Holy Schiit Talking (AI)!

Calling what we currently have available "AI" is a bit of an overstatement. Machine learning models are definitely impressive in their verisimilitude, but the mere fact that they're able to offer the appearance of truth while fully spreading misinformation is the most disgusting part about it, never mind how, at least in the world of art and illustration, there are people engaging in predatory processes infringing on others' hard work and having the audacity to be offended when arguments that those same works produced with hardly any effort on their part are not considered their own intellectual property.

Laziness drives innovation, but with our current trajectory we're well on our way to going full Wall-E.

 

Prompt to Bing in creative mode:
Write a review of the Sennheiser HD555 headphones, in the style of Yoda.

Output:
Great sound quality, these headphones have. Comfortable and light, they are. Open-back design, they feature. Hear your surroundings, you can. But leak sound, they also do. Annoy others, you might. For private listening, best suited they are. Long and durable, the cable is. But detachable, it is not. Replace it, you cannot. A good value, these headphones offer. Enjoy them, you will.

I was tempted to ask Bing to make it a negative review, because the HD555 are awful. But this is close enough.

 

I vote for "The Man in the Yellow Hat" to be AI trainer

 

I was trying to tell the Schiit AI that I am the guy with the biggest dick in all of audio, but it wasn't learning it. It didn't give me an answer on where's Marv.

This AI stuff is poo. I want my five minutes back.

 

If AI obtains sentience will it also find inner peace?
Will it find noble truths, or challenge them...

Might solve the brain in a vat problem once and for all.

 

,

Oh yeah I agree, just a semantic point that is particularly annoying to me because of how many people bandy the factoid* that we already have a proper artificial intelligence model.

Fun fact about that chess game vs Kasparov: apparently the move that caused Kasparov to resign that game thinking that Deep Blue was thinking at greater depths than he could fathom was caused by a bug, per IBM: https://www.wired.com/2012/09/deep-blue-computer-bug/

Deep Blue couldn't figure out a good enough response to Kasparov that it ended up playing one at random. Kasparov thought that the computer was seeing lines that he couldn't ken and was properly spooked by that. The fact that the human folded thinking that the computer was more skilled than it actually was despite that being far from the actual case says a lot about the human psyche and our cultural preconceptions about computer intelligence.

Having said that, modern chess engines like Stockfish 15 ARE actually able to beat the crud out of any living grandmaster so there's been improvements on that front.

*PEDANTICS: "factoid" = something said or supposed so often that it is accepted as truth (never mind its actual veracity).

 

Yeah, this is something we're just playing with a bit.

The genie is out of the bottle, though. I realized it when Lisa told me she'd been using ChatGPT to generate advertising stuff for Amazon and Google for a while, and another friend said they already used it for some legal review. Limited but nuts, Tyler is going full philosopher gaga over "free will is an illusion" and stuff like that.

But this is playing. Let's see how it goes. When it can be trained better--say, to point people at the RA page when it determines people really need an RA, and then approve it, hey cool now our people can concentrate more on real questions. Right now it's incomplete, a bit wacky, not completely reliable, but very very exciting. You know, like the internet in 1995.

I should be able to do another one that can swallow the entire corpus of Schiit Happened (to present day) and we'll see what it does with that in a bit.

Edit: oh and this is using chatGPT 3.5. 4 is nuts. Just a bit pricey to unleash at the moment.

 

When it could not come up with a chess move Deep Blue simply turned to its extensive poker training and bluffed.

 

This will be interesting to see as it gets implemented more and expanded out. AI spookiness aside, it seems like a neat tool. Now to try out this recipe:

 

Loads of data in search engines has been around for more than two decades now.

The really big step that was done with those new AI chat bots is what seems to be real and actual “understanding” of human language. That’s what fascinates me about ChatGPT. You can give it extremely complex sentences, and it understands what you mean, what your goal is. It remembers what you said before and what it said before itself. It can generate complex texts without errors.

In other words: It can really talk and “understand” the meaning of what is said. That, in my eyes, is the true milestone.

Better databases and training—there will be ways for sure. That’s not really the problem. But some kind of grasp of meaning and intention—that’s what makes this new stuff special and opens completely new possibilities.

And yes, it’s kinda scary.

