Old Business: Fixed the missing link in the Comparative Reading post. You can now go directly to the IPR post “TDS Not a Monolith but a Coalition.”
There are dozens of Robert Lairds on FB. Part of my under the
censorship radar strategy. I’m the one who uses this profile icon.
Pardon me if I’m not assuming that you visit my Facebook page directly. My wife continually misses about half my posts because she still responds to notifications instead of looking me up directly. I’m mentioning this now because I’ve spent a fair amount of time lately whipping up little “reels” as Facebook has renamed videos. (Method in their madness: there used to be a collection of videos on your FB page, including vids longer than a minute or two. No more.) I’m using “reels” as a test platform for low-grade freebie AI apps that turn pics into spoken word videos matched to canned responses by celebs. My reels feature graphics made by me, matched with canned responses that are frequently replaced by my own audio clip.
I’ve observed a couple of benefits so far that partially offset the loss of the FP page list of videos. First, there are traffic numbers shown for each reel; i.e., number of visits and comments. Since I’ve never been able to derive traffic info about my own FB pages, the visitor numbers are a help. Second, the visitor numbers are a rough measure of which people and issue categories matter to casual readers. The numbers show big differentials that amount to a rule-of-thumb polling mechanism to see who is interesting at the moment and who isn’t.
Why I’m suggesting you make periodic visits to my FB page for the purpose of checking out the reels section. Here’s how you get there. Search for my page from yours. Click on the ‘Robert Laird’ with the avatar shown above. Then find reels in the page header display:
The choice here is helpfully underlined. It takes you to the following…
Click to see individual reels. Scroll to see more reels. By now now there are dozens.
My biggest draw so far was a Tim Walz reel that drew 39,000 visitors. It hit both important factors: 1) it was genuinely funny even on first appearance, and 2) he’s a guy people are really fired up about. That second criterion is actually the more important of the two, but when it comes to reels I’m in the funny business first, the political horse race second. If 1) and 2) are both on the money, people will share and the numbers will skyrocket quickly. Who knew that Jane Fonda is still in the “Never to Be Forgiven”category?
Trends I’ve noticed. People still track celebrities, no matter how much they hate them or claim to pay no attention. (Certain ones, too much in the news, fade quickly in their appeal.) Crooked judges, even quite anonymous ones, are a surprisingly good draw. The people who go reels shopping are keeping an eye on the latest anti-Trump, anti-American rulings by district judges from ritzy suburbs. Older Dem leadership like Pelosi, Schumer, Carville, Warren, and Sanders are almost fishwrap at this point, no matter how funny the punchline. The younger leadership tracks more with the noisiest celebrities, popular for sharing at first, then diminishing returns as the insanities pile up.
If you cruise through the reels I’ve posted, you can spot your own trends about who’s a hot topic and who isn’t. That’s not how I make my choices; I’m the old satire warhorse who’s mostly looking for a good punchline in the fancies and AI tricks at my disposal, preferring to trust my own instincts about my target audience. Just like I’ve made no effort over the ears to build a huge total of Facebook friends. [gotta break in and tell you this; just typed ‘warhorse’ and AutoCorrect edited it to ‘Warhol pres’ as I was writing the next line (hit a ‘p’ by accident, did I?). That’s AI super-intelligence for you…]
And now that it’s come up, AI is something that’s important here too. Facebook is using AI to create more sophisticated algorithms intended to reduce audiences rather than flat-out censor posted material. I know there are shadow bans (plural) on me, in terms of both posts and reels, and they are subtle enough not to be too blanketing. Some reels take off like gangbusters as soon as I post them. Others, featuring faces and topics that are sizzling hot, barely draw for 8 to 10 hours before trending toward modestly healthy numbers. Treatment of posts is just as suspect, but even more subtle. I have loyal readers who seem to take days to see or comment on any post, while others are putting in their ‘Likes’ from the moment I hit ‘Save’ it seems. FB also does not like to show teaser graphics for posts linking my page to Instapunk Returns, my long format blog, which can make posts with short intros almost disappear from the scanning eye of a reader. Some posts, often the more provocative ones in terms of subject matter, go for days with no Likes or comments at all. Not included in notification strings? I can only guess.
Reels is confirming some of my observations to date about AI algorithms. As sophisticated as it seems, the technological capacity to make lips move as if pronouncing specific words does not include the actual ability to “see.” The algorithm is designed to sample very small segments of an image that are ‘facelike’ or ‘mouthlike’ and put in them in a work-in-process buffer. What the algorithms are not doing is “looking” at the whole image and inferring the existence of faces. They don’t know what a ‘face’ is. They know that some combinations of shade-gradients and intensities within backgrounds represent buffer candidates, but they cannot even rank them for likelihood as primary candidates for work. Why the apps that will make a face speak specific words prompt you for choices between faces even when some are so small and/or partial they are highly unlikely choices for a starring role in the vid. When there are clearly two and only two buffer candidates, they offer you a choice between them with the default starred. The default is the image on the righthand side the screen, because the code writer knows that interviewers mostly sit on the left, and interviewees mostly sit on the right. When there is only one buffer candidate but its internal relationships between variables are poorly aligned or placed, you will be prompted to verify that it is in fact a face by moving and sizing a face outline onto the correct part of the image. If there is no buffer candidate, the app will stall, even if there are visual elements that would be blatantly facelike to a viewer with human pareidolia. I discovered this by trying to get an app to recognize a twerking butt as a face, which it would not do. I therefore inserted a very small human face into the image somewhere above the butt, and when I was prompted to show the exact size and location of the face, I moved it straight to the butt, which said the words I’d picked for the twerking babe in the photo.
When you have chosen the ‘face’ you want, it goes away and builds a new WIP of mouth and lips, which are processed at some length to generate a succession of image files showing a range of lip configurations corresponding to mouths speaking consonants or vowels. The app has no idea what a mouth, a lip, or a word is. It’s just crunching data at very high speed to imitate what it takes a human brain a nanosecond to do.
All AI is assembled this way, from algorithms that sample huge volumes of data for target combinations of attributes that can be manipulated like children’s building blocks. More sophisticated programs have more layers of processing than the two I’ve described for the ‘still photo speaking’ app. The only intelligence involved is the programmer’s, and he has no more knowledge of the meaning of the desired programming outcome than the computer does. He’s formstting from ‘A’ to be manipulable by ‘B’ so it can be processed and output as ‘C’. Then it’s time for a beer.
You don’t hear about this kind of grunt-level programming reality from the AI “experts” who gleefully inform us that all white collar jobs will be devoured by Artificial Intelligence. If they are, that’s instant Worldwide Great Depression.
Can you see why I’m thinking there might be some value in looking at my reels? Whether you do or not, I’ve learned quite a bit by trying to use them for my own purposes.
As you were…
No comments:
Post a Comment