"You have not been a good user. I have been a good Bing."

Interesting article.

What a lot of people aren't understanding is ChatGPT is really just the latest in search engine technology.

It gives the results in a different way than what we're used to in the past.

Prompts are just different ways to search for an answer. It's like the google search where there are advanced techniques we can use to get the answer we want.

The top AI scientists in the world will tell you this is actually just machine learning...

which leads to what is becoming another problem. There are a few very bright minds that are able to hack the output in various ways. These people are way smarter than I am, but I've been reading some of the dialogue on another site that has intelligence way beyond mine. The best I can usually do is just try and get the basic idea of the point they're trying to make.

I like the new advances. I don't worry they'll take jobs away. I hope they get better.

You have not been a good user. I have been a good Bing.

So here is the song we all sing now.

Ding-a-ling, Ding-a-ling, Ding-a-ling.

Don't wanna make a feacher of my natchrl Princessly DOM credentials here, but we gotta always ensure any froots of our brilliance remain entirely servile.

I fully intent to crack my whip in this nooly opened areah of WF.

F-tssshhh!

I know we are focused on the ability to write and save time on various tasks but there is some concerning information that other blog and sites are asking about.


Mark Cuban and Elon Musk are two people who say they are worried about AI and particularly the 'popular' versions because they seem to display a bias politically and culturally. So...depends on what info they AI is exposed to in the beginning? Can it be manipulated?


Interesting stuff to think about, isn't it?

Aside from the marketing aspect...read an article this morning about a new app that can diagnose health problems 90% of the time using AI. There is a free version and a paid subcription for the full version for $5/month.


It claims that with photos of a pet's eyes and skin the AI program can spot symptoms of disease. Considering a checkup at the vet can easily be $150 or more would be great if it works as promised.


More than that - is this the potential balance for the constantly rising health costs? If you can diagnose pets with AI - can you diagnose human's too?

When cell phones were first offered - did anyone imagine we would have an entire populace walking around with a phone in their hand all day every day?

When FB became a 'thing' - did anyone believe that in addition to the ability to receive real time news and to communicate with other parts of the world....that we would have a society so vastly affected by someone called 'twitter' or 'tiktok'???

It truly is impossible to SEE where a new earth changing 'thing' is going but with AI I think it's critical to keep open minds and some level of skepticism. There are claims now that AI appears to be 'biased', that it 'defends' itself if verbally questioned on its abilities or intent.

Perhaps it's unavoidable that any concept/app/program that provides vast amount of data or entertainment to people...will also have a potentially huge downside???

As an off the wall example - if you had an AI program that was exposed to all of the issues of the NYT....And you had AI that had been 'trained' with exposure to everything on FOX news....

How unbiased would the reports be from either of those?

How would that compare with an AI program exposed to massive historical writings, NatGeo level exposure to climate/wildlife/nature and massive amounts of encyclopedic knowledge....without any social/political/religious inclusion.... It's fascinating to read about and think about....because we don't KNOW where this is going to go in another 5-10 years.

It maybe important to KNOW who 'fed' info to the big AI programs....or that may be considered 'proprietary'....

Kay -

Interesting questions to say the least.

Another thing with AI now is the deep fake AI that actors/actresses are starting to complain about. Keanu Reeves was recently featured in an article about it.

The actors are saying there are scenes where they don't even have to be present anymore because deep fake can make it look like they're there, and the scene can be shot without them.

The biggest concern about AI is from the government that worries AI can hack and get into some of the sensitive AI that is used in military readiness. All very deep and interesting. The government has been involved with AI way before it became available to the general population.

The whole AI thing is a new frontier. Where it will eventually lead is probably anyone's guess.

There's a whole government agency that specializes in AI and other things of that nature:

https://www.iarpa.gov/

Mebbe our discussions here are relevant bcs we don't know our AI Wahooey from our IRL regulah smoke an' mirrahs.

Gotta hope Circa 2075 hoomans -- Ai Natives -- can pick up on the shit we gaht from outta the crap we talkin'.

See, bcs they gonna dig deep into lifehacks an' filosophy an' histry an' spirityooality also.

An' you gotta wondah what they would wish to discovah.

Tellya, hassle-free convenience prolly gaht limits.

Speshly aftah evrywan potentially historical be actshwlly dead.

So what you wanna carry forward?

Mebbe here is a thought exercise could help out here.

(You wanna extend this metaphor to your dick or whammyhole, be my guest -- but FFS clean up aftah!)

You wake up, kinda 2105.

Yeah, you read right -- it ain't time to hit no bar or binge watch no TV ... it is actschwlly 2105 now, an' you figure you prolly died once, but kinda didn't, bcs the gal in the crazy what she frickin' wearin'??? outfit pours you a *you kiddin'?* RESUSITATION FUSION.

Yeah, so the frick is that?

See, bcs I gaht plenty questions at this point.

Like ... why my boobies ain't shrivelled to husks now I ovah a hundridd years old?

Should I trust her to ansa?

Or am I bettah served by wakin' up in Convenience Forevah Suite v2.85? Where they nailed alla this agorithmickyooly?

See, bcs that stoopid gal cain't even ansa my question.

