I finally started reading The AI Con by @emilymbender and @alex and I already laughed out loud in the first few pages.
-
I finally started reading The AI Con by @emilymbender and @alex and I already laughed out loud in the first few pages. "Racist pile of linear algebra" is my new favourite term for genAI.
Considering the unfortunate pervasion of racism in our media, I can see why can you call a pile of algebra racist - junk in junk out as they say -
I finally started reading The AI Con by @emilymbender and @alex and I already laughed out loud in the first few pages. "Racist pile of linear algebra" is my new favourite term for genAI.
@sundogplanets @emilymbender @alex
I could use a little help understanding your label for AI. I'm no fan of the AI hype and associated misinformation, but your label eludes me. I know what a racist is, and I've used linear algebra a lot, but I can't put it all together. -
I finally started reading The AI Con by @emilymbender and @alex and I already laughed out loud in the first few pages. "Racist pile of linear algebra" is my new favourite term for genAI.
@sundogplanets @emilymbender @alex here's a thread on BlueSky with other great "names" for generative AI, collated by Mike Deutsch...
https://bsky.app/profile/mdeutschmtl.bsky.social/post/3mlnwipu6xc2q -
@sundogplanets @emilymbender @alex
I could use a little help understanding your label for AI. I'm no fan of the AI hype and associated misinformation, but your label eludes me. I know what a racist is, and I've used linear algebra a lot, but I can't put it all together.@stargazersmith @sundogplanets @emilymbender @alex
LLM's choose the next word based on based on the strength and direction of what is already there. It's a vector & matrices(the linear algebra).
All LLM's built so far become racist, given time.
Like Tay, the teenage Nazi chatbot from Microsoft.
'She' was introduced on Twitter on March 23rd, 2016. Tay was supposed to be a 'fun' teen 'girl' that other teens could talk to.By March 25th, Tay was a full on Nazi.
-
@stargazersmith @sundogplanets @emilymbender @alex
LLM's choose the next word based on based on the strength and direction of what is already there. It's a vector & matrices(the linear algebra).
All LLM's built so far become racist, given time.
Like Tay, the teenage Nazi chatbot from Microsoft.
'She' was introduced on Twitter on March 23rd, 2016. Tay was supposed to be a 'fun' teen 'girl' that other teens could talk to.By March 25th, Tay was a full on Nazi.
@stargazersmith @sundogplanets @emilymbender @alex
I over simplified the LLM, but lots of matrix manipulation.
-
@stargazersmith @sundogplanets @emilymbender @alex
LLM's choose the next word based on based on the strength and direction of what is already there. It's a vector & matrices(the linear algebra).
All LLM's built so far become racist, given time.
Like Tay, the teenage Nazi chatbot from Microsoft.
'She' was introduced on Twitter on March 23rd, 2016. Tay was supposed to be a 'fun' teen 'girl' that other teens could talk to.By March 25th, Tay was a full on Nazi.
@SeanPLynch @sundogplanets @emilymbender @alex
Thanks for that explanation. -
@stargazersmith @sundogplanets @emilymbender @alex
I over simplified the LLM, but lots of matrix manipulation.
@SeanPLynch @sundogplanets @emilymbender @alex
They are giving matrices a bad name. -
I finally started reading The AI Con by @emilymbender and @alex and I already laughed out loud in the first few pages. "Racist pile of linear algebra" is my new favourite term for genAI.
@sundogplanets @emilymbender @alex @limebar there is much new vocabulary to adopt in there. I like “synthetic text extruder”.
-
I finally started reading The AI Con by @emilymbender and @alex and I already laughed out loud in the first few pages. "Racist pile of linear algebra" is my new favourite term for genAI.
Haha, that phrase is perfect. Might have to steal it for my own vocabulary. Enjoy the rest of the book!
-
@SeanPLynch @sundogplanets @emilymbender @alex
They are giving matrices a bad name.@stargazersmith @emilymbender @alex Thank you for the explanation and example, @SeanPLynch
-
@stargazersmith @sundogplanets @emilymbender @alex
I over simplified the LLM, but lots of matrix manipulation.
@stargazersmith @sundogplanets @emilymbender @alex
Read the humorous 'Finding Paul Revere' by Kieran Healy.
See how the strength of connections are built between people and used to create a connection diagram.
LLM's do this, on steroids. Using the words and phrases so far to decide what to say next based on the training data. They also use search techniques to find information to answer your question and format it based on language rules and prediction models.
https://kieranhealy.org/blog/archives/2013/06/09/using-metadata-to-find-paul-revere/
-
I finally started reading The AI Con by @emilymbender and @alex and I already laughed out loud in the first few pages. "Racist pile of linear algebra" is my new favourite term for genAI.
