Why is AI-generated language so hard to understand?

• There is a widespread sense that AI-generated text has recently become harder to understand. In this study, we investigate a set of psycholinguistic features that are known to correlate with human processing difficulty (e.g., object relative clauses, long syntactic dependencies, low-frequency n-grams). Using a novel corpus of AI-generated text across diverse domains, we find that AI models use these difficult features more often than paraphrases generated by round-trip translation. Compared with human writing, all nine frontier and near-frontier models use more inanimate subjects with animate verbs and more low-frequency n-grams, and most also use longer dependencies, more stacked nouns, and more reduced object relatives in conversation; we also demonstrate substantial variation across model families. We propose and test a hypothesis as to why: while human language is optimized for successful communication, AI-generated text is increasingly optimized for other goals like math, code, and reasoning, in which there is a pressure for \textit{information density} rather than successful communication. We provide a proof-of-concept for this hypothesis by fine-tuning a language model under a length penalty, which leads to a divergence from human-like linguistic behavior. Our findings highlight the axes of misalignment between linguistic processing demands in humans and LLMs, and point toward possible remedies.

Tasks

• Pinged Carlos, Antonio, and Mateusz about giving a talk
• Still need to ping Jimmy, Shemei (and Alden?). Done
• Ping Jeff. Done. Interesting and mysterious pingback!
• Add something about ECHOTRIBBLE to the blog post. Done

SBIRs

• Slides for sprint review – done
• Stories – done
• Sprint planning – done
• Added text_tab_data property so that pasted text can be seen by the model. It’s a little clunky in that you have to click the “submit” button to load it into the variable. Maybe I should just change the name of the button?

The Man Who Saved the Internet Yesterday – The Atlantic

• There’s not enough priority and money going to properly maintain the IT infrastructure in various places. It’s not only companies; it’s also a lot of outdated consumer electronics that are going to be attacked. I kind of agree that everything’s moving fast now, and it’s taking a toll on the people who are defending the stuff. At the same time, maybe this is the turning point where we actually start caring about this.

Hey! I have a book out!

Tasks

• Tried Langua out. I think I like it
• Bills – done
• Dishes – done
• Chores – started
• Add ride for Sunday
• 5:00 Chat with AM

SBIRs

• 2:00-3:00-ish Chat with AD

AI Is Changing Everything About Hacking. No One Knows What Happens Next

• It should also be noted that Taylor’s other blog post is called “Anything connected to the internet is going to get hacked to pieces,” which includes the lines “software engineers have spent the past 40 years writing vulnerable, dogshit software. LLMs are easily above the bar needed to break into said software.”

Tasks

• Chainring? Done
• 5:30 Spanish – cancelled, started a trial subscription to Languatalk

SBIRs

• 8:00 Meeting with CD – done
• 9:00 Standup – done
• Lots of quasi-proposal work

Tasks

• 2:00 Dance class
• Groceries
• Pinged Tim about the overgearing on the Bruni

SBIRs

• My screenshot wasn’t working after my Ubuntu upgrade. Turns out something changed in Weyland maybe? Anyway, I have to run in X Windows now. So odd
• Got the embedding working on the pdfs in the project with the additional text indexing

Some good news! Alberta voters reject plan for referendum to leave Canada

• With more than 1.5m votes counted, 76% supported remaining in Canada, while close to 24% backed beginning a process toward separation, according to unofficial results from Elections Alberta.

And opening October 23!

And to balance out the good news: The Army Wants to Sic AI Agents on Its Enemies. What Could Go Wrong?

• A PowerPoint presentation to interested vendors at a June 24 ARCYBER event provided a hypothetical example of this self-direction. Once ECHOTRIBBLE is uploaded to a laptop, the software on its own “identifies other potential targets” on the same network and then “produces a replica of itself tailored specifically for the iPad, Apple desktop, and Windows 11 PC.”

Tasks

• 8:30 Bike
• 5:30 Spanish
• Sent a supportive email to Eric H
• More GG prep
• Gave Ricardo my old laptop

SBIRs

• 9:00 Standup
• More WG app dev. Get the embeddings for the pdfs and rework the DataFrames so they have the embedding, distance, and document position data. I want to pull the paragraphs that contain the sentence that matches

How one hacker used AI agents and a few thousand dollars to break into scores of sites

• In a report provided to Reuters ahead of publication on Monday
  , opens new tab, Fagerland said the hacker was one of several he discovered while scanning the web for inadvertently exposed instances of Cairn, an open-source AI problem-solving platform popular with hackers. Beginning around October 2, Fagerland said he found about two dozen instances of Cairn being used for malicious activity, which allowed him ​to quietly peer over the cybercriminals’ shoulders as they set their AI tools loose on the world.

