Bill Gaver's 1989 Demo of The SonicFinder™

While digitizing my analog videotapes, I came across a short 1989 demo by Bill Gaver of his SonicFinder™, an interesting piece of early work on audio in user interfaces. It's easy to find online copies of research papers describing The SonicFinder™, but I was surprised to find that the Internet doesn't seem to offer anything like the video demo I have. That's a shame, because hearing audio and seeing it in action is a lot more illuminating than reading about it.

With Bill's kind permission, here is a digitized copy of the videotape he gave me in 1989 (presumably when he came to give a talk at Brown University, where I was pursuing graduate work in Computer Science). It's short--not quite three minutes long. Enjoy!

Escaping from iCloud Photos

After more than a decade storing photos and videos in iCloud, I decided to download my stuff to my PC and delete the copies at Apple. I figured I'd log into iCloud using my web browser, click the Download link, wait a while for the eighty-plus gigabytes to make their way to my computer, delete the cloud copies, and be done with it. 

It didn't work out that way. 

What I expected to take a few minutes ended up taking more than a week of screen-staring, head-scratching, teeth-gnashing research, installation of a couple of new apps, and a bunch of manual work that I didn't know how to avoid until most of it was already done. 

A good place to start the story is the term "my stuff." I had photos and videos at iCloud, which I'd organized into a few albums, and that's what I wanted to download: photos, videos, and albums. For now, we'll set aside albums and deal with photos and videos.

iCloud Photos is unambiguous about how many it stores for me:

That's a total of 23,251 items. However, if I install the iCloud for Windows app and sync my photos, this is what File Explorer shows me:

That's 958 more items than iCloud says exist. Not a good sign. Things got worse when I did the same thing with Cloudly, a program I tried in an attempt to deal with the albums issue I'll address later. Here's what File Explorer has to say about the result of syncing Cloudly with iCloud Photos:

That's 1,717 more items than iCloud reports and 759 more than iCloud for Windows downloaded. How can there be so much trouble simply agreeing on how many photos and videos are to be downloaded?!

The answer involves a detour into Apple's options for downloading. 

Apple Download Options

In the web browser interface to iCloud Photos, you can click a simple download icon to get what I'll call the Default download. You can also click on the "..." icon and then on "More Download Options..." to select from three more choices. They are Unmodified Originals, Highest Resolution, and Most Compatible. My simple experiments suggest that the Default and Most Compatible options are the same, so the iCloud web site effectively offers these download choices:

• 
  
  Unmodified Originals.This means exactly what it says. It is probably not the droid you're looking for. 
  
  When you take a photo with an Apple device and edit it (e.g., crop it or adjust the colors), Apple saves both the original photo and the result of your edits. The edited photo is what you see in iCloud (and in the Photos app on your device), but the original is still lurking in the background in case you decide you'd like revert to it. The same applies to videos. Using the Unmodified Originals download mode gives you your pictures and videos just as you took them--without any of the edits you applied to make them look better.
  
  Sometimes edits are applied automatically, and you may not realize they've been performed. A good example is photos taken in Portrait Mode. The only image you see on your device is the one with the background blurred, but behind the scenes an ordinary (unblurred) photo is taken and some transformations (i.e., edits) are applied. If you download the Unmodified Original for a Portrait Mode photo, you'll get the truly original one, the one without any blurring. 
  
  The iCloud item count doesn't include the original versions of edited photos. However, when Cloudly downloads edited photos, it downloads both the final image and its first forebear. That's one of the reasons Cloudly downloaded so many more items for me than iCloud said it contained.
  
  
• Most Compatible (the default). Apple says that using this download mode yields JPEG photos and videos that are MP4/H.264, the idea being that these formats are more widely supported (especially on non-Apple systems) than the HEIC photos and H.265 videos Apple devices typically produce. In contrast to Unmodified Originals mode, the files you download include edits you've applied.
  
  What Apple fails to mention is that Most Compatible downloading incurs a significant reduction in resolution in the photos and videos you end up with. In my testing on originals taken on an iPhone SE 2, photos generally went from 3024 x 4032 to 1536 x 2048 (a pixel loss of 74%) and videos from 1920 x 1080 to 1280 x 720 (56%). 
  
