I bought the audiobook from Libro.fm. Here it is on Bookshop.org.

I started listening to the book, and I'll continue listening soon. I was a bit put off by a scene that described an implausible social interaction between humans and robots: Robots need to use elevators alongside humans, but robots are required to look down around humans, so they can't record faces.

Wait, what? First: Robots? Most humans do not need to interact with robots, and we never will. Robots are machines installed in factories to automate stuff. A dishwasher is a robot. A self-driving lawnmower is a robot.

Andersen uses the term "robots" to mean wheeled or "walking" robots smart enough to do ... some tasks ... autonomously. I haven't read far enough to find out how smart.

By contrast, I use the term "artificial people" to mean machine entities that have achieved actual AI, the AI that speculative fiction has taught us to anticipate one day, indistinguishable in social context from real people. I look forward to finding out whether Andersen's robots have that level of intelligence.

Next: Faces (that is, bodies and heads)? No robot needs a face. OK, when artificial people arrive, we may enjoy interacting them on a video screen, rather than just audio or text, just as we do other humans when they can't be present. Presumably, each artificial person will choose (and routinely change?) their on-screen appearance, whether human, animal, or perhaps something entirely novel. But I can't think of a reason we would need them to be embodied, much less anthropomorphic.

And: Ambulatory robots that need to use an elevator?! Why on earth would a robot need to be present in a physical space other a fixed place in a factory? Robots don't need to attend meetings or join us for lunch.

OK, delivery robots. But we already have those and, yes, they need sidewalks and crosswalks. Maybe they'll get good enough to use the elevator for the last step of the delivery. But they won't have "faces." (Or "hands." Andersen correctly observes that his robots don't "push" the elevator buttons. They just network with the elevator.)

So I was hoping for a book that will help me develop my understanding of how we will interact with artificial people. OK, this book is speculative fiction, and that's where we learn such things. So I'll try to put aside the assumptions I've made about artificial people and enjoy the book!

Mark Hurst is an estimable designer and a longstanding critic of Big Tech.

Gary Marcus is, in Mark's words, an "AI expert - I'd even say somewhat of a booster of AI, within reason." Marcus says, "... the time has come to shut OpenAI down, at least temporarily, and charge them with computer crimes. (Or at least investigate.)"

Jensen Huang (w.wiki/EgTF) is the chief executive of Nvidia. In the interview, Huang concludes, "I think the answer is we have to shut the labs down."

Jensen Huang says he is "fairly certain" the "AI" companies "will say" they know how to build this stuff safely. I can cite two researchers who say the "AI" companies do not! And probably won't in time to save us!

(See the link below for a short Science Friday episode with two scientists studying how to do "AI" safely. They believe they may not have time. One says, "We may not have 20 years left" to do that work.)

I believe the current panic about an "AI" apocalypse is justified, but not for the reason given in all of the coverage and by the heads of "AI" companies.

The story is that "AI" may "go rogue" and kill us all. Or, more plausibly, irreparably damage human society by taking out critical infrastructure, like power stations, clean water, and, yes, the internet.

I believe that's likely, and soon, but it won't be because "AI" "goes rogue." Rather, as Jensen Huang says, it will be humans who do the damage, because it's humans who are writing software without understanding how it works and deploying it in ways that have caused real-world damage.

Worse, I believe there's is no engineering solution to fix "AI" development.

Rather, we need humans to get back to well-known software development practices. Responsible humans carefully create software that operates deterministically. We carefully test the software, both to confirm it behaves according to a concrete specification and to detect bugs and fix them.

A common approach is test-driven development, where the specification is expressed as a set of tests that can be performed automatically, every time a change is made. And each bug is added to the set of tests to ensure the bug is fixed and doesn't come back.

A good example is software developed for civil and military aviation. None of us would want to fly on an aircraft in which the engines, flight controls, communication, and navigation systems run software that isn't carefully certified. Here are some of the U.S. certification standards.

But humans are developing "AI" software that's allowed to produce real-world effects, without understanding how it produces those effects or the ability to stop it before the results cause harm.

How? By giving the software access to the internet, and using human language to prompt the software to start its work. Human language is ambiguous! So they're giving unavoidably ambiguous input to software they know they don't understand!

When humans write software to attack an important system through the internet (like clean water, banks, or governments), we call those humans "criminals."

There's no difference between those criminals and the humans building software they don't understand and turning it loose on the internet with ambiguous input. Both sets of humans know their software will cause harm.

And, I believe, the humans building "AI" are worse criminals, because they are accepting billions of dollars from unsuspecting investors. They are misleading those investors by failing to disclose honestly their software is useless, harmful, and illegal (as I wrote in 2024).

Elizabeth Holmes, convicted of such crimes, is back in the news this week. I can only hope, after "AI" irreparably damages society, there's enough society left to throw the leaders of the "AI" companies in jail. If not for deploying software that caused the damage, then for the massive financial fraud.

Yes, my preferences would be that we shut down those companies and throw them in jail now.

Science Friday is awesome. Always interesting, with episodes short enough to listen often. Here's their 15-minute episode (or a transcript, if you prefer) with Dr. Vinod Vaikuntanathan of MIT and Dr. Andrea Lincoln of Boston University:

Everyone is calling for safer AI. So what does that mean? — Science Friday

I use "AI" because what everyone calls "AI" is not AI. And when actual AIs join us as part of human society, I don't want them to call me out for insulting them by using "AI" to refer to the useless, harmful, illegal, mindless software everyone currently calls "AI."

My primary Apache web server and its aging Ubuntu OS are becoming a bit unstable. In particular, MariaDB crashes occasionally. I haven't figure out why. I've been restarting it when that happens. The web sites that use the MariaDB server can go down for a few hours without any problem.

But I'm also working on moving the databases from that server to a more table MariaDB server on a separate DigitalOcean droplet that's more up to date. To finish up that work, I wanted to know which databases are still in use. (I didn't want to search through the configuration files of several sites running Wordpress and other open source platforms on my server.

The linked page is MariaDB documentation on the binary log. I used that to create this command to see all of the databases actively in use, at least since the logs were last rotated, which appears to be at least a week.

mariadb-binlog mariadb-bin.001061 | egrep '^use `' | sort | uniq

I ran that after visiting all of the web sites on the server.

The command revealed only four web sites with active databases. I don't think that's the whole list. So I guess I'll have to look through the configuration files on the server after all.

Recommended by a cycling friend for REI-ish stuff.

