Good ride on Saturday and a nice party on Sunday.

GPT Agents

• Finished two reviews! Printed out the other three for the flight

SBIRs

• Going down to Huntsville for a few days to be a booth babe. Hot but dry there, rain here.

From BlueSky

No interns today, so reviewing papers instead. One down!

Tweaked the Overleaf document.

Applied for the Reddit Research API

Liars know they are lying: differentiating disinformation from disagreement

• Mis- and disinformation pose substantial societal challenges, and have thus become the focus of a substantive field of research. However, the field of misinformation research has recently come under scrutiny on two fronts. First, a political response has emerged, claiming that misinformation research aims to censor conservative voices. Second, some scholars have questioned the utility of misinformation research altogether, arguing that misinformation is not sufficiently identifiable or widespread to warrant much concern or action. Here, we rebut these claims. We contend that the spread of misinformation—and in particular willful disinformation—is demonstrably harmful to public health, evidence-informed policymaking, and democratic processes. We also show that disinformation and outright lies can often be identified and differ from good-faith political contestation. We conclude by showing how misinformation and disinformation can be at least partially mitigated using a variety of empirically validated, rights-preserving methods that do not involve censorship.

Dog days of summer for sure

Need to start on the tasks Wolfram asked for

SBIRs

• Really interesting discussion with Aaron on CI Agents. It might be possible for token trajectories to maintain their “identity” by looking at the distance between a set of agent responses, and selecting those that have the greatest distance.
• Got some good NNM work done yesterday. I’m almost ready to generate ring buffer text along with activations by token at each layer of the model. And in case I haven’t really mentioned it anywhere else, I think training ring-buffer GPT-2 (or bigger, local models that can quickly be finetuned. You need to be able to access the layers) models via finetuning from a big model with extensive prompt tuning might be a very good way to create local maps.
• 9:00 standup
• Lunchtime ride in this gap!
• 11:00 M30
• 2:00 Conference prep
• 2:30 Hall research
• 4:30 Book club

GPT Agents

• Finish TiiS review – done!
• ICTAI – Downloaded papers. Deadline is August 18, so that’s a bit over 3 days per paper
• 3:00 Meeting
• Can Artificial Intelligence be Open Sourced?
  • This paper explores the potential of open source models to match or even surpass proprietary models in the future. The dialogue reflects a broader debate on the implications of open sourcing AI, weighing the benefits of democratization against the risks of misuse.
• The Public Interest Internet
  • Allow me to open with a wildly speculative question: What if the internet were public interest technology? I mean “internet” the way most people understand it, which is to say our whole digital sphere, and by “public interest” I don’t mean tinkering at the margins to reduce harm from some bad actors or painting some glossy ethics principles atop a pile of exploitative rent-seeking — I mean through and through, warts and all, an internet that works in support of a credible, pragmatic definition of the common good.¹

And so we say goodbye to July. That was fast

11:00 meeting with Wolfram – went well!

SBIRs

• Back to setting up the NNM project codebase. Here’s what I was planning before getting pulled off:
  • First, a program that reads in a config file and generates a file of activations for each layer. This will be a fixed size ring buffer of tokens so that the number of vectors for each layer is fixed for the “evolution”
  • Then code that reads the same config file to produce a set of UMAP files, again one for each layer.
  • Last, the visualization code using Plotly. It either reads the same config or the user can navigate to it. Management of all the files (12 layers of embeddings, 12 layers of UMAP) will be managed by extension. Might want to set up folders though.
• Started up the instance and getting back to work

SBIRs

• 9:00 Standup
• 10:00 AI Ethics talk – good!
• Maybe some NNM – nope. Couldn’t get settled into a coding groove
• Moved the the CwoC Overleaf project to the company repo.

GPT Agents

• More reviews. I have been pummeled by too many words

SBIRs

• Looks like I’m going to Huntsville for a conference
• 9:00 Intern presentation. Good stuff 🙂
• Got all the flight, hotel, and car rental handled
• Looks like book club is on for Thursday

GPT Agents

• Start journal review – started

AC gets its checkup: 7:30 – Noon

SBIRs

• Finish the whitepaper(s). One is more marketing, one is more proposal. Done. Not done, refining.
• 2:00 ML discussion? Kind of. Important no shows
• 2:30 USNA proposal discussion
• Maybe do more on NNM?
• Cancelled the book club for the week

GPT Agents

• 3:00 meeting. I think we’re discussing the invited paper. One of the things I think I’ll suggest is the HST thing that I worked out yesterday. It’s 639 words. And this is also a good thing to look at: From Burnout to Balance: AI-Enhanced Work Models

I wrote a new thing. It’s kind of based on how a conspiracy theory might be floated, but from a left leaning perspective

SBIRs

• Overleaf has changed their personal plan that makes link sharing the same as invites. Upgraded my plan, but grudgingly.
• 11:30 Sprint planning
• Time to get back to the GPT2 + UMAP – started
• Had to update my passwords on bastion and driver
• Remarkably, everything still works!

