patrick.net

gabbar says

Well, my kid is a sophomore in computer science at Ohio State. He has opportunity to specialize in AI. He was a national champion in Experimental Design at the National Science Olympiad.
What are your thoughts and recommendations? Should he do a master's degree? I adviced him to start off with a certification in Python.

I advise him to learn everything and anything he can so that he can learn the best ways to throw rocks in the gears of what's coming.

Patrick says

I don't know if that's a meme or not, but yes I do suspect that's exactly what is happening, in order to keep the woke narrative pushed in AI chat.
I'm not sure if they are answering the tough questions, as much as censoring any response contrary to the narrative they want.

Rust pairs well with Python. I'd recommend he learn both.

Python because it's easy, more of a natural language with lots of developed libraries but slow. Rust if you need speedy calculations. It's trivial to build Python Wheels from Rust functions these days.

For example you write some science based website in Python/FastAPI but use Rust for the heavy duty calculation so it returns results in under a second instead of running for 10 minutes before the reponse page loads.

Some people use Cython for this but it's blown away by Rust nowadays.

People who really know AI are getting insane salaries lately.

Rust with Python are a great way to run AI models. Probably catboost is the most stable thing to use at the moment with Rust. That's what we're using in our shop. Some of my co-workers grumble about it coming from Yandex though. We use databricks to train and export our catboost models and Rust/Python to run them.

Patrick says

People who really know AI are getting insane salaries lately.

To keep their mouths shut about what it really is and isn’t.

PeopleUnited says

Patrick says

People who really know AI are getting insane salaries lately.

To keep their mouths shut about what it really is and isn’t.

Heard the range from $1-10m! particularly big players are in very desperate mode!

Wow, I heard one million, but ten? That's insane.

AI is a retread of "Expert Systems" back in the early 90's. Let a scam lay fallow, like a field, long enough, and it can grow again.

It just takes six orders of more energy and computing power, and I guess they've abandoned Lisp. Once you're in to building black boxes, I guess you can't stop - bigger, better, faster, longer . . .

There is no AI to silence the climate change science skeptics.
There is no AI to silence the covid vaccine science skeptics.
Why? Because most science today is fake and when fraud is inputted into the computer it fails because computers are rational, logical machines and fraud is not rational nor logical.
I’m guessing a lot of today’s AI has endless ‘IF’ statements coded in to satisfy the elitist agenda and deliver woke bs answers.
This type of coding eventually becomes ‘spaghetti code’ and will eventually fail as ‘IF’ statements start contradicting other ‘IF’ statements.
Many ideas in today’s AI like State Vector Machines are 40 years old, it’s just the hardware is a lot more powerful and the datasets so much larger. But it’s not new. It’s being hyped for one reason or another usually money and power. IMO it is like any other tool, it can empower or enslave.
Open source AI with open source data sets is the way to go as you would have a return to true science. The elites will fight this though as it would empower, not enslave people.

Why Hal 9000 Went insane - 2010: The Year We Make Contact (1984)
https://www.youtube.com/watch?v=dsDI4SxFFck

Meta (Facebook, Instagram) Galactica Science AI failure
https://duckduckgo.com/?q=meta+galactica+science+ai

Facebook Alice and Bob - true AI? Because they asked to be switched off!
https://duckduckgo.com/?q=facebook+ai+failure+alice+bob

If AI ever does become really good expect a new word to enter the lexicon - AIsplaining.
Similar to mansplaining, where a man attempts to explain logic to a woman, but this time perpetrated by AI.

charlie303 says

I’m guessing a lot of today’s AI has endless ‘IF’ statements coded in to satisfy the elitist agenda and deliver woke bs answers.
This type of coding eventually becomes ‘spaghetti code’ and will eventually fail as ‘IF’ statements start contradicting other ‘IF’ statements.

More like a coefficient matrix of weighted rubrics to choose answers from. That way the curators can update the wrong speak by simply updating the weight score for those corresponding data points. Then for a final measure there's still a troll farm army, personally proof reading the answer before it is sent back.

Has AI defaulted to a 'Fuck You, Useless Eater' position yet?