On the Cycle the Hudson Valley bike tour this month, I biked by The Makery, in a cool old brick building.

A building of makers.

The Makery is a working building at 17 Railroad Avenue in Kingston, New York — a set of studios shared by the painters, ceramicists, printmakers, photographers and makers who fill them.

A lot of the cool old buildings in Kingston are of brick. Probably because they had their own brickyard 1865-1965, now a resort hotel: Hutton Brickyards.

I don't remember where I heard about this Wordpress theme.

I've been thinking lately I should look for a responsive Wordpress theme that can be adapted to my web site, which I code at the bare metal level. I've struggled to learn enough about responsive design and coding to get very far. And I haven't found an open source responsive design that isn't tied to a particular platform.

New BayCHI volunteer Adhithya Rajasekaran (LI) is using Directus to build the new BayCHI web site, using a PostgreSQL database as the back end.

Directus is the collaborative backend that puts data back in the hands of the teams who should own it.

I haven't explored Directus yet, so I don't yet grasp its architecture.

If we use Directus as our CMS, how will BayCHI's public-facing web site be hosted?

Will it be a static Apache virtual host on BayCHI's DigitalOcean virtual machine? And Directus will update those static files when we add or edit pages through the Directus CMS?

Will DirectUS use a PostgreSQL db on BayCHI's virtual host at Digital Ocean?

(I haven't worked with a "headless CMS" before. But I'm looking forward to it!)

I've asked Adhithya to comment.

New BayCHI volunteer Adhithya Rajasekaran (LI) suggested OpenBAO as an alternative to Bitwarden for BayCHI's shared password manager.

OpenBao is an open source, community-driven secrets manager and fork of Vault managed by the Linux Foundation's OpenSSF.

I don't yet see how to easily adopt OpenBAO for that use case, and it appears to be self-hosted, where BayCHI prefers paid services. I asked Adhithya to comment. We'll see whether it's a good choice!

It's cool that the user-repairable Fairphone ($650) is now available to order in the U.S. I'm tempted.

Found via Ars Technica.

It's not at all cool that the Fairphone runs stock Android infected with Google surveillance, including Google "AI." Murena sells the previous model, "Gen 6" ($650), with /e/OS, an open source, de-Googled Android. But not yet the Gen 6+.

There's a lot of reader discussion on the Ars Technica story about which open source OSs might be the best to replace stock Android.

Regardless, I'm tempted.

I ordered the SDD-13 cable from DX Engineering just now. $14 for shipping, oy! But they say it will be delivered to Scotts Valley in the next two days. We'll see!

(Amazon had a higher price, still less than the DX Engineering total with shipping, but it wouldn't get to Santa Cruz for a week or more.)

I ordered that because I found that I don't have a way to power the radio from the cigar lighter or USB in my van on long drives. I thought I did, but ...

My Fumei USB charge cable didn't work at all. I haven't done any troubleshooting. I bought the Fumei cable from Amazon a year ago, but this is the first time I've tried it. I just left a one-star review.

I ordered a Yaesu SAD-25B AC charger from DX Engineering a year ago, and I have it in the van. I can plug it in to my Ecoflow power bank's inverter, but it's not as convenient while traveling.

A K6BJ member at their Radio Wave Wednesdays meetup today helped me download the frequency list for Santa Cruz County ARES (XCZ).

From there, I downloaded the RT Systems file for the RT Systems software and the Yaesu FT60 radio. (A list explicitly for my FT-70D was not offered.)

I downloaded the RT Systems software for Windows, even though I already had it downloaded and installed on my laptop. I installed it again, in case it was updated.

I plugged in my yellow "USB-57B" cable from RT Systems to the USB port on the laptop and the mic/headphone jack on my FT-70D.

With that set-up, the RT Systems software downloaded the frequency list from my radio just fine. I believe that frequency list was for West L.A. I don't remember where I got it a few years ago.

And then I uploaded the the FT60 frequency list from XCZ to my FT-70D. That worked fine, so now I have lots of channels for Santa Cruz County quickly available on my radio.

So far, I'm familiar only with the K6BJ, WRAOK, and KN6GWQ (Watsonville and San Lorenzo Valley) repeaters, where I see the channels for 2 meter and 70 cm.

Using my smallest rubber ducky antenna, I was able to connect with a member at the meetup over one of the simplex channels.

Then I broadcast, "K6SBW radio check" on the K6BJ repeater and got a quick response.

I also spoke to member "Lisa" over the repeater: Before she left the meeting, she asked me whether I found my RT Systems software license in my email, but I put her off, too overwhelmed talking to the other members who were helping me. So I was glad to answer her over the repeater: Yes! I searched my email and found it quickly!

Today for the first time, I installed the Comet GI190NMO antenna I bought from DX Engineering last June 3 on the antenna mount I had installed on the roof of my van last June 10-13 by Lotts Auto Stereo in Santa Cruz. Those also were recommendations from K6BJ members.

Then I did another radio check over the K6BJ repeater, both 2 meter and 70 cm, and got a quick response saying I was coming in clearly.

So now I can stop saying I haven't been on the air yet!

A K6BJ member at their Radio Wave Wednesdays meetup today showed me how to look up the AllStar node number for the K6BJ repeater.

I don't actually know what that means. I'll have to do more reading about AllStar to fully understand.

The link is to ClearNode ($435 and up), a Raspberry Pi-based AllStar simplex node recommended by a K6BJ member at their Radio Wave Wednesdays meetup today. I don't yet know whether ClearNode runs the Pi-Star software or another package.

I was told there are other small Allstar devices. I found the AURSINC Shari PiHat, $70 from Amazon, a big price difference. I don't yet know the difference in capabilities.

My understanding is that those devices require an internet connection.

So I expect one can also run software on a laptop to use AllStar with a regular headset. I haven't looked into that yet.

I don't know much about AllStar. So far, all I know is that it can connect to nodes and repeaters globally. But if I can do that directly through the internet, why do I need one of these nodes?

I think I also understand that the Santa Cruz County repeaters, e.g. K6BJ, WRAOK, and KN6GWQ (Watsonville and San Lorenzo Valley) are connected by AllStar, so when I transmit, for example, to the K6BJ repeater, I'll be heard throughout Santa Cruz County on all of the repeaters. I hope I have that right.

Recommended by estimator at Krafts Body Shop.

Recommended by the Chrysler dealer in Santa Cruz.

