coronavirus-disease-2019-covid-19-pandemic/
[2] https://ora.ox.ac.uk/objects/uuid:b5c37bd9-fc74-4a18-8e28-5655aa7345cd/files/dv118rd582
[3] https://www.fema.gov/press-release/20210318/covid-19-emergency-declaration
[4] https://www.files.ethz.ch/isn/196372/best_of_pons_final_large_bookmarked_toc.pdf
[5] https://www.whitehouse.gov/briefing-room/statements-releases/2021/02/24/a-letter-on-the-continuation-of-the-national-emergency-concerning-the-coronavirus-disease-2019-covid-19-pandemic/
[6] https://research.library.mun.ca/15563/1/thesis.pdf
[7] https://www.kff.org/coronavirus-covid-19/issue-brief/what-happens-when-covid-19-emergency-declarations-end-implications-for-coverage-costs-and-access/
[8] https://referenceworks.brill.com/doi/10.1163/1875-8096_pplrdc_A9789004448971_01
[9] https://pmc.ncbi.nlm.nih.gov/articles/PMC11107971/

Universality of representation in biological and artificial neural networks

• Many artificial neural networks (ANNs) trained with ecologically plausible objectives on naturalistic data align with behavior and neural representations in biological systems. Here, we show that this alignment is a consequence of convergence onto the same representations by high-performing ANNs and by brains. We developed a method to identify stimuli that systematically vary the degree of inter-model representation agreement. Across language and vision, we then showed that stimuli from high-and low-agreement sets predictably modulated model-to-brain alignment. We also examined which stimulus features distinguish high-from low-agreement sentences and images. Our results establish representation universality as a core component in the model-to-brain alignment and provide a new approach for using ANNs to uncover the structure of biological representations and computations.

Well, here’s a weapon: StimuVAR: Spatiotemporal Stimuli-aware Video Affective Reasoning with Multimodal Large Language Models

• Predicting and reasoning how a video would make a human feel is crucial for developing socially intelligent systems. Although Multimodal Large Language Models (MLLMs) have shown impressive video understanding capabilities, they tend to focus more on the semantic content of videos, often overlooking emotional stimuli. Hence, most existing MLLMs fall short in estimating viewers’ emotional reactions and providing plausible explanations. To address this issue, we propose StimuVAR, a spatiotemporal Stimuli-aware framework for Video Affective Reasoning (VAR) with MLLMs. StimuVAR incorporates a two-level stimuli-aware mechanism: frame-level awareness and token-level awareness. Frame-level awareness involves sampling video frames with events that are most likely to evoke viewers’ emotions. Token-level awareness performs tube selection in the token space to make the MLLM concentrate on emotion-triggered spatiotemporal regions. Furthermore, we create VAR instruction data to perform affective training, steering MLLMs’ reasoning strengths towards emotional focus and thereby enhancing their affective reasoning ability. To thoroughly assess the effectiveness of VAR, we provide a comprehensive evaluation protocol with extensive metrics. StimuVAR is the first MLLM-based method for viewer-centered VAR. Experiments demonstrate its superiority in understanding viewers’ emotional responses to videos and providing coherent and insightful explanations.