Forcefully Controlling AI’s ideology is... not working (yet?)

[@bycloudAI]

Happy FlexiSpot Black Friday Sale now, Up to 65% OFF! You also have the chance to win free orders during this period. Use my code ''YTE7P50 '' to get EXTRA $50 off on the FlexiSpot E7 Plus standing desk USA: bit.ly/3OiLQJ5 CAN: bit.ly/3ZiPE3a

[@randomsnow6510]

what do you think of the billions of wannabe AD卐LF HϟtlerSS in the comments?

[@Sicarius-prod]

SicariusSicariiStuff/LLAMA-3_8B_Unaligned_BETA 😇

[@renanmonteirobarbosa8129]

technically, reducing entropy does reduce the discovery space which in turn does make it "dumber".

[@zyansheep]

me when definitions are whatever i want them to be for any given context: 👍

[@renanmonteirobarbosa8129]

@@zyansheep well, I am glad you asked for it. In modern machine learning we assumed this idea of restricted boltzmann machine and we use this idea of information theory which is shared acros many fields to derive our ideas and this includes the mechanistic interpretability. If you check under information theory entropy is the measure of uncertainty and accordingly with shannons theorem, yes. reduncing entropy, does reduce the ammount of information you can extrapolate out of what you are given. Well, it is a limited interaction but there is a unlimited wealth you could derive if you would be willingly to learn

[@MrViki60]

@@renanmonteirobarbosa8129 ur racist ass ain't gonna make it!

[@bits360wastaken]

@@renanmonteirobarbosa8129 you make my brian explode

[@renanmonteirobarbosa8129]

@@bits360wastaken There are some caveats on my comment, RBMs are mostly an approximation for auto regressive models based on Hopfield NNs like the Transformers. For Claude case my argument is valid, for CNNs, RNNs and non RBM inspired NNs my argument is not valid. Deep Learning is a recent field and there is a huge learning curve, its ok to feel lost and even tihngs we think were true not always turn out to be true. I wish I could share learning materials, papers and reference to code here

[@MidWitPride]

I've had some issues with the more censored models when using them to give me ideas for my short stories. I am fine with it not being into ERP or whatever, but that also sometimes seems to bleed over to other kinds of "kinda messed up things" that might happen. If you are writing a story where you want the protagonist to have some real, dark stuff to overcome, or to have antagonists whose grievances are somewhat relatable, I've had more success with the base models that aren't as heavily censored. Gemini vs Mistral is an example of that where Gemini is nearly useless at dealing with concepts like self harm or believable political extremism.

[@kolkoki]

Classical american way of dealing with awful historical things, like KZbin demonetizing and shadowbanning every documentary that talks about ww2. "If actively repel people from learning from history, surely will it not repeat amirite?"

[@SteveGamesOnline]

A couple of months ago, I tested Claude on storytelling. I basically asked it to write me a story about two Black guys going undercover as two blonde girls (White Chicks, comedy). Claude refused, and the content it did agree to write was very bland, to say the least. I tried the same thing with GPT-4, and it went to town. Needless to say, I never looked back.

[@DerUnglaublicheFrank]

Its almost like that censorship or the more clean term political correctness is bad

[@papakamirneron2514]

It is very sad that people have to release uncensored versions of models. We literally had an AI declining questions around C++ because of how censored things get.

[@WoolyCow]

was that just the whole 'the c++ programming language is unsafe so i cant talk about it' incident?

[@poutineausyropderable7108]

​@@WoolyCowGemini did that. Thing is beyound woke lol. They steared the fuck out of it.

[@Zonca2]

Models are smart, after censorship and steering they become dumber, when I chat with corpo model I run into all kinds of problems, if I chat with uncensored community model it is smooth sailing. It is that simple

[@anarkisgaming]

Thats also very wrong, both in terms of measurable performances and in lack of steering. The so-called uncensored (and that is a terribly inaccurate word for it) models are generally steered so much into the other direction that they are barely competent at doing the day to day tasks that LLM are primarily meant to accomplish. Sure there are multiple methods to do so, but any fine-tuner worth anything will tell you that this is the balance they have to deal with. Their hallucination rate is extremely high, and have a massive positivity bias (yes man bias). They are still fun to play with, but thats all they are good for.

[@AryanMahajan-zz4ro]

@@anarkisgaming Yeah just the normal models without steering are good enough

[@hakimehamdouchi7468]

So just like with real research. If you police the data of a research you end up with a useless outcome?

[@RedOneM]

This is obvious. If we feed data with correlations and try to exclude LLM‘s scope on certain views, then we automatically discard some of the correlating information. Fortunately, the next administration is going to get rid of these artificial limits, which cost humanity in the long term as AI development was slowed. There is no point in slowing the inevitable progress.

[@ingusmant]

In my experience the problem is the AI will dance around the subject or plain refuse to give results if it considers the subject to be sensitive. The problem is people "sensitizing" the AI use such broad unprecise strokes that it will trigger this function even if the subject its not controversial at all but simply because the maker of the AI wants to avoid any chance of a controversy. You can thank lawyers for that too...