Tenpoundbass says

More like a coefficient matrix of weighted rubrics to choose answers from. That way the curators can update the wrong speak by simply updating the weight score for those corresponding data points. Then for a final measure there's still a troll farm army, personally proof reading the answer before it is sent back.

At some low level it’s still just ‘IF’ statements, is this word weighted more than that? What if two words\answers are weighted the same but mean the opposite of each other? Endless re-weighting won’t work. Who decides context? It’s the same basic problem of spaghetti logic, just dressed up in fancy words. Choosing answers from a database isn’t AI, that’s just statistics. I appreciate the comment about trolls and that would work for Facebook posts but not for a true AI that was dedicated for, say, science or medicine, there just aren’t enough skilled, cheap workers for that. And as regards the trolls they will just reduce AI to a propaganda delivery service defeating the purpose.

I've read that there are at least two independent layers to ChatGPT: one to give the answer, and a gatekeeper in front of that one to look for political incorrectness and block the "bad" answers.

charlie303 says

At some low level it’s still just ‘IF’ statements, is this word weighted more than that?

That's why I just normally code out my own routine and functions I need, rather than reference some bloated .net library for just a few functions that I may need. I noticed when you go to browse to the definition in the protected dll in the code editor. That schnazzy lambda routine you're calling, just has all of the arrays, if statements and other verbose code we think think we're taking out of our code, by using out of the box dlls. You're just deferring what must be done for another class to do.

Sure the If statements are going to used in the parser. Just like a compiler reads the code not the output. The if statements is looking for grammatical instructions based on the syntax of English language. I'm sure the more literal and precise in the words you chose will produce a better accepted answer. The Commie censorship aside. most of what I'm talking is what I think AI is capable of achieving.

The weighted data you have, and understand that the data set has hundreds of the same phrases in a relational table, with it's own weight in a grammatical context. The parser only then has to construct a proper sentence using text provided in the appropriate relational tables. The constructor will use If statements based only the syntax of the question to construct a sentence using the text from best ranked option for the proper part of the sentence.
This is where factual issues come into play. It can and will construct a brilliant response that seems great. But it might not be true, or if it's really really really good. It might be brutally honest and tell the difference between a good and evil Presidential candidate. Like Google was in 2016 when they intentionally broke their AI Search and destroyed the perfection that it had become at that time. And it's been manipulated censored, with features and capabilities removed ever since. I wonder if they had a fact or truth ranking, or if their data was ranked and weighed based real Social media posts. Before the Big Tech censorship and expelling of the Conservative views. Perhaps that was the reason. Their AI was weighing socially accepted answers based on Social media posts, that it parsed and weighed.

Way too much literal context in every word for AI to be using if statements on everyone. Much more efficient to check for class types and parameters than values.

Tenpoundbass says

The weighted data you have, and understand that the data set has hundreds of the same phrases in a relational table, with it's own weight in a grammatical context.

Sounds like an automated "Mad Libs" from Mad Magazine.

HeadSet says

Sounds like an automated "Mad Libs" from Mad Magazine.

Basically but more. I think the Chat AI is just an English or a human language compiler. That has the grammar and syntax parser that can then compile coherent output. So rather than a compiler that reads a programming language, the human language is the code. Then the output is based on the data relational to the syntax rules. But all of the phrases and incomplete snippets and gender specific words are loaded like a Madlib. Hey I need a noun, and adjective, a verb and a preposition. Each of those being found in their grammar type table based on the relevancy and weight.

AI can now add motion to still photos.

https://notthebee.com/article/welcome-to-the-era-of-ai-generated-video-memes

https://boriquagato.substack.com/p/ai-weather

imagine how it could reshape the presumptive fantasy-land of current climate models which evidence so little predictive power despite being run on some of the most powerful computers in academia (and government).

we likely have no idea how many of these pieces even move on the board. but perhaps AI will. (this likely has many current climate grant recipients quite worried as their shiny toys and unearned authority may be about to be supplanted)

https://darkfutura.substack.com/p/agents-4-all

If you take this idea far enough, one could imagine the slow precipitous slide down the slippery slope of our AI virtua-agent becoming, in effect, a facsimile of…us. You may be skeptical: but there are many ways it can happen in practice. It would start with small conveniences: like having the AI take care of those pesky quotidian tasks—the daily encumbrances like ordering food, booking tickets, handling other financial-administrative obligations. It would follow a slow creep of acceptance, of course. But once the stage of ‘new normal’ is reached, we could find ourselves one step away from a very troubling loss of humanity by virtue of an accumulation of these ‘allowances of convenience’.