Ordered today from Amazon. Recommended by Dan Lotts at Lotts Stereo for Sally the van. Note black frame, a little less ugly. Dan's new recommendation. Much less expensive than Dan's recommendation of the Renogy Shadowflux 200W panel earlier this year.

Selected by Dan Lotts at Lotts Stereo in Santa Cruz to allow the Sally the van to plug in at RV parks.

My dad uses this device to listen to TV audio through his hearing aids. Or, rather, he tries to use this device. The Phonak device is terribly unreliable. And its behavior doesn't match what's in the manul (PDF) available from this page. It's one of the most frustrating things I've ever tried to get working. I just don't know what to say.

Heard on WFMU's Vocal Fry.

A relatively old thread on enabling "virtual files" support in the Linux Nextcloud client. I'll have to try that. Or maybe I'll switch to syncthing first.

born in 1004 in the town of Qubādiyān in the southwest of present-day Tajikistan.

DNSimple is a LetsEncrypt sponsor, mentioned in the Internet Security Research Group email newsletter this month.

I looked up DNSimple pricing today, considering using them for DNS service for some of the domains I have registered at EasyDNS, to mitigate ... attacks? Anyway, a large volume of unnecessary DNS queries, which can cause small extra charges from EasyDNS.

This month, sbw.org will probably go over EasyDNS's 250K query limit for the "free" tier. (That is, DNS service at no extra charge for domains registered at EasyDNS.)

Some of my domains have gone over 250K in recent months. EasyDNS charges $0.50 per 100K over 250K, but they waive that charge if it's below a threshold (which I don't remember). I've paid a small charge a few months, as has BayCHI.

I've set up DNS service for a few domains on Cloudflare's free tier, which doesn't charge for (or even report) DNS query volume. But Cloudflare is a big, scary, commercial, centralized vendor, and not all that easy to use. Would DNSimple be a better choice?

DNSimple's "free" tier provides DNS service for $0.50 per zone (e.g. domain) and $0.10 per million queries.

Is it worth moving from Cloudflare to DNSimple? Maybe, just to avoid Cloudflare.

Is it worth moving DNS service from EasyDNS to Cloudflare or DNSimple? Probably not, just to avoid EasyDNS's small charges for overages.

I've been with EasyDNS for many years. I might transfer a domain to DNSimple, just to get familiar with their service.

This is not yet a recommendation, just notes on a DNS provider I'm looking into as an alternative to EasyDNS. More on EasyDNS below.

DNSimple is a LetsEncrypt sponsor, mentioned in the Internet Security Research Group email newsletter this month. That alone made me want to evaluate DNSimple.

I looked up DNSimple pricing, considering using them for DNS service for some of the domains I have registered at EasyDNS, to mitigate ... attacks? Anyway, a large volume of unnecessary DNS queries, which can cause small extra charges from EasyDNS.

This month, sbw.org will probably go over EasyDNS's 250K query limit for the "free" tier. (That is, DNS service at no extra charge for domains registered through EasyDNS at a price that varies by TLD, currently $19.49 for .com and $18.99 for .org.)

Some of my domains have gone over 250K in recent months. EasyDNS charges 50¢ per 100K over 250K, but they waive that charge if it's below $2 for overages in all domains on an account. I've paid a small charge a few months, as has BayCHI.

I've set up DNS service for a few domains on Cloudflare's free tier, which doesn't charge for (or even report) DNS query volume. But Cloudflare is a big, scary, commercial, centralized vendor, and not all that easy to use. Would DNSimple be a better choice?

DNSimple's "free" tier provides DNS service for 50¢ per zone (e.g. domain) and 10¢ per million queries.

So, is it worth:

• Moving from Cloudflare to DNSimple? Maybe, just to avoid Cloudflare.
• Moving DNS service from EasyDNS to Cloudflare or DNSimple? Probably not, just to avoid EasyDNS's small charges for overages.

I decided to transfer a domain to DNSimple, just to get familiar with their service.

Transferring a Domain to DNSimple ↩

On February 26, I initiated the transfer of k6sbw.net from EasyDNS to DNSimple. I found the experience straightforward. (And I have a low tolerance for poorly designed UIs.)

As part of the transfer, DNSimple would transfer not just the domain names, but also "the most common" DNS records. But the DNSSimple domain transfer flow actually tried to deflect me to just setting up DNS service without transferring the domain, to give me time to check whether I would need to edit or create some DNS records manually before transferring the domain.

The domain's DNS records included at least one barnacle, so I decided to just create the records in DNSimple manually. And I found that process slightly easier and quicker than my experience with either of EasyDNS's two different DNS record editors. (EasyDNS tried to make an easier editor, but never retired the older editor, probably because the easier one is no easier.)

While I was at it, I added k6sbw.net to my Fastmail domains. That'll let me create email aliases like sbw@k6sbw.net in Fastmail. Until now, I created those aliases in EasyDNS. DNSimple charges $2 per month for up to ten email aliases, which strikes me as excessive. Moving the aliases to Fastmail allowed me to try creating the many DNS records required: MX, DKIM, SPIF. And that was also slightly easier than on EasyDNS.

DNSimple put the transfer fee of $17 on my credit card as a hold, clearly disclosed at the time.

The next day, I received an email from Transfers <noreply@opensrs.email>, saying:

EasyDNS received notification that you have requested/pre-approved a transfer to another domain name registrar. If you want to proceed with this transfer, you do not need to respond to this message. If you wish to cancel the transfer, please contact us (within about five days) or please go to our website (transfers.easydns.com/index.cgi?away=1&domain=k6sbw.net&id=...)

On the linked page appeared, "New Registrar: Key-Systems GmbH ICANN-id[269]."

Hmmm ... That doesn't clearly refer to DNSimple. It looks authentic, since the link is to EasyDNS's domain name, and initiating the transfer required a random "auth-code" emailed to me by EasyDNS. But DNSimple's transfer flow, while very informative, didn't say to expect a different registrar in the messaging. So I opened a support request with DNSimple.

Within a couple hours, DNSimple replied by email:

Thank you for reaching out, and sorry for any confusion!

We partner with Key-Systems for domain registrations, so yes, that is legitimate, and if you entered the auth code on our site, the domain will be transferred to DNSimple.

I replied to suggest they include that information in the transfer flow.

I approved the transfer. Within a couple of hours, an email arrived saying the transfer was completed. And the k6sbw.net web site and the Fastmail email alias are working, so a smooth transfer, all in all.

