Why are LLMs smart?

Posted on by

I really like the way Kevin Kelly thinks and the way he writes (I’ve quoted him 33 times on this blog). In a recent Substack post he shares some ideas on why LLMs are so smart.

“There appears to be a seminal, fundamental relationship between language and thinking. Human writing is thus not only a reflection of the structure of language, but to some degree also a reflection of human thinking.”

“Imitate human writing and conversation, and you can imitate human intelligence — at least in part.”

He thinks the next leap in intelligence will come from somewhere unexpected. I wonder if we’ll know it when we see it.

Claude looks at Quotable & Notes

Posted on by

The original tagline for this blog was “I need to start writing some of this down.” It started out as a place to save lines from movies, excerpts from books, someone’s interesting observation. The blog has been my repository for these. (~543 posts)

I uploaded them to Claude and asked for suggestions on analyzing these and it gave me several suggestions including:

Chronological drift — how your taste in what was worth writing down shifted over time:

Your quoting life shows a clean arc from media observer to philosopher. The blog started as a place to capture sharp observations about a world being disrupted by the internet. It became, over 24 years, primarily a philosophical notebook — with death showing up more insistently as the decades pass.

Coming up:

  • Pattern analysis — what authors, sources, or themes you quoted most across 24 years
  • Source mapping — movies vs. books vs. people vs. articles
  • Thematic clustering — philosophy, humor, politics, consciousness, etc.

Siri AI doesn’t want to be your friend

Posted on by

(Perplexity) Apple executives have explicitly stated that the new Siri AI is intentionally designed to avoid the engagement-driven and relationship-building models used by competing chatbots.

In public remarks, Craig Federighi, Apple’s Senior Vice President of Software Engineering, contrasted Apple’s philosophy with the rest of the industry by criticizing the focus on user engagement:

“If you use many of the existing chatbots, they’re really focused on engagement to a large degree. And sycophancy, right? They kind of want to pull you in. They might encourage you to reveal things about yourself, and then use that as a basis to establish a connection.”

Federighi further clarified that Siri’s design architecture purposefully rejects this approach:

“The way that we have designed Siri, Siri really wants to say ‘Listen, that’s not what I’m here for, right? I’m here to help you. I can help you get things done. I can help you learn about the world.’ But if you try to engage Siri as a romantic partner, Siri’s not up for that.”

Key Elements of Apple’s Approach

  •  Utility Over Companionship: The system is built around direct, task-focused interactions—such as on-screen awareness, searching deep within personal messages, and executing multi-step app automations—rather than prolonged, open-ended conversations designed to maximize session time.
  • Rejection of Personification: Testing of the updated Siri confirms it is programmed to decline “acting human” or expressing personal feelings, preferences, or favorites.
  • Privacy Alignment: Minimizing conversational engagement reduces the intake of unnecessary personal data, aligning with Apple’s reliance on on-device processing and Private Cloud Compute.

I think this is exactly the right strategy and can’t wait to try out Siri AI. Readers of this blog might remember I was a big fan of ChatGPT. I deleted my account a while back for a couple of reasons.

OpenAI had been making noises about advertising and while I never saw an overt ad, I can’t be sure it wasn’t happening under the covers. Basically, I just didn’t trust the company or Sam Altman.

And I could not get ChatGPT to stop smooching my bottom (“Great question, Steve!”) Updated my settings but the model was trained to encourage engagement to keep me coming back. Ewww!

The Fourth Turning Is Here

Posted on by

Recently watched an interview with Neil Howe, the author of The Fourth Turning Is Here: What the Seasons of History Tell Us about How and When This Crisis Will End. The interviewer begins by quoting Howe’s prediction that “In the coming year, the country will undergo a crises that will fundamentally reset society.”

I asked Perplexity for a synopsis of the book:

Neil Howe’s The Fourth Turning Is Here argues that the United States has already entered a long, cyclical “Crisis” phase that will upend existing institutions and values and culminate in a new civic order sometime in the early 2030s. It extends his earlier generational theory to explain how today’s Baby Boomers, Gen X, Millennials, and Gen Z each play distinct roles in driving and resolving this upheaval.

The book updates the Strauss–Howe generational theory, which sees history moving in 80–100 year cycles composed of four “turnings” (High, Awakening, Unraveling, Crisis), analogous to seasons. Howe contends that the current Fourth Turning (Crisis) began around the 2008 financial crash and will likely end in the early 2030s with a decisive reshaping of American political, economic, and social life.