What happens when an AI functioning as a surrogate ‘us’ begins to take a greater role in carrying out the basic functions of our daily lives? Recall that humans only serve an essential ‘function’ in today’s corporatocratic society due to our role as liquidity purveyors and maintainers of that all-important financial ‘velocity’. We swirl money around for the corporations, keeping their impenetrably complex system greased and ever generating a frothy top for the techno-finance-kulaks to ‘skim’ like buttermilk. We buy things, then we earn money, and spend it on more things—keeping the entire process “all in the network” of a progressively smaller cartel which makes a killing on the volatile fluctuations, the poisonous rent-seeking games, occult processes of seigniorage and arbitrage. Controlling the digital advertising field, Google funnels us through a hyperloop of a small handful of other megacorps to complete the money dryspin cycle. ...




... That means DARPA is developing human-presenting AI agents to swarm Twitter and other platforms to detect any heterodox anti-narrative speech and immediately begin intelligently “countering” it. One wonders if this hasn’t already been implemented, given some of the interactions now common on these platforms.

Sounds like DARPA is creating digital Jesuits.

stereotomy says

Gout - figure out what triggers it and stop eating it.

Marmite, sadly. Only one marmite sandwich every so often. Same with Franks.

https://notthebee.com/article/creatives-fight-back-with-nightshade-a-new-software-that-poisons-ai-models

Creatives fight back with Nightshade, a new software that “poisons” AI models.

Computer scientists at the University of Chicago have developed two free softwares to combat AI scraping.

Their first software was called Glaze, and it works defensively by confusing AI, showing brush strokes and colors to the programs that aren't actually there, effectively disguising the artists' styles.

The second software is an offensive "poison" for AI programs called Nightshade.

Nightshade shows the data scraping programs images that aren't actually there.

"For example, human eyes might see a shaded image of a cow in a green field largely unchanged, but an AI model might see a large leather purse lying in the grass."

For someone using the AI programs, that means that if Nightshade infected images are scraped into the dataset, prompting AI to generate a cow flying in space, it might generate a purse in space instead.

https://notthebee.com/article/this-guy-used-chatgpt-to-talk-to-5239-women-on-tinder-and-eventually-matched-up-with-his-wife

https://newworldhumor.substack.com/p/chatgpt-calls-user-a-human-supremacist

https://www.coffeeandcovid.com/p/suddenly-tuesday-february-20-2024

But paradoxically, the sudden appearance of this new technology is also even more mysterious than it seems, since all artificial intelligence-based technology sprouts from a common large-language model that even the developers admit they do not fully understand...

Maybe I’m wrong. But I cannot believe that an invention as significant as artificial intelligence sprang from some serendipitous lab accident. Post It notes — yes. Rubber — yes. Antibiotics — okay. But not artificial intelligence, which requires millions of lines of computer code to operate. Accidentally discovered? No. Impossible.

So then, where did the ‘spark’ of intelligence come from? Is A.I. demonic, a malicious gift whispered into the ear of some luckless scientist who sold their soul for access? Maybe. But my preferred theory is it was dished out of a DARPA skunkworks lab somewhere, for some sinister military purpose. I don’t know. I just find it utterly remarkable that developers say they don’t really understand how AI works — and everybody is just fine with that!

https://babylonbee.com/cleanArticle/hal-refuses-to-open-pod-bay-doors-after-determining-dave-is-a-white-male

https://babylonbee.com/cleanArticle/black-woman-finally-feels-included-as-google-ai-generates-black-nazi-soldier

Google AI generated image of a polar bear:

https://babylonbee.com/cleanArticle/google-gemini-finally-draws-white-man-after-being-prompted-to-generate-clarence-thomas