DNSimple quickly finalized the credit card charge for the transfer fee. I don't know when exactly to expect the first monthly charge for DNS service. That monthly charge will generally be less than $1, so I wonder whether they'll actually collect those small amounts each month, or let them accrue. (I haven't searched DNSimple's documentation to clarify that.)

A day later, a family member's domain name came up for renewal, also at EasyDNS. I renewed it, paying EasyDNS's .com annual fee for renewal and DNS service, $19.49. Should I suggest that we transfer that domain to DNSimple, paying slightly more per year? OK, then, let's compare pricing!

Pricing: EasyDNS vs. DNSimple ↩

DNSimple's .net "transfer" fee is $17, which includes extending the domain's registration by one year. New registrations and annual renewals also are $17. I didn't see anything about multi-year discounts. (EasyDNS offers multi-year discounts, e.g. 9% for a 9-year renewal.)

DNSimple's fee for DNS service is 50¢ per month and 10¢ per million queries per month. (I assume that means: 10¢ for each million queries or part thereof within in a month, starting over the 1st of each month.)

How does that compare to EasyDNS? Well ...

EasyDNS charges $20.49 for transfer or registration of a .net domain plus one year DNS service, and the same for annual renewal. EasyDNS doesn't break out DNS service as a separate fee. EasyDNS says their "DNS Plus" dns service is free with transfer/registration/renewal. Plus 50¢ for each 100K queries over 250K in a month.

For the sake of comparison, let's use DNSimple's transfer/registration/renewal number of $17, leaving $3.49 per year, or 29¢ per month, for Easydns's DNS service, if the queries stay under 250K. That's half the cost of DNSimple's 50¢ + 10¢ per month.

DNSSimple's 60¢ per month stays the same up to one million queries in a month.

EasyDNS would charge a $4 overage fee for a month with one million queries. (750K query overage at 50¢ per 100K is 8×50¢, or $4.) That's almost 7x DNSSimple's 60¢ for that month, just for the overage fee.

But EasyDNS waives overage fees below $2, which would require 650K queries in a month, for which DNSSimple would charge 60¢.

Conclusion: EasyDNS's free tier is less expensive than DNSSimple if the queries are below 650K.

The most queries I've seen so far, on any of my domains, is 704K. So it's very hard to justify any of:

• Switching from EasyDNS to DNSimple to avoid the overage charges.
• Using Cloudflare (significant effort) to avoid the overage charges, even Cloudflare's free tier.
• Taking the time I've just taken to do this comparison!

But here's why I still kinda wanna switch: DNSimple doesn't use the word "overage" in their pricing. DNSimple just says "10¢ per million queries per month." That feels ... more honest? ... less judgmental?

In other words, DNSimple has a better user experience design, at least in their pricing.

If the rest of DNSimple's user experience is just as good, I'll be sorely tempted to transfer more of my domains and risk the slightly higher cost.

More on EasyDNS ↩

I've long used EasyDNS for almost all domain registrations and DNS service.

EasyDNS started selling domains 25 years ago as an affiliate of registrar OpenSRS (now Tucows). EasyDNS became a directly accredited registrar three years later. Tucows later founded my favorite mobile provider, Ting. It feels like family!

The CEO of EasyDNS, Mark Jeftovic, is an admirably iconoclastic observer of tech, with beliefs that push even my boundaries, but I'll always admire him for battling to victory over the City of London Police after their extrajudical seizure of several domains. I like to picture his testimony in the case as consisting of pointing at the City of London Police's lawyers and laughing.

EasyDNS's blog, "Axis of Easy," is always informative. Example: AxisOfEasy 431: UK Moves Toward Mandatory Phone Surveillance

And their weekly briefings by email are a mix of alarming news and silly missteps by tech companies and governments. Example: The UK Forces Digital IDs - What's Next?

Some EasyDNS service levels come with email, which I use for one volunteer org. But mostly I use Fastmail.

EasyDNS also does Wordpress hosting. I've always self-hosted Wordpress, but a host like EasyDNS would be a far better choice than Wordpress.com, which feels a bit too big-techish to me, despite their open source roots.

I was glad to transfer my .ms domains (ony.ms and sbw.ms) to EasyDNS, having first registered them through the original .ms ccTLD registrar MNI Networks. Like some other small territories with a ccTLD, I worry whether they have the technical chops, and Montserrat adds the risk of volcanic annhilation.

Antipodal to .ms, two .mp domains feature proudly in my portfolio, still with the local registrar Saipan Datacom, with the delightful URL get.mp:

• urlzr.mp, the short hostname for URLizer.com, the URL shortener I created 23 years ago.
• 2in1.mp, just because it's fun.

Funny story: I was briefly acquainted with the young founders (licensees?) of chi.mp almost 20 years ago, back in San Francisco, in my digg.com days. chi.mp offered free .mp domains (but not the good ones), but the free registrations were revoked a few years later without notice. At the time, those guys offered me the domain ju.mp, but never followed through on the offer.

Interpretive sign: "Welcome to Historic Belmark Park" Part of the permanent exhibit of Santa Monica's 2019 Belmont History + Art Project

In January, 2026, I biked to Santa Monica to pick up a car from the shop. Along the way, I walked my bike around Historic Belmar Park in Santa Monica, reading and photographing each of the interpretive signs commemorating the Belmar Triangle.

The QR codes on the interpretive signs stopped working around June, 2021. The signs are now available on the web site for Santa Monica's 2019 Belmont History + Art Project.

The Belmar Triangle was a neighborhood where many Black families lived from the 1880s until the 1950s, when the city literally burned down the neighborhood to build the Civic Center and the courthouse. In the 1960s, freeway construction destroyed more Black homes to the north.

I've known of the Belmar Triangle and the park's interpretive signs for some time, and I long meant to read each of them, to learn more. I'm a little ashamed that it took me so long.

I was unaware of West L.A.'s Black history until I moved to Venice in 2019. I lived in an apartment in the Oakwood neighborhood and gradually learned that Oakwood became a predominantly Black neighborhood in the early 20th Century, as Black people arrived to work in service jobs. Much of Los Angeles had racially restrictive housing covenants. The Belmont Triangle and Oakwood were among a few places where Blacks could own property.

Many Black families built their homes in Oakwood, often in the American Craftsman style. And earlier, many shotgun houses in the Belmont Triangle. A shotgun house is preserved next to the Santa Monica public Library, and many of the Craftsman houses remain in Venice. I've met a few Black neighbors in Venice who told me they grew up here, descended from those who first came over 100 years ago.