GPT Agents

• ICTAI 2024 paper bidding started. Need to do this today, probably – done
• An interesting back-and-forth with Jimmy.

I have done laundry that included a few items from the Luxembourg trip, so I guess I’m officially done with travel for a while!

SBIRs

• 9:00 Sprint demos
• 3:00 Sprint planning – delayed?
• Spent a good deal of time talking to Ron about the interns and setting up the Overleaf project

GPT Agents

• Review paper for ACM Transactions on Interactive Intelligent Systems

Chores

• Bills – done
• Lawn – done
• Clean house – done
• 11:30 Meeting with Tim – done
• Groceries – done

SBIRs

• 9:00 USNA meeting with Ron – done

GPT Agents

• I’m thinking of writing up something about providing value to consumers vs. extracting value from consumers. How this manifests in modern consumer culture (many examples of both, though timeless fashion vs. fast fashion might be a good one). Importantly, modern AI (in particular LLMs) has an innate (statistical) understanding of each of these value propositions. These systems, like any technology, can be used for either, but in this case, they can help provide a level of reflective insight, both to consumers and producers of technology platforms (private and government). I think it would be possible to generate some proof-of-concept examples along the lines of the browser plugin that I’m starting to work on. Maybe even directly reference romance scams? They are an extreme example of extracting $$  while promising to provide companionship to the lonely.

Still muggy, but not as hot:

SBIRs

• See what’s going on with my expenses – progress!
• Finish first pass at white paper? I think it’s just copy previous work over from MDBE? – done
• 9:00 standup – done
• 9:30 – 11:30 Thunderbolt – done
• 2:00 CNAS Catastrophic AI webinar

GPT Agents

• 3:00 – Starting meetings back up (Discuss conferences and upcoming work ATT&CK social and browser plugin)

Looks like the last hot, muggy day for a while

SBIRs

• More white paper. Slow progress, but progress. This will have to slide some because of the thunderbolt thing. Talk to Protima about her part. Maybe tomorrow though.
• For USNA meeting with Ron on Friday:
  • Make a copy of the NIST slide deck and tweak for USNA. Add the ATT&CK concept as a ‘Next Steps’ slide.
  • Make an initial pass at a social hacking version of the ATT&CK matrix using the concepts from the slide deck and other student ideas. Ground the ideas in sources with a solid lit review.
  • Text: Killer Apps: Low-Speed, Large-Scale AI Weapons
  • Image:
  • Video: Work by Jon Gratch for sure
  • Audio: Impact of Voice Fidelity on Decision Making: A Potential Dark Pattern?, The Voice: Lessons on Trustworthy Conversational Agents from ‘Dune’
  • Try to use GenAI (and other methods?) to attempt to implement PoC versions of the attacks
  • Try to use GenAI (and other methods?) to attempt to implement PoC defenses
• 10:00 Thunderbolt meeting – done. Added two slides to the deck and a bullet point.

Ouch!

SBIRs

• Write up stories for Social att&ck (For ETF) and WhiteHat AI browser plugin (For IUI)
• Call about Identrust – done!
• White paper. Get charge number. One of the things that I think I’ll mention is very low bandwidth loop management. Something on the order of a coordinate in embedding space (probably projected down to a reasonable number of dimensions. This is sent out as regularly as possible so that other model’s predictions of each other can be validated to a degree. If the distance in embedding space is too great, the likelihood that there is a mutual understanding is low(er)
• 9:00 standup. Get ride in before the heat – done
• 2:30 AI Ethics – done
• 3:30 Thunderbolt – done. Two slides by tomorrow COB? Tricky problem. I guess we just have to sound smart? I think cells could be an interesting direction to think about.
• Write up something for our interns by Friday to go over with Ron. And follow up with the interns on the following Friday.

Gradio is the fastest way to demo your machine learning model with a friendly web interface so that anyone can use it, anywhere!

SBIRs

• Update Jira – done
• Submit expenses (scan parking!) – done
• Write up accomplishments – done
• 1:30 Meeting – done. Not much that we can really do here, I think
• Ping Jonathan Gratch and Cosmin Munteanu – done
• Write up stories for Social att&ck (For ETF) and WhiteHat AI browser plugin (For IUI)
• Restart book club for when Aaron gets back?
• Call about Identrust
• Looks like I gotta write a white paper

GPT Agents

• Ping folks to set up weekly meetings again – done