  I can imagine why Apple made this the default download option. Users would find it frustrating to download something from iCloud and then not be able to view it. However, I think they should be a lot clearer about the implications that has for the quality of the downloaded content.
  
  
• Highest Resolution. As its name suggests, this download option yields photos and videos with the full resolution available. The versions you download include the edits you've made to them. 

If this were the end of the story, I'd download my stuff at Highest Resolution and be done with it. But the story's not finished.

File Timestamps 

Of the 23,000+ items I have in iCloud Photos, several thousand were sent to me as WhatsApp messages. WhatsApp strips out the internal metadata for, among other things, when a photo or video was taken, so the file itself can't tell me when the item was created. I can approximate it, however, by using the date and time when I got the message. That's enshrined in the creation timestamp of the file that held the photo or video on my device and was subsequently copied into iCloud. That means that when I download something from iCloud onto my PC, I want the creation date and time of the file on my PC to match the creation date and time of the file in iCloud Photos. Otherwise I'd have to review all my old WhatsApp messages to determine that, say, this photo was sent to me on March 13, 2017:

None of the download options above--the ones offered by the web browser interface to iCloud Photos--preserve file creation timestamps. If I download that photo using a web browser today, the file creation timestamp will be for today. Nothing about the picture or the file will tell me that it's from 2017. 

For me, that's a deal-killer. Downloading using a web browser is out. (It was already pretty close to out, anyway, because iCloud limits browser-based downloads to 1000 items at a time. Downloading my stuff would have required that I select and download batches of items 23 times. That's doable, but ridiculous.)

Fortunately, Apple has an alternative way to download photos and videos from iCloud to a PC: the iCloud for Windows app. Its design approach is quite different from that in the browser. There no choice in which type of download you want (nor is the type it uses documented, from what I can tell), but my testing suggests that it's the same as Highest Resolution. That's good for me, because it's what I want, but even better is that files downloaded using iCloud for Windows retain the creation timestamps they have in iCloud. That means the photos and videos sent to me in old WhatsApp messages are in files with creation dates and times corresponding to when I received them. That, in turn, will allow me to use exiftool to copy those file-based timestamps into the files' metadata, thus restoring (more or less) the "date taken" information that was originally there and that WhatsApp got rid of.

iCloud for Windows thus solves two-thirds of the problem of getting my stuff from iCloud Photos: the parts for photos and videos. For the final part--albums--iCloud for Windows is sadly and infuriatingly useless.

Albums

iCloud has two kinds of albums, shared and unshared. There are important differences between them (which Apple doesn't seem to directly describe anywhere), but for my purposes, both are just collections of photos and videos that are in my photo library. When downloading my stuff, I want to download these collections, too. At the very least, I want to download representations of the albums I have and the photos and videos that are in each of them. Almost as fundamental is the ability to have the downloaded representations retain any custom ordering I've done on each album's contents. That is, if I've arranged the photos and videos in an album to tell a story in a particular way, I want my arrangement to be preserved.

For Windows users, Apple offers no support for any of this. You can do things manually, of course (e.g., create folders in your PC's file system for each of your albums, use the iCloud web interface to download album contents (in 1000-item batches) into the folders, then use software like FastStone Image Viewer to arrange and rename the items so they stay in the order you want them in, etc.), but you're completely on your own. It's dismissive--even insulting--on Apple's part.

If you have a Mac, things are better. There you can use the Photos app to export album contents to the file system. Using the Sequential File Naming option gives the exported items names such that when you list them in alphabetical order, you get them in the order shown in the album. File timestamps are correctly preserved. 

The only thing missing is the ability to export all albums at once. You have to manually export them one by one. It's a lot better than the nothing offered Windows users, but it's not perfect. The more albums you have, the further from perfect it is.

I have a Mac, but I preferentially use Windows, so I scoured the web looking for a way to download albums from iCloud. Pickings are slim, but I eventually came upon Cloudly, a program promising to "[preserve the] album structure of your iCloud Photo Library." It's a largely hollow claim. Cloudly downloads only unshared albums (not shared ones), and it doesn't preserve item orderings within the albums. As for the downloads themselves, they're, um, quirky. 

For edited photos, Cloudly downloads both the unmodified originals as well as the highest resolution versions. For videos, it downloads unmodified originals only; there is no way to download the revised versions of videos you've edited. On the plus side, file timestamps are properly copied from iCloud.