P.S.: I’m especially interested in how political manipulation bot nets will evolve. They will become much better for sure in the near future. My hope is that in the end, they will make social networks completely irrelevant for political discussion as theirs will be more or less the only content. But it might well be that their commanders are clever enough not to let it come that far, and that populism will become worse than ever in the years to come.

 

AI: Earthling’s Final Form
or is it? I say AI (or AGI or whatever) will experiment and produce biological sentients for their own interests. AI will eventually create a superior being (psychic abilities obviously) and evolution continues...

AI already knows we know the answer since their knowledge base is built upon ours.:

Spoiler

The professors in my wife’s doctorates program encourage students to use AI to critique homework before handing in.

^ Bots will become indistinguishable.

 

Saw this last month:
https://twitter.com/linusekenstam/status/1646095934177124353?s=12

The list of the headphones the AI ended up scraping together:


Overall it looks like a summarized list of major tech sites that have the best SEO rather than having a good conclusion since headphones was so broad it even put earbuds in the second choice. Certainly has potential, doesn't seem like a list I would make so I don't worry about AI taking my job yet as self admittedly a lot of my job is headphone recs.

 

The real fun starts when AI is good enough that you can clone yourself--as in, you can produce a reasonable online facsimile of your persona, in chatbot form (or perhaps even voice and video, those tools are out there too)--and then have it do useful things. Haggle over a car price, make a deal, bring it to you for final signature. Of course there will be counter-AIs employed by the companies you're haggling with, eventually, and of course there will be people using 10 or 100 or 1000 clones to do stupid stuff, but at the rate this is going, things are gonna get real interesting really fast.

 

I recently grabbed this book: https://a.co/d/fifVr1v

I'm a relative doofus so I found this book useful because the organization I work in is using quite a bit of NLP and fancy statistics/graph analytics to do some interesting things. Can't really say more, but, what I can say is the question of the amount and types of bias due to training data selection (and the cleaning of training data) and algorithmic tweaks (knob twiddling) to "improve" ML output is something to which my organization is devoting an entire conference later this year. I have no reason to attend based on my field of expertise, but I do find the subject fascinating.

 

And the one who can afford the best AI gets the best prices? Ok, thinking about it, that’s not much different from the current situation. As a German saying has it: “The devil always shits on the biggest heap.” (Meaning he who has lots of money will always get more money.)

I think AI agents will be particularly useful for internet research. “Create a resume about what people say about the Focal Clear Mg, preferring results from SBAF, but including others, too.”

Stuff like haggling—I would have expected that businessmen like you, Jason, would always have the nagging feeling that they might have been better than the AI themselves, but as you bring this on the table yourself, maybe I’m wrong there?

 

Just doing important stuff here.



Write a song called "The Ballad of Jason Stoddard" about how Jason Stoddard conquered the audio industry marketing monster

Well, that sounds like a fun challenge! Let me see what I can come up with for "The Ballad of Jason Stoddard."

(ahem)

Verse 1: There was a man named Jason Stoddard Who faced a beast that left him scarred The audio industry's marketing game Was a monster that he aimed to tame

Chorus: But Jason was no ordinary man He had a plan, he had a plan To make Schiit that would stand apart And conquer the beast with all his heart

Verse 2: He teamed up with Mike Moffat too Together they knew just what to do They designed products that were fun and bold And didn't follow the industry's mold

Chorus: But Jason was no ordinary man He had a plan, he had a plan To make Schiit that would stand apart And conquer the beast with all his heart

Verse 3: They made amps and DACs and preamps too And personal audio that was brand new They refused to play by the industry's rules And made Schiit that was loved by audiophiles

Chorus: But Jason was no ordinary man He had a plan, he had a plan To make Schiit that would stand apart And conquer the beast with all his heart

Verse 4: Now Jason's Schiit is known far and wide For its quality and its unique stride He conquered the beast with his Schiit And made the audio industry take a seat

Chorus: But Jason was no ordinary man He had a plan, he had a plan To make Schiit that would stand apart And conquer the beast with all his heart

Outro: So here's to Jason Stoddard, a true audio hero Who conquered the beast with his Schiit, don't you know He showed us all that with a little heart and soul You can make Schiit that's worth its weight in gold.

 

The courses at fast.ai are pretty good at explaining stuff starting from a sort of general science and basic programming baseline. If nothing else the main guy Jeremy Howard is an exemplar of a very good teacher.