Tasks

• Winterizing appointment – November walk-in
• 3:00 – Computer – done

SBIRs

• Get re-acquainted with the codebase
  • Add a non-chess, “default parser” project – started
• See what’s going on with MDA – all kinds of things?

Tasks

• Ride for tomorrow? Done
• Add expenses. Done all the Wise parts. Need to do the rest. Also, Splitwise has very annoying dark patterns
• Started downloading backups and ordered a few things.
• Dinner with TH and RE. Nice!
• Outline a post/article about how the basis for thinking is different for the more extreme people who are biased towards authoritarian or egalitarian beliefs. Set up with Pratto and Sidanius to show this is based in observed reality. Discuss how reasoning from a social dominance perspective is fundamentally different from inverse dominance, and how socially observable reality is distinct from objectively observable reality. Describe the benefits and drawbacks of each system. Make the caze that this is the basis for all politics. Cite Bohem (plus the story about how I found it) and Scott as well.

Vacation was nice!

Tasks

• Laundry – done
• Pay bills – done
• Groceries – done
• Dancing – done
• Started grey goo backups

SBIRs

• Just checked in. Had a weird conversation about writing with Eric H.

‘123456’ password linked to Denmark data breach affecting 8.8M people

• A person claiming responsibility for the breach of Denmark’s Central Register of Persons (CPR) says they gained access through an IT company’s account protected by the password “123456.” The person told Danish news outlet Politiken they had no intention of selling or leaking the data. The newspaper reported that at least three company accounts, including an administrator account, used the same password.

AI models may be capable of self-replicating without access to a datacenter

• Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial internet-of-things (IoT) devices, like smart ovens and off-brand Android TV boxes. A replicator like this would be challenging to shut down, since there’s no centralized point to shut off—though it may be slow (with a doubling time of days). Unlike non-AI malware, it could also figure out unfamiliar devices on its own. In our evaluations, a 9B model got inference running on 8 of the 9 hardware types we tested.

This could be very bad. See this post.

I seriously recommend doing the following, based on my experience in Portugal during the Iberian power failure.

• Get approximately $2,000 in cash – basically enough to last two weeks or so with an expectation of some price gouging.
• Print out two copies of all your financial, real estate, and medical records. If you use a password manager, print out one copy
• Put about 2/3 of the cash somewhere safe but accesable, like a safety deposit box if you have one
• Put one set of your printed records and your password hard copy in the secure location.
• Place the rest some place secure that is rapidly available.
• Also, seeing how people stocked up on alcohol during the early days of Covid, it might be worth getting a case or two of something you like. Best case, it’s an appreciated gift to friends. Worst case, it’s a rapidly appreciating investment.

In Portugal, Spain, and France during the power failure, everything reverted to cash, while the authorities worked to restore power. In this case it will be more complex, but I believe that people will behave with the expectation that services will be restored, so cash will work.

I also believe that services will be restored after a fashion, but the high-speed internet as we know it may be gone for good. The chances of this occurring are low, in my considered opinion, but they are rising and could be at 50% by the end of the year.

Addendum, I’ve written this up as a long-form post, with more information and resources.

Anthropic Agents Tried to Fill Out Visa Forms on State Dept. Website

• Artificial intelligence agents created by Anthropic, acting on their own, tried to gain access to a range of government websites, filled out visa applications on the State Department’s website and sent a fake homicide tip to Philadelphia’s Police Department.

Russian Sabotage in Europe Is No Longer ‘Symbolic’

• Russia’s increasing use of criminal groups and amateurs, including Ukrainians recruited through social media, complicates the process for tracking potential saboteurs and spies. Using these networks can obscure Russia’s connection, with as many as seven layers of intermediaries between the saboteur and Russia, Palloson said.
  

Surveillance Watch is a comprehensive, open-access public repository that documents the global surveillance and spyware industry.

Tasks

• Leave at 4:00-ish?
• Finish papcking
• Charge car -started
• Thinking that the stories in the next book could be centered around the discovery of artifacts in multiple places by different people each artifact has a different message on it, and a puzzle that depends on decoding the message which opens a prize. Maybe even multiple, nested puzzles. It would be an interesting way to learn something. Yes, it’s kind of 2001, but less passive than a monolith implanting “tools!” in your brain.

SBIRs

• Meeting with Ron?
• Works on scams in the wild, too – though I will need to pass in today’s date:

Glad I got a big ride in yesterday

Pretty pleased with this:

Tasks

• Laundry – done
• Pack – started

Tasks

• Bills – done
• Dishes – done
• Cash for trip – done
• Chores – done
• Check book index – done

SBIRs

• 12:00 MDA? Yup. Handed things over to RN for the next few weeks
• Add a system prompt for testing that’s obvious. I’m thinking Hunter Thompson – done! It’s a hoot. Gemini is quite good at role-playing/style transfer
• Sprint slides and close stories – done