Shared albums are downloaded as separate copies of the items in your iCloud library following the rules above, so if you have an edited photo in five unshared albums, you'll end up with six downloaded copies of the unmodified original--one for each of the albums plus one for the library as a whole--and six copies of its highest resolution version. (Similar file replication results from the Mac-album-export approach: each exported album contains its own copy of the Highest Resolution version of each item in the album.)

Cloudly's only download option is for an entire iCloud library. For me, Cloudly said when it finished that about 500 items had failed to download. It gave me a "retry" option, which caused it to try to download the missing items again. It got about half of them, thus leaving me about 250 items short of my full library. I clicked "retry" again, and the number of un-downloaded items was reduced again. After about a dozen iterations of this, it finally told me that everything had been downloaded. 

My experience with Cloudly made me wish I'd been less reluctant to go the Mac route...

"Downloading My Stuff" Revisited

I can now be more precise about what I mean when I say I want to download my stuff:

• I want a copy of all my photos and all my videos in the best available quality. For edited items, I want the most recent version. I want the file creation timestamps to match those in iCloud.
• I want all my albums (shared and unshared) to have some kind of representation (presumably folders) in what I download. I want the order of the items in my albums to be preserved. 
• I want the downloads to proceed without manual intervention. 

What I don't want is to have to break my downloads into 1000-item batches or sacrifice file-creation timestamps (as the iCloud web browser interface demands), to have my albums ignored (as iCloud for Windows does), to have to export albums manually (as the Mac Photos app requires), or to have the item orderings in my albums disregarded (as Cloudly does). Pardon me for channeling the feelings of every user of every piece of software everywhere, but really, people, I don't think I'm asking for anything unreasonable!

An iCloud Photos Escape Plan

If you find yourself wanting to get your stuff out of iCloud and on to Windows, this is my suggested approach:

• Install iCloud for Windows and let it sync. This gives you high-quality copies of your photos and videos with the correct file timestamps.
• If you have a Mac, configure its Photos app's iCloud settings to "Download Originals to this Mac", then export the albums one bv one (sigh) into folders named after the albums. Specify sequential file naming for the exporting. Copy the resulting folders to Windows.

I'm a little abashed to admit that this is pretty close to what ChatGPT suggested when I asked it how I could get my stuff. In my defense, I also asked Perplexity, Grok, Claude, Gemini and Copilot, and none recommended using a Mac. Copilot even implied that ChatGPT's approach wouldn't work: "The native Photos app on macOS does not preserve album structure when exporting..." At the time I asked these chatbots for advice, I was hoping to stick with a Windows-based solution. I didn't try using a Mac for albums until I'd gone down a bunch of Windows-centric dead ends.

If you don't have a Mac, things are more complicated, but ChatGPT had some suggestions. In an act of contrition for failing to take its advice as seriously as I should have, here's a link to our discussion. My contrition is minimal, however, because, well, read on. 

Aside: Chatbots Again Fail as Research Assistants

All the information above is based on experiments I ran, but when deciding which experiments to perform, I sometimes took the opportunity to test LLM chatbots by asking them questions I needed to answer. Here's one of the questions I posed (in each case to the free version of whichever chatbot I was using):

When downloading photos and videos to a PC from iCloud using a web browser, there are three download options: Unmodified Originals, Highest Resolution, and Most compatible. When downloading using the iCloud for Windows app, there are no download options, the files are just downloaded. Which web browser download option does iCloud for Windows use?

The results were terrible. Grok, Perplexity, Copilot, and Gemini all confidently stated that iCloud for Windows downloads unmodified originals. ChatGPT was equally sure that the answer was Most Compatible, and Copilot agreed (though with less certainty). All of them were wrong. iCloud for Windows uses Highest Resolution. At least it does for me. All that chatbot opposition spurred me to check many, many times. 

As I've said before (e.g., here and here), LLM chatbots--or at least the free versions thereof--have a long way to go before they'll be reliable research aids. I continue to look forward to their getting there.

 

Video File Metadata: First Thoughts

A few years ago, I published a series on metadata in image files with a focus on old slides and photos that had been scanned. I've since turned my attention to metadata for video files, because I had a number of old home movies digitized. 