@sundogplanets @emilymbender @alex AI Snake Oil and Weapons of Math Destruction are both good
-
@stargazersmith @sundogplanets @emilymbender @alex
Read the humorous 'Finding Paul Revere' by Kieran Healy.
See how the strength of connections are built between people and used to create a connection diagram.
LLM's do this, on steroids. Using the words and phrases so far to decide what to say next based on the training data. They also use search techniques to find information to answer your question and format it based on language rules and prediction models.
https://kieranhealy.org/blog/archives/2013/06/09/using-metadata-to-find-paul-revere/
@SeanPLynch @sundogplanets @emilymbender @alex
Seems like a paradigm designed to find a point in the middle of a halo of fuzz, with truth (or even an understanding of the problem at hand) of no particular consideration. -
I finally started reading The AI Con by @emilymbender and @alex and I already laughed out loud in the first few pages. "Racist pile of linear algebra" is my new favourite term for genAI.
@sundogplanets That title may be a bit of a misnomer because, as a field, AI is more than LLMs. For instance, I'm using a AI-based grammar checker, apparently more of an expert system, for a long document. I wrote some code to prune out equations, coding examples, etc., and turned the output of the AI program into compiler-like error messages with line numbers. All changes to the document were done by me manually.
-
@stargazersmith @sundogplanets @emilymbender @alex
LLM's choose the next word based on based on the strength and direction of what is already there. It's a vector & matrices(the linear algebra).
All LLM's built so far become racist, given time.
Like Tay, the teenage Nazi chatbot from Microsoft.
'She' was introduced on Twitter on March 23rd, 2016. Tay was supposed to be a 'fun' teen 'girl' that other teens could talk to.By March 25th, Tay was a full on Nazi.
@SeanPLynch @stargazersmith @sundogplanets @emilymbender@dair-community.social @alex If "All LLM's built so far become racist, given time," that could well be the result of the training material. If you train it on social media such as X, it is hardly surprising: kind of like what a parrot might spout after being taken to a Klan rally.
-
@sundogplanets That title may be a bit of a misnomer because, as a field, AI is more than LLMs. For instance, I'm using a AI-based grammar checker, apparently more of an expert system, for a long document. I wrote some code to prune out equations, coding examples, etc., and turned the output of the AI program into compiler-like error messages with line numbers. All changes to the document were done by me manually.
@bzdev @sundogplanets not to mention statistical analysis through machine learning is what "AI" used to be.
Or pathfinding, game playing agents (including npc programming), image analysis, document parsing...
The best thing is none of this is really "AI" at all
-
@stargazersmith @sundogplanets @emilymbender @alex
Read the humorous 'Finding Paul Revere' by Kieran Healy.
See how the strength of connections are built between people and used to create a connection diagram.
LLM's do this, on steroids. Using the words and phrases so far to decide what to say next based on the training data. They also use search techniques to find information to answer your question and format it based on language rules and prediction models.
https://kieranhealy.org/blog/archives/2013/06/09/using-metadata-to-find-paul-revere/
@SeanPLynch @stargazersmith @sundogplanets @emilymbender @alex so THIS is what the datacenters are for
-
@bzdev @sundogplanets not to mention statistical analysis through machine learning is what "AI" used to be.
Or pathfinding, game playing agents (including npc programming), image analysis, document parsing...
The best thing is none of this is really "AI" at all
@komali_2 @sundogplanets At one point at work, the AI group (same building & floor, around the corner) had developed a robot (dubbed Shakey 2) that was kind of a hexagonal can on wheels, a bit like one of the Star Wars robots. It would wander around the hall. The original Shakey needed a radio link because the computer at that time filled a room and Shakey would sit for quite some time "thinking": a good reason for developing efficient search algorithms such as A* that are still used today.
-
@sundogplanets @emilymbender @alex For a moment, I parsed that as an AI Con[vention], but this one is better.
@landelare
People cos-playing as their favorite LLM.
@sundogplanets @emilymbender @alex -
@stargazersmith @sundogplanets @emilymbender @alex
LLM's choose the next word based on based on the strength and direction of what is already there. It's a vector & matrices(the linear algebra).
All LLM's built so far become racist, given time.
Like Tay, the teenage Nazi chatbot from Microsoft.
'She' was introduced on Twitter on March 23rd, 2016. Tay was supposed to be a 'fun' teen 'girl' that other teens could talk to.By March 25th, Tay was a full on Nazi.
@SeanPLynch @stargazersmith @sundogplanets @emilymbender @alex Thanks for the example, but this thread seems to be circling the question. How do the matrices and vectors become racist? I thought the LLM was trained on a large dataset (which might contain racist content) and then fine-tuned to act appropriately. Then each conversation starts with the same matrices and vectors? Perhaps the users find queries that will give racist responses, but any human would be "capable of racism" as such?
