Color me impressed with up-to-date LLMs

I only recently started playing around with LLMs, and I started with somewhat dated models.¹ Yesterday I signed up for ChatGPT and I can see why some former sceptics have gone fully bought in. The 4o model is impressive.

My main investigative tool was to pick things that I know something about, and ask the model a question from those fields neither trivial nor really hard but requiring some nuance. I tried to select ideas with a range of popularity because I believe that the volume of writing on a subject may influce the "skill" the model exhibits. My subjects so far:

1. Discuss the Epistemological implications of Godel's incompleteness theorem.
2. Summarize Freemon Dyson's "Time Without End" paper. (If the model did well on that I followed up with a question on which of the papers conclusions had been overcome by more recent developments in cosmology.)
3. Summarize Lamb's "Anti Photon" opinion piece.
4. Discuss the similarities and differences between the movies Ghost Dog and Leon: The Professional.
5. Explain the continued strong culture of short form composition in speculative genre fiction as compared to general fiction
6. Differences between classical and modern guitar
7. Prepare a packing list for a lightweight wilderness survival kit to carry on my day hikes in the desert southwest.

My highly unscientific observation suggest that items 1, 6, and 7 are the more popular topics and should provide mode source material for the models. The movies question (4) is weird: each movie has been extensively discussed so there is a lot of source material for each movie, but I'm not sure how much human generated comparison text is out there. Then I would rank 2 and 5 a more common than 3.

I didn't get any major factual errors from ChatGPT4o. I's main fault was that much of the writing was bland and characterless on all responses. Of course, the desireablity of "character" is context dependent: an encyclopedia or other raw factual source should be pretty neutral and that's what most of the responses sound like. ALso I didn't do any prompt engineering to elicit character: I just gave the model the question and let it go.

I do have two speific comments on ChatGPT's responses. First, it's interpretation of Lamb's paper if different enough from mine that I wasn't particularly happy with it, but I wouldn't be surprised to learn that other trained and capable physicists subscribed to that interpretation. Second, the answers it gave for the movies really sounded like they were drawn almost entirely from reviews for each film in isolation:² Ghost dog this; Leon that. Over and over again.

---

¹ My initial goal was to investigate the possiblity of using a LLM as a purely local coding assistant (that effort continues). At work our security guidelines imply that no internal code or design details should be release to the wider world or sent over an unsecured internet connection. Obviously just using CoPilot from VS Code is out. But I started this investigaction on a personal machine to find out if the tooling would meet requirements before going in search of management buy-in. Alas, my "best" machine is totaly unsuited for even late 3rd gen models. I can run models up to about 1 billion parameters smoothly and easily. A 7B parameter model (zephyr) is too slow for use in a work-flow (seconds-per-word), though it's fast enough to learn about the models. I have put a 22B parameter model on the machine, but it runs at minutes-per-word.

² Indeed that observation is what prompted my comments above.

Not an advertised feature

The child has a tablet and likes to stream stuff on it in the car, which means tethering it to one of our phones (yes, we're wondering about making the next one cellular enabled). You can imagine the various minor inconveniences that implies, but it also means that when my wife set off on the pre-school delivery run leaving her phone in the upstairs bathroom (where the tablet had no trouble connecting to it), she rapidly learned that something was wrong.

Back they came and (using her smartwatch) she called me from the driveway to tell me she didn't have her phone. Cue double take.

New use for the table: phone abandonment detector.

The Rip-off from Cleveland Depths.

Title in reference to Fritz Leiber's 1962 novella "The Creature from Cleveland Depths".¹

The complex of product options that go by "activity tracker", "health monitor", or "wellness monitor" (and smart watches that include that suite of functionality) aren't exactly the "tickler"s in the story, but they share a lot of DNA with them. They are, after all, automated systems we wear explicitly so they can nudge us toward better versions of ourselves. You can get other parts of the tickler package as apps. Either for your phone or, if you choose the right smart watch for the wrist mounted device itself.²

One of the psychologically interesting things is that the prime statistics these things recorded when they first took off was ... steps. They were glorified, electronic pedomenters.³ Pedometers have been around for ages and never made such a big splash before. I suppose gamification is the key to that. Anyway, today's rantlet is about the way a wrist mounted pedometer fails us: you don't get credit for pushing a shopping cart. I can trek up and down every aisle of the local, big-box, warehouse club and the fool machine registers a few hundred steps because pushing the fool cart more or less immobilizes my wrist. I only get credit for putting things in the cart. The old waist-mounted mechanical jobs managed that kind of use better.

Okay, so it's seriously a first-world problem. You're right. And it's of no actual consequence. You're right about that, too. It still annoys me.

---

¹ Available from Project Gutenberg or in audio form from Librivox.

² I wear a Withing ScanWatch so I can't get a bunch of apps on my wrist, but I do get reasonably comprehensive health tracking, weeks of battery life (I only every charge it while I shower), and the aesthetic appeals to me.

³ Indeed, the cheapest ones still are, but many do a lot more these days.