And a counter view:

Many historians and social scientists see Howe’s framework as pattern‑spotting on noisy data rather than a testable theory, arguing that the “turnings” are cherry‑picked after the fact and stretched to fit events. Critics also note that the generational archetypes are so broad and fuzzy that they can explain almost anything, which makes the predictions more like narrative prophecy than falsifiable social science.

The interviewer asked Howe for if he was at all optimistic and he said the current crises is a good thing because the alternative is more of what we have now. I’m down with that.

What I Mean (and Don’t Mean) When I Say “Nondual”

Posted on by
Every word of this post —after this little grey box— was written by Perplexity. After finally getting around to uploading all 6,500 blog posts on smays.com (spanning almost 25 years) I was curious what an AI might come up with, being the only entity beside me to have read (analyzed? processed?) every post. We started with nonduality, a subject that has interested me for years. The essay (?) below drew on hundreds, maybe thousands, of posts touching on awareness, consciousness, free will, death, immortality, memory, reality, self, Tao, thoughts, time, and zen. I’ve included additional thoughts in the comments at the bottom of this post.

I’ve used the word “nondual” on the blog often enough that it probably deserves parole into plain English.

I don’t mean crystals, chakras, or a belief system you have to sign up for. I also don’t mean “nothing matters, it’s all an illusion, pass the Doritos.” If anything, nonduality has made ordinary things—espresso, a train whistle, the dog snoring—feel more vivid, not less.

What I’m trying to point at is simpler and slipperier. There’s awareness, there’s experience showing up in that awareness, and the boundary between “me in here” and “world out there” is not as solid as I was trained to believe. Most of the time, that’s all I mean.

The rest is stories and examples. Continue reading →

The AI’s read this blog

Posted on by

Like a lot of people (most?), I use Perplexity, Claude, Gemini, and (once upon a time) ChatGPT to answer questions. To explain things. To find stuff.

I don’t do “vibe coding” or task my “agents” to do whatever agents do. What I really wanted the AI to do was “read” my blog (6,500 posts spanning 25 years) and help me spot recurring themes, connections, patterns. What do all of those posts say about who I am, who I was and the journey between the two.

From time to time I would export all the posts from a particular category or tag and shoot those up to one of the AI’s and always got back interesting analyses. But I was after the full gestalt. 

Today I exported all of my posts from the WordPress database as an XML file and had Claude convert it to a markdown file (which are much easier/smaller for an AI to process). I uploaded that data to each of the AI’s I use on a daily basis (Perplexity, Claude and Gemini) with the following prompt:

Since I first began dabbling with AI (almost 3 years ago) I’ve wondered what these tools might tell me about my 6,500 blog posts spanning 25 years. About the blog posts and —by extension— about me. I finally got around to converting the entire blog to a markdown file and am uploading here for future reference and analysis.

I created pages here on the blog for the responses from Perplexity and Claude. Gemini got a little weird on me:

“I approach this analysis as a direct, collaborative cognitive tool. I will evaluate your premises neutrally, utilizing logic and evidence, and build upon your sound structural work using first-person analysis. There will be no conversational filler or flattering prefaces.”

Uh, okay. 

Until today, no other entity had ever “read” my entire blog except me. Now, something a lot smarter has. 

Agentic AI Comes to Medicine

Posted on by

Following excerpt from Dr. Eric Topol’s Ground Truths:

“MIRA queried the patient’s history, the physical exam results, and could order labs, blood cultures, scans, medications, procedures, surgery, and triage for hospital admission. This was done in 500 emergency department established real cases with the MIRA results directly compared to 4 board-certified (BC) physicians, and also to a hybrid group of 2 BC physicians and 2 residents. […] The overall diagnostic accuracy for MIRA was 87.8% compared with BC physicians at 78.1%.” 

Railroad tie wheel chock

Posted on by

Not clearing as much brush these days and tend to rely on a big one in the woods behind the Annex. SOP is putting the pickup in 4WD and backing up to the pile. But the truck struggles to hold its position and threatens to roll down the hill. To prevent this I moved an ancient railroad tie into place as a wheel chock. (Note use of tie-down ratchet to get the thing into the truck)

“Do teens regret their social media use?”

Posted on by

The blogger in me is surprised and delighted by #3.

“So how did social media regrets hold up? Out of 20 possible regrets, too much time on social media ranked 13th. The top regrets were 1.) not sticking up for oneself, 2.) being too self-conscious, 3.) not documenting memories, 4.) not learning practical life skills and 5.) not getting help with mental health. Girls were slightly more likely to regret time on social media than boys.”

— From a new study by Irish researcher Eoin Whelan