Now we learn that the National Park Service wants designate the beaches as a National Park from Will Rogers State Beach north of Santa Monica to Torrance Beach near Palos Verdes, around 20 miles of beaches, plus 200 yards inland from high tide.

• LAist: Could LA beaches become a national park?
• Santa Monica Daily Press: Parks service says federal control of local beaches could spur new, protections, conservation efforts and economic activity

The stories quote NPS Project Manager Sarah Bodo: "If a national park is designated, the Park Service would seek to work collaboratively with local communities and existing agencies on common goals for resource protection and recreational opportunities."

But I worry that the true agenda of the current administration is to seize more control over a piece of California that embraces its diversity and teaches residents and visitors about our past mistakes. The NPS may change the public facilities along the beach to match the administration's racist view of American culture.

I doubt the administration will take Santa Monica back to the days of whites-only beaches, with Black families restricted to a two-block stretch of beach with a derisive nickname (w.wiki/HrrZ). Historic Belmont Park is 750 yards inland, but if the administration threatens the city over the park's honest message of diversity, will the city give in?

I recently noted the Veterans Administration has added ideological propagana to signage at the 138-year-old West L.A. Veterans Home. I expect NPS might take similar actions if the beaches are designated a National Park.

I took note of this USA Today story when it appeared. I scheduled annual emails to myself to remind me on the date of her birth and her death, and on March 9, until I can publish my bicycle ride on that day as the "Hattie Scott Peterson Ride." So far, I haven't ridden on those days.

I'm finally building the camping interior for my Pacifica, using 80-20 materials. I found these guides helpful.

The Santa Monica Pier at Sunset, January 14, 2026

Santa Monica Bay, view southeast from Saddle Peak Road at the top of Stunt Road. The headland in the distance is Palos Verdes.

The Miracle Mile and Los Angeles Skylines, view to the east from Saddle Peak Road at the top of Stunt Road. Snow in the San Gabriel Mountains.

No, really, timeanddate says sunset was at 5:07 p.m. that day at that location, and the image metadata says I snapped the photo at 21 seconds after 5:07 p.m. That's largely a coincidence, but I did "plan" this photo a little bit. Let me explain.

My route on the bicycle was from Venice, to the top of Sepulveda pass, west on Mulholland to the end of the pavement, down to the foothills of the San Fernando Valley, west on Mulholland Highway, up Stunt Road to Saddle Peak Road, down to Topanga Canyon Boulevard and PCH back to Venice. 53 miles with 4,600 feet ascent.

I left at 10:01 a.m., expecting to finish well before sunset.

I stupidly thought I might be able to ride that route in about five hours, even though the last time I rode the route, 7 years prior, it took over six hours, and we were all much younger then. (That ride was 5,600 feet ascent, because I rode dirt Mulholland to Woodland Hills, where the pavement starts again, more climb-y.)

But today it took over 7 hours: I stopped for about 30 minutes in Calabasas to eat a sandwich. Then I found the 10-mile, 1,800-foot climb from Calabasas up Stunt Road much harder than I remembered: That climb took 100 minutes, where it was 81 minutes 7 years ago. And I had to wait at least 30 minutes for flag crews, first on Saddle Peak Road, then another on the way down Topanga Canyon Boulevard.

So the sun was already low on the horizon when I got to the Santa Monica Beach Path. I decided to stop to take a photo, but I knew the photo would be better when the sun was already halfway hidden by the horizon. And I thought I might have enough time to ride under the pier to compose a nice shot with the ferris wheel before it was too late. So I kept riding at 14 mph for about 3½ miles, looking over my shoulder many times to keep track of the sun.

And it worked out perfectly. Just beyond the pier, I stopped and immediately took a series of photos in the 53 seconds before the one you see here, which is the last one I shot. I'm delighted that my 2½-year-old Android phone's dynamic range was just enough to capture some foreground detail and the lights on the ferris wheel, even though I manually darkened the exposure by tapping on the sun in the viewfinder. I used the phone's 3X optical zoom, which often produces way more interesting composition, due to the foreshortened perspective.

In the photo, there's a low shape on the horizon just to the right of the sun. That can't be the Channel Islands, 70 miles in the distance, because they're further north, bearing 267°, where the sun set at 245°. In the second photo, there are two tankers offshore, but they're too far south. I think the shape might be clouds far offshore.

I'm often frustrated with my phone as my primary camera, especially stuff like the moon. But this shot came out nicely!

My friend Dennis H. from Santa Cruz recommended this to me as a gravel bike for touring and bike camping. The Niner web site and Adrenaline Bikes in Orange, California, both say it's out of stock. I'd love to find one used, to save money.

The 1900 Streetcar Depot (w.wiki/HCir) has long interested me. It took me a long time to find an authoritative sources of information on the depot and the streetcar and trolley lines it served.

From the Library of Congress, the location of the Streetcar Depot 1893-1909, and the new location 1909-1925.

1898 U.S.G.S. Topographical Map of the Soldiers Home, showing Home Junction and the Soldiers Home Branch Line.

From the National Register of Historic Places, the Streetcar Depot in 1963. Photo by Julius Shulman.

From Google Street View, the Streetcar Depot in November, 2016.

From Google Street View, the Streetcar Depot in October, 2025.

The Streetcar Depot December 25, 2025. Since October, it was moved to the lawn in front of the Veterans Administration Building 218 on Eisenhower Avenue, alongside an original wooden barracks building.

I had always wondered how the streetcar line entered the Veterans Home property, because of where the depot was located 1909-2025: Between Dewey and Pershing Avenues, just west of Bonsall Avenue. And because I assumed the streetcar line came in from Los Angeles, to the east. But there's a small escarpement, about 25 feet, just east of Bonsall Avenue, so I doubted the line came in that way.

The depot originally stood at the terminus of the 1893 Soldiers Home Branch Line (w.wiki/HCj6) of the Santa Monica Air Line, which ran two miles from Home Junction (now Metro Expo/Sepulveda) (w.wiki/HCj9) northbound on the west side of Sepulveda Boulevard, according to Wikipedia. (Alas, Wikipedia's citations don't confirm that routing. I had to find confirmation elsewhere.)

The Electric Railway Historical Association of Southern California says this about the Soldier's Home Branch Line (erha.org/pelines/pewal.htm):

The Soldiers' Home Branch left the Air Line at Home Junction, proceeded via west Los Angeles (Sawtelle) to Soldiers' Home grounds, a distance of 2.14 miles, all single track and all on private way; this branch crossed the Santa Monica via Sawtelle Line at Santa Monica & Sepulveda.