Changing media landscapes and "smart" TVs

We have two smart (or at least occasionally clever) TVs in the house. Neither of them was a fancy model because we got the first one while I was still badly underpaid and we got the second one while I still had the habits of the badly underpaid.¹ One of them simply has a fixed set of apps in it'e ROM; upgrades are not possible, but then neither are certain classes of persistent attacks that require the attacker to make local changes.² The other, the older one as it happens, has a loadable app system, but the manufacturer has stopped updating the existing apps or providing new ones.

Our daughter watches various shows on her tablet and on our TVs. Because they are smart TVs we're able to, for instance, bring up things on Netflix. We haven't felt the need to restrict her screen-time yet because (a) her pre-school is a screen-free zone and (b) she spends as much or more of her time at home engaged in play with physical toys as she does on screens. We figure as long as she is becoming adept at the real world we won't fret about her interest in the virtual world (though we do curate what apps she can have on her tablet).

Anyway. The kiddo has some new favorite shows that are on a platform that is new to us. Putting it on our laptops and her tablet is easy enough, but we can't get it on either TV because it wasn't a thing when they froze the available content.

Which is a problem.

I'm thinking that in the future I'll get a dumb TV well supplied with ports so that I can plug in a computer of some kind (be it a streaming stick, a raspberry pi, or some kind of mini IPX machine). so that we can upgrade the "smarts" for as long as the display is useful. Unless some manufacturer is making smart TVs based on an open platform. Anyone know? For that matter are their any streaming sticks built on open platforms?

---

¹ Honestly I"m trying to hold onto certain of those habits even though our financial situation is rather improved. The trick it to figure out which of those habits are a good idea in general and which would be better replaced by buying a little further upscale.

² If I had known this about it prior to buying I might have hesitated. But it was a deal of the day Woot just when we needed a larger TV, so the clock was ticking. It is, at least, a nice display.

Audio goodness

I have a gamebox¹ and like to play games where the sounds carries information. So I need stereo, and I need enough volume to hear what is going on in the game soundscape.

But ...

• These games can be pretty noisy and other people in the house don't want or need to be disturbed by my entertainment
• The household also includes a toddler and a couple of people with moderate to severe health issues. I need to be able to hear when I have to stop playing right now and go take care of something important.

So getting a good pair of over the ear headphones and shutting the doors (the go-to solutions for hearing your game) are not really good ideas.

Now, a couple of years ago I spotted a solution: a wearable speaker. But at the time my finances were rather tight and before they recoverered I had forgotten. Then one popped up on woot: a JBL Soundgear.

The experience of using this thing for gaming is immersive. The sound is rich and seems to emenate naturally from all around you. It's hard to believe that everyone around you isn't hearing the same thing, but if you hand the device to a friend to try out you hear essentially nothing while they're wearing it. Of course you still hear local environmental noises which isn't ideal if you are listening for a telltale footfall in-game, but I chose this route because I need to be connected to my real-world environment.

These things might not be for everyone, but I'm pleased.

---

¹ My gamebox is a PlayStation 4. Alas, Sony cheaped out on the bluetooth audio support (in an effort either to save a few cents per unit or to squeeze a few more bucks out of the consumer by controling the supply of compatible devices. You'd find it very easy to convince me of the latter. So I also bought a Avantree DG80 USB to bluetooth audio device. Works like a charm.

Little surprises #6: "Stop helping me!" edition

I got a little technological help today. Unfortunately it was not the helpful kind of help.

I attend some remote precense meeting that are mediated by a bespoke telephone-based system rather than by internet. This is driven in part by the institutional inertia that comes with having had the system for a long time, but mostly by security concerns. So I dial in, switch to speaker, and prop my cell phone against my monitor stand. It has to be a cell phone because we don't have a landline.

The conference involves our sponsors, so we keep a separate chat channel open for communication amungst ourselves.

On the conference today I finsihed answering a questions and waited for a response. And waited. And waited. And noticed that the line seemed deader than usual. I enquire on the chat channel if there is a problem with the conference, and a colleague (another parent of young children as it happens) writes back "I don't know, but someone has to get in the car seat to get a ice cream."

A horrified glance out the window is enough to confirm the worst. There is the babysitter loading the toddler into my wife's car. The phone has linked to the car and my colleagues and sponsers are hearing all about the proposed trip to get ice cream.

I want it to do that, but not now

I have no one to blame but myself. Unless I can blame Apple a little, too. That sounds good. I'll do that.

Usually I want the phone to pair with the car as soon as it comes on and without asking me if that is a good idea. But maybe the behavior should be varied when there is already a phone call in progress? Especially if I am already using the speaker. Or something. I'm frustrated that progress in AI has produced a lot of powerful spying tools and general creepiness, but damn all context sensitive decision making for real life.

Phone case suggestions?

The snap-in belt holster for my iPhone 8 is worn out. The phone falls out at undesirable moments several time a week. The phone itself has been fine as the case (an OtterBox defender) is pretty sturdy, but I don’t want to lose the fool thing.

I’m looking for recommendations on a new holster (or a whole new case).