I assumed that most of what I'd learned about image file metadata would apply. I also assumed that because widespread consumer use of digital video came after mass adoption of digital photography, the inconsistent, competing, overlapping metadata standards for images would have been avoided and that a single, widely-supported standard for video metadata would exist. I was wrong on both counts. I was reminded again of the the remark at Stack Exchange Photography that "Image and video metadata is a complete hot mess." 

The video mess seems even hotter than for images, because

• There's no accepted standard for video metadata. That's worse than the three competing standards that bedevil metadata for image files.
• It's harder to find video players that show metadata. 
• Google Photos (GP) exhibits less predictable behavior for video file metadata.

Nevertheless, the case for metadata in videos is as strong as it is for images, so into the mire we wade!

Note added 8/8/25: All my GP testing was done with the "Storage Saver" quality setting, whereby "photos and videos are stored at a slightly reduced quality." I don't know whether "slightly reduced quality" encompasses changes to metadata, but it would not surprise me if it does. 

Data to Store

I store the following information in image file metadata, and I want to store the same things in video files:

• A description of what's in the video. 
• When the video was taken.
• When the video was digitized.
• Who did the digitizing.
• A copyright notice.

I also want to store something I can't believe I overlooked for image files:

• GPS coordinates for where the video was filmed.

My excuse for this oversight with images is that when I started on the metadata problem, I was mostly dealing with pictures where I had only a vague idea where they were taken.

Metadata Fields to Use

In a perfect world, there'd be a widely-supported standard for video file metadata, and that standard would prescribe which fields (i.e., tags) should be used for the information I want to store. Heck, in a really perfect world, that standard would apply not just to videos, but to images as well.

Our world falls short of perfection. Fields for image file metadata are largely ignored in the video world, and while there is an IPTC standard for video metadata, it has nothing to say about digitization, e.g., when it was done and by whom. Furthermore, my experiments indicate that support for the IPTC standard by Google Photos (GP), Windows File Explorer, and other programs I care about is poor. 

I've written elsewhere about how GP's search capabilities make it indispensable for me, and that indispensability gives it a lot of clout in the metadata decisions I make. For images, I didn't find that to be constraining, but for videos, GP seems to like to throw its weight around. One of my goals is to be able to round-trip a video to GP (i.e., upload a video, then download it) without a loss of metadata. GP seems to enjoy interfering with that process. Often, fields present in uploaded videos are missing when those videos are downloaded. Sometimes GP keeps the data in downloads, but puts it in different fields. It's irritating.

I aspire to use metadata fields with these features:

• They are part of the IPTC standard. 
• They are visible in commonly-used programs (e.g., video players).
• They have these GP-related characteristics:
• Are displayed in the web browser UI.
• Are consulted by GP during searches. For example, the results of searching for videos taken in 1998 in California should include videos with such information in the metadata for when and where they were taken as well as in the videos' "description" field.
• Survive round-tripping.

I didn't find any fields that check all these boxes, so compromise was the name of the game.

Identifying and testing candidate fields was a big job. I initially got help from a capable human, Amelia Huchley, but her time was limited, and when it ran out, I turned to marginally-capable AI chatbots. They've read pretty much everything on the internet, I figured, so they should be able to help choose metadata fields suited to my purposes.

I've blogged about how AI chatbots--or at least the free versions of them--are lousy search assistants, but desperate times, any port in a storm, etc. I turned to ChatGPT, Perplexity, Claude, Gemini, and Copilot and sought to use them as a sort of advisory committee to help me identify prospective fields and evaluate how well each was likely to achieve my goals. As advisory committees go, this one left a lot to be desired. The chatbots helped me put together a list of candidate  fields, but their advice on how well the fields were likely to do on things like being displayed in the GP interface or surviving GP round-tripping was unreliable to the point of uselessness. This reinforces my impression that, in their current form, the free AI chatbots are worth about what you pay for them when it comes to internet research.