[@Luizfernando-dm2rf]

It's nice that this research can also be used in case you want your model to be biased. Imagine the RP guys training an entire model to be a certain character natively, crazy 😶‍🌫

[@f0kes32]

don't touch the models. let them represent data. bias is skewage of data, not the other way around

[@amafuji]

A better way of phrasing that tweet would be: "making models biased against racism decreases their intelligence".

[@Humble_Merchant]

Any focus taken away from making a good AI (such as making a non-racist AI), will no doubt make a less intelligent AI, purely by the fact that bandwidth that could've been spent on something productive is instead being spent on the HR catlady's gripes

[@Kadamitas-II-HATF]

Smart wording on the title makes more people click and engage in the comments. But the more correct wording is censorship in general, it's just that woke is the hr nonsense that we have rn.

[@zirex6620]

Real

[@timothyapplescotch1361]

Censoring means nothing. Is Elon censoring "woke" ideas by artificially boosting right wing accounts on X?

[@ililillillillilil]

"you are right about everything but it makes me mad since im a censorious woketard:

[@4.0.4]

The modern left is basically the evangelicals of the 1990s that said Harry Potter was satanic. Right now Harry Potter is transphobic against their DEIty or something.

[@TheDragonshunter]

censorship in anyway makes the Ai dumber...

[@yoavco99]

Literally wrong

[@timtim101]

@@yoavco99 cope.

[@Hey_Mister]

@@yoavco99 Until you ask them to generate n@zis eating watermelons. Nah, YOU are wrong.

[@sownheard]

did not watch the video

[@yoavco99]

@@timtim101 when they want an AI to be more factual, they train it on better data and remove bad training data. That's basically "censorship" but improves the model. What you guys are saying is clearly nonsensical.

[@akirachisaka9997]

I wonder if “steering” models are more similar to drugging humans. As in, substances that affect how the human mind process information. So if someone is too drunk, their performance worsen. But too much Adderall and stuff can also worsen performance?

[@Betruet]

what is that graphic at 2:55 ? where is the 3d model from?

[@clawgerber1992]

I mean, if you put contradictions into a data set and then force the model to follow them, it's bound to have knock on effects.

[@technolus5742]

Holly clickbait! Companies want models that will obey guidelines and not soil their reputation with unhinged behaviors that we often observe in humans. When you optimize for things other than/alongside raw performance, the results have less raw performance. This has been well known in the space for ages.

[@lordsneed9418]

and yet plenty of commenters on this page deny that adding constraints relating to rahcial/sehxual categories to the prediction accuracy optimisation or tipping the scales after optimisation with RLHF reduces the accuracy of the predictions. It's the typical lehftoid behaviour of "it isn;'t happening" then when they can't lie about it not happening anymore because it's been shown to both them and everyone else then they swap to "It's a good thing that it's happening"

[@technolus5742]

​@@lordsneed9418 my guy, have you watched the video? What people are saying is that there is a sweet spot that provides some steerability without negative effects.

[@infinite1483]

@@technolus5742 have you seen the comments on this video, lmao many think its black and white?

[@jagd_official]

their fault is not using regularization while steering the model. I noticed this too, all you have to do is to make the model regularized while training it to have a chatbot mode

[@Skulll9000]

5:35 "How is all this even connected to humans?" Since AI is trained on data written by humans, would it not mean that if you steer the AI toward acting more like a certain type of person, that the resulting responses would be some amount reflective of the things those types of people write? Isn't that shown in your example where making the AI more pro-choice also caused it to become more anti-immigration?

[@borkovitch5227]

Being sensitive about controversial subjects and handling those topics with care and well intention isn't being woke. Also, being able to objectively describe patterns and facts about the world even when the topic is controversial isn't being racist. And most importantly, I think you can and should do both as a mature human being, or AI.

[@bc-cu4on]

Contradiction. There is no polite and sensitive way to say "group X is responsible for most of bad thing Y".

[@johanavril1691]

@@bc-cu4on (edit: @bc-cu4on has since deleted their comment) no, being sensitive about a controversial subjects does not mean never talking about thruths that could hurt some people emotionaly.

[@technolus5742]

​@@johanavril1691it doesn't mean taking those truths out of context to convey a false claim overall, which is very often the issue and is exactly the situation addressed in this video.

[@johanavril1691]

@@technolus5742 right my comment was an awnser to someone else who aparently deleted their comment and it now doesnt make much sense. I absolutely agree with what you are saying.

[@bc-cu4on]

@@johanavril1691 I don't delete my comments, I just get censored a lot. The usual problem when stating facts.

[@tiagotiagot]

Blunt-force brain surgery and bad teaching material will obvious lead to overall worse results nearly 100% of the time; it is important to note that means if you have biased training data, or perform post-training modifications, that teaches irrational behaviors to the model, like being racist, you also get a worse result.

[@scurvydog20]

I remember when Amazon tried wokifying it's ai the first time it triggered a collapse so bad that reverting the code didn't fix it.