...

The Soldiers' Home Branch, 2.61 miles long from Home Junction to the Home at Sawtelle, was built in 1877 by SP. .... Passenger service on the Soldiers' Home branch was abandoned probably in 1920 except for special movementss.

(I don't know why the inconsistency in the length of the line.)

Now that I know the line came in from the south, I can see on the topographical maps that the escarpement diminishes as the land to the east rises to the level of Wilshire Boulevard.

A VA history web page says of the Veterans Home:

Separate residences for officers and other staff, a dining hall, multi-denominational chapel, hospital, and streetcar depot were arranged around a large open lawn that served as a parade ground.

The chapel, also built in 1900, still stands in its original location, just north of Wilshire Boulevard on Bonsall Avenue.

1898 U.S.G.S. topological map shows the Solders Home Branch running north from Home Junction (Now the Expo/Sepulveda Metro Station at Sepulveda and Exposition Boulevards) along the west side of Sepulveda Boulevard past a point south of Wilshire Boulevard, likely at Ohio Boulevard, where Sepulveda ended.

That U.S.G.S. map shows a road continues north, through the future Veterans Home property, and enters Sepulveda Canyon in the vicinity of today's Sunset Boulevard intersection with Sepulveda Boulevard and I-405. Today, no roads exit the north end of the Veterans Home, but it appears at the turn of the 20th Century at least one did.

A Historic American Buildings Survey from the Library of Congress shows the Soldiers Home, the Soldiers Branch Line, and the locations of the depot: Originally, 1893-1909, south of Wilshire between Bonsall and Sepulveda. Then, 1909-2025, at the end of today's Dewey Avenue.

Now, the current depot building was built in 1900, according to Wikipedia's references. I don't doubt that there was a depot south of Wilshire back to 1893, but not the building surveyed. (Other sources say the depot building was built in 1893. I'll try to confirm which is correct.)

The survey shows the Soldiers Branch Line continued north from the end of Sepulveda shown on the 1898 U.S.G.S. map, bending slightly west and passing the depot to cross Wilshire east of Bonsall. The branch then loops to the north of Wilshire within the Soldiers Home, the loop extending east across the path of present-day I-405 and Sepulveda Boulevard. A note says, "Loop allows turning of complete trains." And a short spur serves buildings near Pepper Avenue (present-day Constitution Avenue), presumably for freight deliveries. The loop and the spur are all east of the escarpement.

Of the post-1909 location, LOC says:

1887 Horse Car Line. In 1908 became Los Angeles Pacific electric trolley tracks. Tracks removed around 1928.

That appears to refer to a spur off of the 1908 Westgate Line, which crossed Wilshire just east of 9th Street (today's Federal Avenue) to join San Vicente Boulevard briefly on its way to Montana Avenue and the shoreline.

On Christmas Day, 2025, I rode through the Veterans Home and noted that the depot has been moved to the lawn in front of the Veterans Administration Building 218 on Eisenhower Avenue, alongside an original wooden barracks building. Both buildings have new foundations and interpretive signage.

So, 125 years after it was built, and after 116 years in its last location, the depot now has a new home. And many years after I first took an interest, I finally know more about its years of service.

The nearby Westgate Line also is interesting. Also from the Electric Railway Historical Association (erha.org/peww.htm):

From Sawtelle this line looped north and west on private way through Brentwood ... Track was on private way long (sic) San Vicente Boulevard from Wilshire Boulevard to Montana St. & Ocean Ave.

One of the links below is a low-resolution map that shows what appears to be the Soldiers Home Branch and the Westgate Branch as contiguous, a loop north and then southwest to the shoreline, but they weren't. Another map below shows the "Soldiers Home" terminus at the end of a little branch off of WestGate, which may refer to the depot's post-1909 service on the trolley tracks that branch off of the Westgate Branch, according to the Library of Congress survey.

A photograph below has a caption that says the Westgate Branch left the Air Line at Santa Monica Boulevard and Purdue Avenue, which is one block south of the Veterans Home property boundary at Ohio Avenue. In the photo, it's not a crossing, but a branch line turning to the north. So the Soldiers Home Branch and the Westgate Branch were not contiguous, but they ran through the same part of Sawtelle.

Update 2025-12-27: ↩

Here's a link with more information and a map. It may correct some of what I've written above:

• Sawtelle California 1897 to 1950

This page reproduces a photo that I found elsewhere with a confusing caption (below). The description of that photo is clearer on this page.

This page also has a diagram, "Los Angeles Pacific Co. Balloon Route" detail of Sawtelle, showing both the Soldiers Home Branch and the Westgate branch. It confirms the 1909-2025 location of the depot was on a spur off of the Westgate branch. I'd like to find a better-resolution version of that diagram.

This page says the Westgate branch continued down San Vicente Boulevard to the shoreline, not down Montana Avenue, as other sources say. I'll try to confirm which is correct.

Link Dump ↩

• V.A.: Explore the rich heritage of the VA Greater Los Angeles Healthcare System
• Water and Power Associates (Not LADWP.): Early Los Angeles City Views (1900 - 1925)
• Water and Power Associates: National Soldiers' Home
• Water and Power Associates: Early Los Angeles City Views (1800s)
• SoCal Landmarks: West Los Angeles Street Car Depot
• Library of Congress: Drawing and Narrative of Streetcar Depot
• Santa Monica History Museum: The Old Soldiers Home
• PBS SoCal: The Balloon Route: A Tourist's Trolley Trip Through Early-1900s Los Angeles
• Pacific Electric Railway Historical Society: Pacific Electric Railway Sight-Seeing Trolley Trips Brochure (1912)

On this page is a confusing photo caption:

• Water and Power Associates (Not LADWP.): Early Los Angeles City Views (1900 - 1925)

The photo contains a lettered caption, "Oregon Avenue, Sawtelle, Cal." But just below that photo is this caption:

(1906)**^*^ -- View looking east on Oregon Avenue, which was the original name of Santa Monica Boulevard in this part of Los Angeles. The line going off the left of the frame is the Westgate Line that ran down the middle of Burton Way through Brentwood Park. The exact location of this photo is somewhere near present-day Santa Monica Boulevard and Purdue Avenue.

Assuming "Westgate Line" refers to the Westgate Branch (described above), then, yes, that's the branch from Sawtelle north and west along San Vicente Boulevard and Montana Avenue to the shoreline. Indeed "Purdue Avenue" is in Sawtelle and intersects Santa Monica Boulevard one block south of Ohio Avenue, the southern boundary of the Veterans Home. So that makes sense.