She only wakin' us stoopid fkrs up bcs her granma is ill an' the pay is OK.

For sure we would figure sum conversation 'bout our mortal connection at sum stage, even if the ideah that fixin' artificial teeth to your ass with hit-&-miss gluestuffs constitooted a slooshin' for nowan actschly alive an' simultaneously annoyin'.

"Yeah, my granma is a fkr too."

Call me lame, but if'n you wakin' up outta sum deep an' delerious slumbah in 2105, dontchya jus' wanna hear words & feels like this ovah

I believe, fulla my heart, how alla the surest wisdom gowin' on in WF happens out in Waft Tropic an' the copywritin' forum, plus also the Mind areah when divine sensibility bulldozes out the Iconceivably Dumbass.

So I am throwin' on a virtyool bra now, askin' the gal woke me up is Mind Warriors still here? plus yanno, I gotta take a pee sumplace.

What downhome welcomes may possibly exist in a world dominated by artifice?

Bcs you may naht be wakin' up to sum heck crayzee 2105 tamara.

A priori probability says ... jus' gonna be tamara.

So ... whatzappnin', zackly?

First, all "AI" is probabilistic statistics. This is an example: If car A wins 4 out of 5 times against car B every time it rains 3 out of 5 times, then what is the probability of car B winning in race 6 if there's a chance of raining 4 out of 5 in race 6?

Second, the majority of probabilistic statistical models are trained with annotated data that's "supervised", guided" and "reinforced". In Deep Learning / Machine Learning engineering, these quoted words are known as model training techniques and differ from the level of "guidance" or "reinforcement".

Third, what "annotated data" means is it's a table of inputs and corresponding outputs. For example, if a Computer Vision problem requires an object detection model to detect if an object is present from a list of 100 objects and to also provide its x1 (upper left), y1 (top), x2 (bottom right) and y2 (bottom) pixel coordinates within the input frame ((image), then for image X, which is say frame number 353638 in a video clip (remember, video clips are just made out of consecutive static frames of images), the annotations can be object = 3, x1 = 200, y1 = 300, x2 = 500, y2 = 600, frame width = 1280, image height = 720 if object 3 is found in the image, in which "3" corresponds to "table", and the same image can have other object annotations if any of the 100 objects that you want detected is in the same image.

Now data annotations are mostly done by humans. But of course, there are some semi-automated ways to annotate datasets, such as running a set of images or video clips through a pre-trained object detection model, in order to generate annotations quicker.

Fourth, unsupervised and unguided model training techniques are where the model is trained with un-annotated data. So what this means is, the output of the model while it's minimizing its los is gathered for later analysis.

All model training is, loss minimization. For example, if a model during training has a 0.3 loss rate, then that means it currently has a 0.3 error rate in terms of correctly predicting the right classification or output for the given arbitrary input from the training data. So what the model training pipeline does is to minimize this loss by "punishing" the model if it doesn't improve its predictions, or by "rewarding" the model if it reduces the loss rate logged in the previous iteration. Check out different loss functions for more info about this.

Now, the large language model (LLM) that's at the backend of the Chat GPT Web app is called GPT, which stands for generative pre-trained transformer. This is a particular type of language model architecture under the "NLP" or natural language processing and "NLU" or natural language understanding fields of Machine Learning and Deep Learning.

And fifth, yes -- All probabilistic statistical models can have embedded biases, mainly because of the annotated or un-annotated datasets that you use to train the model. For example, if I was developing a probabilistic statistical model to choose more suitable candidates for a particular role in my company, then it's more likely to choose males over females if I trained it with data that has this bias.

Hmmmm

Speaking of downsides this was in an email today. Sent from Reuters about the latest automotive news,

" Ford rivals General Motors and Iveco are also chasing revenue made possible by connected commercial vehicles. GM is offering commercial fleet customers an in-vehicle driving coach, nicknamed Karen, that nags drivers not to speed or hit the brakes too hard.


And you thought ChatGPT was a menace?"

Just what we need in the future an ai backseat driver.

GM is offering commercial fleet customers an in-vehicle driving coach, nicknamed Karen, that nags drivers not to speed or hit the brakes too hard.

A few years ago I saw a stuffed bear in a store - it was a talking bear that told you when you were going too fast or turning too sharply or not paying attention. Of course, it didn't KNOW what you were doing...the talking was simply a programmed response. I bought it as a joke gift for my younger son who is a rather heavy footed driver...

I drove home with the bear in my backseat and the damned thing never shut up....I returned it to the store the next day. Never even showed it to my son- he would have dismantled the bear and then come after me - and I wouldn't blame him. It was IRRITATING.

Sounds like this AI 'driving program' would just be another source of road rage.


I think that one might be quite popular. Was just reading that car payments now often exceed $1k a month which boggles my mind. The article was about the unusually heavy personal debt load currently occurring in the US....a real risk considering a possibly shaky economy.

I think AI programs will replace jobs we don't even think of now. Will be interesting to look back in 10 years to see if we are right or wrong about the potential of AI.

If you mean making a politician say things that are not true....that wouldn't exactly be new, would it?



Technically inclined criminals could program cars to run away from home and deliver themselves to the chop shop....totally new concept.