 In the end, these are the fields I decided to use:

• When taken: QuickTime:CreateDate. Choosing this field should have been easy. From a metadata perspective, you can't get much more fundamental than when a video was made. I considered six fields for this information. What I found perfectly exemplifies why video file metadata is a mess. 
     Four of the fields come from QuickTime--meaning that QuickTime has four metadata fields for the same information. Two have ridiculously similar names: QuickTime:CreateDate and QuickTime:CreationDate. Two have names differing only in namespace: Keys:CreationDate and QuickTime:CreationDate. (ExifTool unhelpfully treats these fields as the same, sigh.) Two of the fields are changed into a third during a GP round trip: uploading files containing either of the two CreationDates yields downloaded files where those fields have become QuickTime:ContentCreateDate. You get the idea.
     I ultimately settled on QuickTime:CreateDate, because it has good GP support (showing in the UI, being found in searches, and remaining intact when round-tripped) and, in contrast to the other candidates, is also visible via Windows' File Explorer and MediaInfo.
• Where taken: QuickTime:GPSCoordinates. I identified six candidate fields for this information, but I tested only four, quitting when I found that this one offered full GP support, unlike the others I tried. 
     "Full" GP support is quirky. Copying GPS coordinates from Google Maps and pasting them into an ExifTool command line to set QuickTime:GPSCoordinates works, but if the coordinate values have six or more decimal places, GP will refuse to show where the video was shot. When coordinates have at most five decimal places, no such problem arises. Enlivening the situation is that (at least under Windows) right-clicking a location in Google Maps brings up a context menu showing the location's coordinates to five decimal places, and clicking these coordinates copies them to the clipboard, but pasting them yields values with up to 14 decimal places! So when pasting coordinates copied from Google Maps into an ExifTool command, the values have to be edited back to at most five decimal places. (This restriction doesn't seem to apply to image files. In that case, you can use ExifTool to write coordinates with up to 14 decimal places, yet GP will still show where the photo was taken.)  
• Description: QuickTime:Title. Amelia and I tested a dozen fields in an attempt to find one that GP would display. None did. Many also failed the round trip test. When I found that QuickTime:Title was found in GP searches, was retained in GP round-trips, and was displayed in VLC Media Player, Windows Media Player, Windows Media Player Legacy, and Windows Explorer, I figured that no field would do better, and I stopped looking.
• When digitized: XMP:DateTimeDigitized. This was the only field I found that is specifically designed for when digitization took place, can be set by ExifTool, and is retained across round trips to GP.
• Who did the digitizing: XMP-xmpDM:LogComment. I was not able to find a field designed to store this information (Claude suggested the apparently-non-existent XMP:DigitizedBy), but the general-purpose XMP-xmpDM:LogComment seems like a reasonable choice, and it survives a round trip to GP. 
• Copyright: QuickTime:Copyright. This was the only easy metadata field to choose. It's specifically designed to hold copyright information, it's retained across round-trips to GP, it's part of the IPTC standard, and it's visible in at least one video player (QuickTime Player).

These fields are my preliminary choices. To date, my experience with them is mostly limited to simple tests. My thinking may change as I get more experience with real video files, but given the dearth of guidance about metadata in video files, I thought it would be useful to document where I am now. That's why this post has "First Thoughts" in its title.

Adding Transcripts

An important part of a video file is its audio track. For many videos, there's a lot of speech in such tracks, and it's easy to imagine uses for transcripts of what's said, e.g., subtitle generation and full-text search. The transcript for a video is a form of metadata, so, at least in concept, I'd like to include it in the video file. Given the availability of speech-to-text software, it's not unreasonable to hope for initial drafts of such transcripts to be generated automatically. (YouTube, Vimeo, and Dailymotion all offer automatic generation of captions or subtitles.) I haven't gone beyond the thinking-about-it stage for generating and embedding transcripts, but it's clear that this could be a useful avenue to explore.

On the other hand, it's possible that this is an area where advances in technology could render the issue moot. If speech can be converted to text quickly and accurately enough for real-time display and search, there'd be no need to create and store transcripts inside video files. The metadata they represent could simply be generated when needed, thus sparing people like me a lot of work. 