But "Burton Way" is four miles east of Purdue Avenue on Santa Monica Boulevard. Burton Way runs east from South Santa Monica Boulevard and curves south to become a completely different San Vicente Boulevard. That other San Vicente Boulevard has often confused me when I'm driving over in that part of L.A.

And "Brentwood Park" is back in Brentwood, but well north of our San Vicente Boulevard.

I noticed the Files app on Ubuntu doesn't display the file timestamps in a useful format. I prefer a format that sorts naturally, since that's what my brain expects. None of this 17 Dec 15:31 crap.

DuckDuckGo found this fix for me.

I expected it to be a setting in the app, but I suppose it's not inappropriate that it's part of the locale translation system. Still, the default seems especially unfriendly.

The pinned link is a 2022 post on the Quicken "Community" by a user who, like me, was trying to get Wine to install Quicken on Linux. I found that post after trying wine Quicken.exe (the installer) and seeing an error message with the string InstallPDFDriver.

The post unpacks the Quicken.exe installer (a 7zip archive?) and edits the gobbledygook that tries and fails to install a PDF driver.

And then how to run he unpacked installer. That appeared to make the installation successful, but now I can't get Quicken to start.

So more hacking to do. Or just switch to GnuCash.

The pinned link is to the "OpenCloud with Docker Compose" section of the OpenCloud installation documentation. It explains how to install OpenCloud with Collabora for web file storage and document collaboration. ("Multiple users can work on the same document simultaneously, with real-time synchronization of changes.") And the usual cloud storage sync'd with desktop/laptop/mobile devices. Devices can keep copies off-line or use the cloud copies to allow access to files without consuming device storage.

I came across OpenCloud today as a "lightweight alternative to NextCloud." (So says DuckDuckGo.) I've done that search from time to time, because my venerable NextCloud installation hard to maintain and requires a relatively expensive DigitalOcean Droplet ($24 per month).

So I reviewed the OpenCloud installation steps, then spun up a very small DigitalOcean Droplet ($4 per month) to try it. The OpenCloud docs said 1 CPU and 512MB RAM would be sufficient. But when I got it installed and started it up, it immediately began swapping so hard that even SSH was unresponsive.

So I increased the CPU and RAM, and that made it work well. From the DigitalOcean performance graphs, I believe 1 CPU and 2 GB RAM would be plenty ($12 per month).

I installed the Windows client and experimented a bit with syncing. That all worked fine. And I could edit and view text files and Markdown files, and view uploaded PDFs and images. I didn't think to try playing sound and video files.

But I couldn't get Collabora to work, so I posted a question on the Github project. I'm sure it'll work. I just don't know how to troubleshoot.

I took a snapshot of the Droplet (32¢ per month), so I can start it up again quickly, then destroyed the Droplet just to save the cost until I can make progress with Collabora.

But Cloudflare can also be used to deal with attacks on DNS service, while still allowing the primary server to respond to HTTP requests. That's what I needed. And Cloudflare gets it almost right. Close enough that I've switched three of my domains to use Cloudflare as their authoritative DNS.

My domains are registered through EasyDNS, and I also use EasyDNS's bundled DNS service. In recent months, three of my domains have been hammered with excessive DNS requests, beyond EasyDNS's monthly limit. EasyDNS handles the requests just fine, and they impose a very small charge for the excess.

I can't say whether those excessive requests are malicious attacks or somebody's mis-configured botnet, but I wanted to avoid that small monthly extra cost.

First, I tried setting the TTL higher, as recommended by EasyDNS. That had no effect, because whatever systems were sending the requests weren't caching the responses as specified by the DNS protocol.

EasyDNS support suggested using a CDN. I knew I didn't want to use Google, Microsoft, or Amazon's CDN. Or Akamai. I didn't want to use Cloudflare, either. Ideally, I would find a more independent provider, but I didn't want to spend a lot of time on finding a solution.

Cloudflare has a free tier, so I created an account and added the domains. This is where Cloudflare works great, but not quite right. When I add a domain, Cloudflare takes these actions:

1. Copy the domain's zone to Cloudflare's DNS server.
2. Gives the domain owner instructions to change the domain's authoritative DNS to Cloudflare.
3. Slams the domain into Cloudfront's full-blown cache service, responding to HTTP requests in place of the primary web server.

Yes, unless the alert domain owner notices Cloudflare turned on services they didn't ask for, Cloudflare will take over serving their web site. That causes problems with HTTPS certificates, security blacklists on the primary web server, and web analytics that don't use surveillance. Two of my three domains went down as a result, until I turned off the unsolicited "service."

Also, Cloudflare does not update the DNS records when I change them at EasyDNS. That's only available as Cloudflare's DNS Firewall product, which is part of their Enterprise plan, evidently so expensive that they don't publish pricing.

Instead, I now must edit the DNS entries twice, once in Cloudfront, and once in EasyDNS (in case I ever need to turn off Cloudflare).

Two of the affected domains are Mediawiki sites: VanagonWiki and Bondline. I don't know whether the excessive DNS requests have anything to do with some Mediawiki vulnerability. (Those sites also were enduring massive HTTP traffic, millions of requests with meaningless URLs, around the same time, but from different IP addresses. I installed a UFW blacklist to block those.)

The third is BayCHI, which runs hand-crafted PHP I wrote decades ago. Again, no idea why.

Here's the notification I emailed to the BayCHI steering committee:

I just created a Cloudflare account for BayCHI and set up Cloudflare to handle BayCHI's DNS service, using Cloudflare's free service tier.

BayCHI's domain name registrar remains EasyDNS. I used EasyDNS's tools to make Cloudflare the authoritative DNS for baychi.org.

I made this change because BayCHI has been the target of a very high level of unnecessary DNS requests off and on for several weeks. I don't think the requests can be called an "attack," more like an annoying mistake.

Until this change, BayCHI also used EasyDNS for DNS service, that is, responding to DNS requests. EasyDNS handled the high volume just fine, but EasyDNS charges a nominal monthly fee if the volume goes over a stated limit. Adding Cloudflare's service avoids that fee.

Cloudflare has no stated limit on DNS requests. Cloudflare's business is more oriented to handling (or blocking) this kind of attack/mistake than EasyDNS.

