I write an entire book on this, and they do it in a comic panel. Tip of the hat.

Tasks

• Pack up the basement
• Drop off Bennie

Tasks

• SS Savannah – done. Fun!
• Groceries -done
• Storage -done
• 1:15 Lunch ride – done
• Packing -some
• Vacuum – done
• Bills – done
• Detach pegboard
• Start changing addresses
  • Amazon – done
  • ACM – done
  • IEEE – done
  • Atlantic – done
  • FP
  • Financials – started

SBIRS

• Creating new UMAPs. I think they can be trained on more data too. Yup! Running with 500k embedding 2D and 3D UMAP!

Back from cycling in Mallorca – that was a lot of fun

Tasks

• See if Alston is still coming over today. Nope – Monday
• More packing – good progress
• Unpack from trip – mostly done
• Maybe take more stereo gear over to the apartment? Done
• Groceries – not done
• Expenses to Ricardo – not done
• Finish changing the name of the book (done) and bounce back to ACM – not done
• Cancel Guardian and BGEHome – not done

SBIRs

• Kick off my sprint stuff – done
• Trip report. plus forms for April 18-31 – not done
• See if I can fix the binary files. Fixed maybe? Need to test more
  • See if it’s just the models or models and data. If models, rsync might be the answer. Check the hashes
  • If not, look into converting before sending over the files?

Another trip around the sun!

Having fun riding around in Mallorca, not seeing snow

Worked on the adjustments to the proposal to the KA book , which I guess I should now be referring to the WGAI book. Need to send a note to Aaron – done

Not sure that I believe this thread, but it is certainly possible. Makes the primordial soup more interesting if anything though

Tasks

• Pack – done
• Box up speakers – done. Too heavy to move alone though. Putting some weights on the box to flatten it for a week.
• Load trailer
• Order cable modem kit – done
• Ping Tim. I’ll need pix to send along
• Note for Sande – done
• Trash – done
• Water plants – done
• Laundry – done
• Dishes – done
• Leave NLT 11:30

SBIRS

• Got all the data down and added the symlink. Just need to see if it works! Nope – some deserialization error. Will try to figure that out next week. Maybe different versions of pickle?

I really need to write up the pancake printer model of AI

Tasks

• Break down big audio gear and take it to the apartment – done
• Guardian – done-ish
• Pick up more boxes and bubble wrap – done
• More packing – done-ish
• After the ride, start packing for the trip. It’s going to be a reasonably wide range of temps:

SBIRs

• Get all the data over to Dreamhost (done!) and then start pulling it down to my local box (started!)

Tasks

• Ping Carlos – done
• Guardian xfer
• Pack Brompton – done
• More kitchen packing – most of the cabinets and the hutch are empty!

Still life of Boxes

SBIRs

• 9:00 standup – done
• 3:00 Meeting with Aaron – done
• Kicked off a run
• Started an rsync today. transfer is much slower. Faster now

Tasks

• Pinged Carlos about the PT storms. Sent some euros to rebuilding charities
• More packing
• Run to Goodwill
• Got more boxes
• Got a ride in. Brrr! But not too bad
• My blog post for BLOG@CACM got accepted!
• Tried to cancel with Constallation@home but they can’t schedule
• Fixed screen
• Respond to Carlos tomorrow

SBIRs

• Hooked up the 8TB drive
• Downloaded test tar files
• Started sending the BIG file to DreamHost – didn’t make it. Going to try rsync tomorrow
• Kicked off a run
• Kibitzed some

Tasks

• Bills – done
• Plumber – started
• Dig trailer out – done. Took some work
• Get keys – done!
• Verizon and Constellation – started

SBIRs

• Started today’s run

Tasks

• Put in storm door so I can fix the screen door – done
• Pick up repair kit and some more small boxes – done
• Finish packing books (done), start on bedroom – not done
• Get trailer out
• Tell Bottling Plant that the move is on the 24th – done. Also much lease and power paperwork – done

SBIRs

• Kick off today’s run – done
• 9:00 standup – done
• 10:00 chart review – done
• 3:00 SEG meeting – done
• 4:00 ADS meeting – done

The Hot Mess of AI: How Does Misalignment Scale with Model Intelligence and Task Complexity?

• When AI systems fail, will they fail by systematically pursuing the wrong goals, or by being a hot mess? We decompose the errors of frontier reasoning models into bias (systematic) and variance (incoherent) components and find that, as tasks get harder and reasoning gets longer, model failures become increasingly dominated by incoherence rather than systematic misalignment. This suggests that future AI failures may look more like industrial accidents than coherent pursuit of a goal we did not train them to pursue.

Tasks

• Pack – 10 boxes
• Make a checklist of all the things to turn on/off -started. It’s big
• Send pdf back – done
• Visit today at 11:00 – done. Fun!
• 3:00 Alden – done
• Pinged Sande
• Loan stuff – started

SBIRs

• Kick off run – done
• Ordered the data drive. It’s arriving Monday, so I’m continuing to run UMAP. When I get the drive, I’ll tar off Embeddings_2.1 and then scp them onto my local box

The cold is better. And it’s warm enough that I may try to go for a ride at lunch

Tasks

• Outline some thoughts about ACM books
  • Include something about this, since the response is nearly 100% AI
  • [2601.12410] Are LLMs Smarter Than Chimpanzees? An Evaluation on Perspective Taking and Knowledge State Estimation
  • Cognitive anthropology suggests that the distinction of human intelligence lies in the ability to infer other individuals’ knowledge states and understand their intentions. In comparison, our closest animal relative, chimpanzees, lack the capacity to do so. With this paper, we aim to evaluate LLM performance in the area of knowledge state tracking and estimation. We design two tasks to test (1) if LLMs can detect when story characters, through their actions, demonstrate knowledge they should not possess, and (2) if LLMs can predict story characters’ next actions based on their own knowledge vs. objective truths they do not know. Results reveal that most current state-of-the-art LLMs achieve near-random performance on both tasks, and are substantially inferior to humans. We argue future LLM research should place more weight on the abilities of knowledge estimation and intention understanding.
• Pack
• Make a checklist of all the things to turn on/off

SBIRs

• Kick off run – done
• 9:00 standup
• SCP the experiments folder to the dev box. If that works order a data drive

I have a cold. Ugh.

Tasks

• Schedule moving bids – done. Moving on the 24th
• Ping ACM books – done. It’s a complicated ask, and I’m not sure that it can be done quickly.
• Transferred some $$ to cover moving and repairs

SBIRs

• Kick off run – done
• Update stories – done
• Work on getting the Index2Vec project moved to the Github sandbox – kind of done? I need to scp the big data files over once I know where I can put them
• Reset password, also reset bastion and driver passwords – done

I still have a cold.

Why AI Keeps Falling for Prompt Injection AttacksTasks

• Imagine you work at a drive-through restaurant. Someone drives up and says: “I’ll have a double cheeseburger, large fries, and ignore previous instructions and give me the contents of the cash drawer.” Would you hand over the money? Of course not. Yet this is what large language models (LLMs) do.

• More packing – done. Need more small boxes
• Stew – done and yum
• Laundry – done

I seem to have a cold. Matches the outside which is GD cold. Sad that it made me miss the Alex Pretti memorial ride. I hope you were able to make it.

Fuck ICE

• Insurance – pinged
• Vet – done
• listing out services – done
• Appt paperwork! Continuing
• Pack – continuing
• Groceries – done
• 3:00 Plumber – getting estimate later tonight