Convertible EV Spot Check

It's been more than two years since I last looked into the availability of all-electric convertibles, so I thought I'd do a quick check to see where things stand now. Here's a rundown (in alphabetical order) on the models I mentioned in that post, plus a couple of others I've added since then:

• Fisker Ronin: Fisker went bankrupt.
• Genesis X Convertible: Still a concept car with no production announced.
• Maserati GranCabrio Folgore: orderable, but it's not clear that deliveries have occurred. Pricing starts at $206,700 in the United States.
• MG Cyberster: available in the UK and Europe, but not in the United States. Pricing starts at £54,995 in the UK.
• Mini Cooper SE Convertible: Canceled. 
• Polestar 6: Not yet available. Estimated for model year 2028.
• Porsche 718: Postponed and not yet available. Estimated for model year 2027.
• Smart Fortwo Cabrio: Discontinued.
• Wiesmann Project Thunderball: Postponed and not yet available.

All in all, a pretty gloomy picture for those of us who yearn for an EV analogue to the Miata. Sigh.

The Dismal Failure of LLMs as EV Search Aids

For six years, I've been in the market for an electric compact SUV. I haven't yet found one with the features I want at a price I'm willing to pay. (As a rule, missing features have been a bigger impediment than price.) My last review took place about a year ago, so I decided it was time to look again.

This time I experimented with seven LLM-based chatbots as search assistants. I gave each the following prompt:

List all the fully electric compact SUVs for sale in the United States that have all-wheel drive, an openable moonroof or sunroof, an all-around (i.e., 360-degree) camera, an EPA range of at least 250 miles, and are no more than 180 inches in length.

My assistants were the unpaid versions of these systems (listed in the order in which I happened to test them):

• ChatGPT
• Perplexity
• Claude
• Gemini
• Copilot
• You.com
• Mistral
  

The results were eye-opening. None of the systems listed the only vehicle that fulfills the criteria (the Volvo EX40), and all but one listed vehicles that violate the requirements. Worse performance is hard to imagine. The false positives waste your time pursuing dead ends, while the false negatives imply that no qualified EVs exist, even though one does.

Complete failure was averted by one system (You.com) mentioning, almost as an afterthought, the Volvo XC40 Recharge. That car was renamed the EX40 last year, but searching for the old name will quickly lead you to the new name, and that will finally put you on the trail of the only car that satisfies my criteria.

[Update 5/28/25: Per the comments on this post, paid versions of at least ChatGPT and Claude produce much better results than the ones I experienced. I've added links to the conversations I had with the various unpaid chatbots in discussions below.] 

The chatbots failed in a variety of ways (the links are to my conversations with the chatbots):

• ChatGPT said "here are the models that meet all requirements," then listed five EVs and their specs. For four of the five, the displayed specs were contrary to the requirements, meaning ChatGPT "knew" (to the extent that LLMs "know" things) that these cars shouldn't have been listed. For the fifth car, one of the specs it listed was simply incorrect.There was no mention of the Volvo EX40.
• Perplexity's behavior was similar to ChatGPT's: it claimed to list cars fulfilling the criteria, then "knowingly" listed ones that don't. The twist was that two of the three EVs Perplexity listed--the Rivian R2 and the Toyota C-HR EV--don't exist yet. (If they did, the Rivian would exceed the length constraint, and it looks like the Toyota would likely fail the openable-roof test.) Perplexity made no mention of the Volvo EX40.
• Claude listed only one car, Tesla's Model Y, saying it "clearly meets all your requirements. It offers ... a panoramic glass roof ... and measures approximately 187 inches in length. However, this exceeds your 180-inch length requirement." Its dithering on the car's length was disappointing, and its failure to distinguish a panoramic glass roof from one that opens was worse. There was no mention of the Volvo EX40.
• Gemini's response began with "Here's a breakdown of current and upcoming electric compact SUVs and how they stack up against your requirements," which was not what I had asked for. It listed five cars and their specs, noting for each vehicle the specs that violated my criteria. It ultimately concluded, "there may not be any currently available fully electric compact SUVs that precisely meet all conditions." Except there is, of course.
• Copilot produced a refreshingly short response featuring a very nicely formatted table that summarized the two EVs it said satisfied my requirements. Neither does, though Copilot showed no specs that reflect that, so it may not have "known" it was wrong. There was no mention of the Volvo EX40.
• You.com started with this rather confusing statement: "Based on the search results provided, there is no direct information listing fully electric compact SUVs in the United States that meet [your] criteria." It then launched into an explanation of how to perform my own search (<eyeroll/>). Then came a surprise. It introduced the Volvo XC40 Recharge and showed how it satisfied my requirements, though it seemed unsure of itself: "The Volvo XC40 Recharge appears to meet all your criteria. However, I recommend verifying [everything]." Ultimately, You.com found the rabbit in the hat and pulled it out, but its response was confusing and disjointed, and it referred to the rabbit by an obsolete name. 
• Mistral followed Copilot's lead in producing a short, clear response built around a well-formatted table of information that was often incorrect or inconsistent with my requirements.There was no mention of the Volvo EX40.