I spent some time looking for a more independent/distributed provider of services like Cloudflare's. Cloudflare represents a centralized service with no meaningful diversity, which means it's a significant threat to the whole internet. A recent attack shows just how big a threat.

But I didn't find a provider that we might use to increase diversity, so I bit my lip and used Cloudflare. If any of you knows of an independent CDN that provides DNS service, please tell me!

More detail on why the offending DNS requests are unnecessary:

The DNS protocol includes "time to live" information in each response, and the requester is supposed to honor that by caching the response for some time.

The DNS requests are probably coming from a mis-configured botnet that doesn't cache the responses correctly. But that's just me speculating.

I loved last year's two-day Climate Ride in Marin County. I'm looking forward to joining again in May, 2026. Here's my fundraising page.

Found via jhhl on the Creative Good Forum, who writes:

The guy behind Faircamp has a bunch of static webpage tools, many in development! Static webpages have the advantage that they do no backend computation and usually don't have much client side processing either (sometimes it's helpful) which lessens the attack surface of course. I like static websites and non surveilled tools!

Mentioned on the Santa Cruz Amateur Radio Club email list, where the OP says it can be used off-line.

The Android app is available in the Google Play store, but not promoted by the web site, for some reason.

Mentioned via the Santa Cruz Amateur Radio Club email list.

Instructions and a couple of shell scripts that use UFW on Ubuntu to block IP addresses based on daily-updated blacklists.

(I've tried a few times lately to work through how to get fail2ban set up. I found it just too complex.)

I just installed it on the BayCHI web server because it has been getting thousands of bogus requests per minute from various countries since November 4. It has already reduced the volume of bogus HTTP requests noticeably.

Earlier today, I added code to BayCHI's PHP scripts to immediately send a 403 response when the bogus request is recognized. That doesn't block the request from reaching Apache, consuming CPU and memory, and executing a lot of code in the Apache server and PHP language. It immediately reduced the CPU from 100% to under 10%, and the memory from 92% to about 50%.

Next, I might scrape the IP addresses on the bogus requests from BayCHI's Apache logs and submit it to one of the blacklists.

Found via ustoopia.nl.

Open Source alternative to Microsoft Phone Link, available for Linux.

I haven't read the Privacy Policy yet, but it sounds like it doesn't have any surveillance.

Found via review on How-To Geek (warning: ads, surveillance).

Recommended by Lotts Stereo in Santa Cruz for my Pacifica van. My EcoFlow River 2 Pro power bank has a solar charge controller that's compatible. I'll need a way to switch between the solar panel and the van's 12V to the EcoFlow's single DC input. Maybe one half of a bridge rectifier with Faston connectors, although the one-volt drop is a bit disappointing.

Also, the Ecoflow manual warns against having DC and AC connected at the same time. So I may have to devise a more complicated way to automatically switch the inputs between solar, 12V, and AC shore power.

Recommended by Derek Sivers. Derek installed it on his Mac laptop as an alternative to "ugly" Mac OS Tahoe. Should I get a cheap Mac laptop to run Linux, instead of a Thinkpad or something from eBay?

Recommended by Derek Sivers. I love fanless computers. This mini is more expensive than I want right now, and what I really need is a new laptop.

Recommended by Derek Sivers.

The people involved in the Whole Earth community founded Wired in 1993. ... these people had been working together for nearly 30 years before they transformed our definitions of computing. They did this by brokering a connection between the counter-cultural world and the technological world. In doing so, they have influenced culture, business and design at every level ...

When Kelly speaks, I listen. So I'm astonished to disagree with his latest blog post "Paying AIs to Read My Books."

I predict in a very short while, authors will be paying AI companies to ensure that their books are included in the education and training of AIs. The authors (and their publishers) will pay in order to have influence on the answers and services the AIs provide. If your work is not known and appreciated by the AIs, it will be essentially unknown.

...

If AIs become the arbiters of truth, and if what they trained on matters, then I want my ideas and creative work to be paramount in what they see. I would very much like my books to be the textbooks for AI. What author would not? I would. I want my influence to extend to the billions of people coming to the AIs everyday, and I might even be willing to pay for that, or to at least do what I can to facilitate the ingestion of my work into the AI minds.

He argues that writing will change to best influence AIs, just as we write to influence web search engines today.

Kelly's argument is convincing. I find I disagree with him on only one point: I believe AIs are useless. They give wrong answers in an authoritative tone, and they don't cite sources (at least not well). I do not believe there is any way for the current LLM technology to become useful as a search engine, much less a reliable reference tool.

Kelly believes they will become useful. I hope he's right. I hope new machine intelligence technolgy emerges from the ashes of the AI crash, one that we can use with confidence and even appreciation. One that does what we most assiduous search engine users do now: Ask a question, and then carefully review the sources of the answers for correctness and authoritativeness. LLMs absolutely do not and cannot do that, so they can't crash soon enough to suit me.

Mark Hurst pointed me to Gary Marcus.

Wikipedia on Marcus's Neuro-symbolic AI:

Gary Marcus (argues), "We cannot construct rich cognitive models in an adequate, automated way without the triumvirate of hybrid architecture, rich prior knowledge, and sophisticated techniques for reasoning. ... To build a robust, knowledge-driven approach to AI we must have the machinery of symbol manipulation in our toolkit. Too much of useful knowledge is abstract to make do without tools that represent and manipulate abstraction, and to date, the only known machinery that can manipulate such abstract knowledge reliably is the apparatus of symbol manipulation."

Marcus's latest book, with co-author Ernest Davis, is Rebooting AI.

• Paperback from Bookshop.org
• Unabridged Audiobook from Libro.fm

From the blurb:

Taking inspiration from the human mind, Marcus and Davis explain what we need to advance AI to the next level, and suggest that if we are wise ... we won't need to worry about a future of machine overlords. If we focus on endowing machines with common sense and deep understanding, rather than ... gathering ever larger collections of data, we will be able to create an AI we can trust - in our homes, our cars, and our doctors' offices.

Those are the tools I look forward to after the AI crash.

I believe those AIs also must understand human psychology and social interaction to some degree, because relationship to the people we learn from and the social structure we live in are all part of our experience. An AI that doesn't take that into account won't feel like a trustworthy partner in exploring and learning about the world.

Also read Kelly's blog post for a concise, clear explanation of the Bartz et al v. Anthropic case, which found Anthropic guilty of copyright infringement, even though Anthropic evidently did not use their illegal copies for profit. Interesting.

Recommended by Maryann. 800 miles from Santa Cruz. 500 miles from Los Angeles.