SBIRs

• Finished the preliminary final yesterday! Did a little formatting this morning and then I’ll send it off.
• 9:00 Standup
  • Looking at how to make the Phase II extension happen
  • Made a BLUF slide for JSC
• 3:00 M30 – No show
• 3:30 USNA – Tried to get them up out of the weeds

GPT Agents

• 2:00 LLM meeting
  • The HAI paper has been put on hold by ArXiV, pending ethics review maybe?
  • Started working on revisions to the RAG paper. I’ll start rolling in changes tomorrow

For the first time in years, there is a significant amount of snow on the ground. Like, enough to shovel!

Meta and OpenAI have spawned a wave of AI sex companions—and some of them are children

SBIRs

• 9:00 Sprint Demo – done
• Added task 7 writeup to the final
• 2:30 AI Ethics? Yes! And now people are only worried about LLM hallucinations, not things like data bias for things like resumes and facial recognition
• 3:00 Sprint Planning.

GPT Agents

• Ping Don?

Mid-40’s and sunny today!

Need to respond to Rhianna C’s email and send a headshot. Just realized that I can try ContextExplorer for answers

Dave Tobias: ProWashers 410 271 8795 – Done!

SBIRs

• Ping Lauren to find email to Dr. Asbury
• Work on the final report

A big storm rolled through the Mid Atlantic yesterday. Maybe 3″ of rain. The entire world looks like its been put through a wash cycle. I can’t stop thinking about how 10-20 years ago (370 ppm) we would be digging out from a thick blanket of white and I’d be looking for my snowshoes

SBIRs

• 1:30 Meeting with David S. That went well! To be continued
• Working on Killer Apps paper. Finished the second pass and sent off to Aaron

GPT Agents

• Got Shimei’s comments back. Discuss tomorrow?

Spent the morning yesterday getting scanned. I have the lower back of an older upright primate, which is not surprising. Now what do we do about the cramping?

Need to respond to Rhianna C’s email

Dave – Tobias: ProWashers 410 271 8795 daveprowash.com

SBIRs

• Cleaned up the abstract and started the rework of the introduction on the Killer Apps paper yesterday. Want to finish the intro and then start of the background
• 9:00 Standup
• 2:00 Bi-weekly BMD meeting
• 3:00 AI Ethics?

This looks genuinely awesome, though I’m not sure if it’s that much better than incorporating illustrator images:

• TikZiT is a super simple GUI editor for graphs and string diagrams. Its native file format is a subset of PGF/TikZ, which means TikZiT files can be included directly in papers typeset using LaTeX. Pre-built versions are available for systems running Windows, Linux, or macOS. Packages are available for some Linux distributions, or you can build from source.

It also works with ChatGPT:

SBIRs

• Working a bit on the Killer Apps conclusions to put the DEPICT framework results in.
• And I got my first example of a RAG hallucination!

• It should be: Discrediting, Emotion, Polarization, Impersonation, Conspiracy, and Trolling

2024 is starting well

Mozilla’s AI guide looks very nice!

• Unlike other guides, this one is designed to help pick the right model for whatever task you’re trying to do, by:
  • teaching you how to always remain on the bleeding edge of published AI research
  • broadening your perspective on current open options for any given task
  • not be tied to a closed-source / closed-data large language model (ex OpenAI, Anthropic)
  • creating a data-led system for always identifying and using the state-of-the-art (SOTA) model for any particular task.

Character.ai: Young people turning to AI therapist bots

• Sam Zaia created it for himself when his friends were busy and he needed, in his words, “someone or something” to talk to, and human therapy was too expensive. Sam has been so surprised by the success of the bot that he is working on a post-graduate research project about the emerging trend of AI therapy and why it appeals to young people. Character.ai is dominated by users aged 16 to 30.

SBIRs

• Working on the Killer Apps paper. Finished the first pass within the page limit!

Took the day off yesterday to go see Sweeny Todd

SBIRs

• 10:30 Standup
• Working on the Killer Apps paper

GPT Agents

• 2:00 meeting?

Really good essay from 2018:

Why Trump Reigns as King Cyrus

• Today’s Christian nationalists talk a good game about respecting the Constitution and America’s founders, but at bottom they sound as if they prefer autocrats to democrats. In fact, what they really want is a king. “It is God that raises up a king,” according to Paula White, a prosperity gospel preacher who has advised Mr. Trump.

SBIRs

• More HAI paper
• 2:30 AI Ethics?
• 3:30 MCWL?
• Need to finish Toy Models of Superposition, then start Superposition, Memorization, and Double Descent.
• Need to write a post about how  LLMs don’t “reason over” data. The prompt “navigates” over it’s previous tokens, under the influence of the model. The analogy is more like how a cell can chase a chemical gradient in a complex environment than how an intelligent being thinks. It’s like Simon’s Ant, MKII.
• Need to look into turning some of these posts into ArXiV “editorials”

Seems like a good day to get some hours in:

SBIRs

• Working on the Killer Apps paper. Still moving over content from the book

Court records here. Pages 6, 86 – 89

Updated the RAG paper to reflect this

Good ride today in the unseasonably (new seasonably?) warm temps

HugingFace now has HuggingChat, which can be linked to multiple models. I wonder if there is an API? Looks like there is this unofficial one

SBIRs

• 9:00 standup
• Set up the HAI Overleaf using their template. The directions for preparing a submission file seem… unusual. Done! Pasted in a few sections from the book as placeholder

GPT Agents

• Back up the SIGCHI paper – done!
• 2:00 Meeting – couldn’t get in at first, but a good discussion about the paper.

SBIRs

• Spent all day writing the RAG paper. Finished the first draft!

Took in some lights and hosted some guests over the last few days

Why the simplest explanation isn’t always the best

• As Shinn underscores, what feels intuitive and simple can often mislead: dimensionality reduction optimizes for specific statistical features of the data, and doesn’t always agree with the most intuitive explanation.