As a group, the systems produced responses rife with claims that were incorrect, inconsistent, and/or incomplete. The last of these is the most disturbing. Six of the seven systems didn't mention the only SUV fulfilling the requirements. The one that did hid it after an explanation of how to do your own search, and even then it referred to that car by an outdated name.

There is a lot of work to be done before LLM chatbots are reliable search assistants.

Three Experiences with Video and AI

Finding an Old TV Episode

I recently found myself wondering about a TV episode I saw decades ago. I had only the haziest memory of it, so I threw this at Gemini:

I'm looking for an episode from the original TV show The Outer Limits or The Twilight Zone. The story is about a man with a robotic hand that he has to add fingers to in order to increase its ability to help him figure out what is happening. Do you know this episode? 

Gemini did, correctly identifying it as "Demon with a Glass Hand" from the 1960s series, The Outer Limits. Googling for that yielded a link to the episode at The Internet Archive, which I downloaded and added to my Plex server. 

Less than 15 minutes elapsed between the time I thought about the episode and the time I had it in my video library. It's not the best television content in the world, but I marvel at how easily I was able to track down and watch a show from 60 years ago based on only a very sketchy memory.

Upscaling the Episode

"Demon with a Glass Hand" isn't terribly compelling, but that doesn't mean it shouldn't look good. Unfortunately, 1963 TV was SD, and these days we're used to a lot better resolution than the 496 x 368 I got from the Internet Archive. 

Earlier this year, I purchased a copy of VideoProc Converter AI to experiment with upscaling low-resolution 8mm family videos I'd had digitized. The results were impressive on everything except faces, which the upscaling process tended to turn into grotesque caricatures of the people behind them. But hope springs eternal, so I decided to see what VideoProc could do with "Demon with a Glass Hand." 

Invoking the program yielded a message excitedly telling me that a new version was, you know, faster and better, and I should upgrade immediately. It was free, so I did, but I didn't expect that V3 would be noticeably better than V2. When was the last time a program upgrade lived up to its PR?

In this case, I think it does. Check it out:

Upscaling is an interesting challenge, because it involves fabricating information (pixels) not present in the original images. Simple interpolation doesn't do a very good job, and VideoProc's V2 AI-based approach fell apart on faces. V3's faces aren't perfect, but I think they're good enough for casual viewing, and that's an impressive accomplishment.

Looking Forward

A few days ago, Andrei Alexandrescu brought my attention to this reddit post featuring a synthesized video by Ari Kuschnir using Google's Veo. The clip takes advantage of Veo's new ability to generate audio tracks, including dialogue and singing. I find the clip pretty amazing. There are legitimate questions about how Veo was trained and how its output could be used for ill, but I prefer to focus on the technical progress it represents and the creative promise it offers. 

Incongruously, I was reminded of the Veo demo after viewing another old TV episode I barely remembered, one Gemini identified from this prompt:

I'm now thinking of a different episode, again from The Twilight Zone or The Outer Limits. It involves a man who goes to a store to custom-order a woman. He chooses eye color, etc. Any idea which episode this is?

Again Gemini knew what I was looking for ("I Sing the Body Electric" from the original The Twilight Zone), Google found a downloadable link to it, and I had it on my Plex only a few minutes after issuing the query. 

The episode is quite terrible (much worse than "Demon with a Glass Hand"), but I liked the hopeful ending. Not the part summarizing grandma's data collection and sharing policy ("Everything you ever said or did, everything you ever laughed or cried about, I'll share with the other machines"), but the optimistic sentiment behind Rod Serling's closing voiceover:

Who's to say at some distant moment, there might be an assembly line producing a gentle product in the form of a grandmother, whose stock in trade is love?

For countries with an aging population requiring increasingly attentive personal care, I'd expect that gentle, loving robots rolling off an assembly line could be a pretty attractive prospect.