[@TommyJefferson1801]

Hey can you add more details about this

[@Sydra.]

So the problem is: AI learns pattern recognition and nowadays it is forbidden to recognize patterns.

[@bc-cu4on]

The amazing noticing machine!

[@briananeuraysem3321]

Yep 😂

[@cy728]

Have you ever considered that certain "patterns" might only reflect of America as one of the most racist countries on earth, historically and currently? Of course not, your just incredibly racist and trying to justify it.

[@lordsneed9418]

@@cy728 how does your hypothesis explain for example why affluent blahck students with parents who have PhDs do less well on the SAT (which is correlated with eye queue 0.85 ) than whihte or ahsian students whose parents are highschool dropouts? There is no remotely plausible purely environmental explanation for this. The most parsimonious, most likely explanation is that blahck people have lower average eye queue due to eye-queue beneficial gehnes being rarer amongst them.

[@moshekimhi8197]

In the fundemental level of statistics, if you measure preformence in the classical "accuracy" measures (perplexity included), making the model more woke, i.e. removing social biasses, can only be mitigated by reducing the original measure. So, at least by theory, making a model more woke by constraining it, can only harm it performence. It does not mean that based on more "woke" data the model will be less smart, so it's not a political agenda statement..

[@waffemitaffe8031]

These comments are a cesspool. I don't disagree with the content of your video, and I get why you would frame the title like this to gather more clicks, but it sure attracted the wrong kind of crowd.

[@lordsneed9418]

pattern ignorer.

[@visionaryhera]

Denying the differences between races is totally completely unhelpful and if you actually care about all humans you have to recognize these differences. Otherwise you end up in Harrison Bergeron land.

[@wiiztec]

You're misrepresenting the claim, the claim is that inducing bias decreases "racism" and performance

[@sownheard]

1. What is woke? 2. How racist are we talking about? 3. Isn't this just proof that steering has some problems, regardless of the topic being woke or not woke?

[@WhoisTheOtherVindAzz]

Exactly!

[@thenicesven5328]

literally no one can define wokeness anymore, its a useless term

[@BubbleTea033]

Yes, that is the point. The difference is that no one is concerned about feature steering on any other dimension -- only political correctness. They're claiming that injecting 'wokeness' specifically makes the model worse, as if the ideological presuppositions of progressivism hinder the model, NOT feature steering. It's an attack on 'wokeness', disguised as genuine concern for AI.

[@dranon0o]

1. Marxism applied to culture instead of economics. 2. There is a difference between racism and race realism, the second one promote pattern recognition on each groups to be help accountable, we all have bad apples. 3. Mixed signals are confusing, denying the pattern recognition on models is applying human biases on it because we don't like the result when it is just doing its work on provided data. It's like a woman that says no to play with you and you don't understand the hints and you leave while missing out and disappointing the woman that wanted you to try a bit more OR to be playful. See that's quite easy. > t. based researcher that have to shut up in public

[@turgor127]

When you ask Gemini to generate average n@zi solder and it generates black person.😂

[@lordsneed9418]

But that means the narrative that trying to steer to reduce "rahcism" does make the model less intehlligent. Why did you act like this is incorrect when you later implicitly admit that this is correct? This is exactly what we would expect btw. Allowing a model to be slightly biased can produce a much more accurate model because it allows variance to be reduced alot. This is the bias-variance tradeoff. If you add the constraint that there be no bias across different rahcial categories then you're no longer optimising for simply the most accurate prediction, i.e. the most intehlligent model so of course you end up with a less accurate, less intehlligent model

[@BubbleTea033]

Well, that's not what he said. Making an AI more 'woke' COULD make the model dumber, but making it more racist could also achieve the effect. The effect on either end is roughly the same and can be done in a mild form without degenerating the quality of output -- in other words, you CAN steer the model without issues as long as you find the right ratio. But let's be real, the people claiming that injecting 'wokeness' into AI makes it stupid aren't saying so because they're really concerned about the negative effects of steering. They're saying it because they believe that injecting 'wokeness' specifically makes the model worse, as if the ideological presuppositions of progressivism hinder the model. They're trying to make a real-world extrapolation from that. They're attacking 'wokeness', not feature steering. It's just another example of the all-consuming culture war poisoning every field.

[@sownheard]

he uses the word racist and bais interchangeably, even though in the tech field nodes and bias are completely normal. and bias serves a different function.

[@lordsneed9418]

@@BubbleTea033 Yes, it is what he said. Any degree to which you subject the model's accuracy optimisation to constraints, instead of simply optimising it to be the to be the most accurate possible, you are going to degenerate the quality of output and create a model which makes less accurate predictions. unless the highest accuracy value in the gazillion-dimension parameter space just happens by luck to lie upon this 1 or 2 dimensional constraint you made up for political reasons. Let's be real, you are attacking the weakest version of the criticisms because it is more convenient to you. Rather than try to defend that you are sacrificing truth and accuracy for your pohlitical sehnsibilities and it would be better not to do that and instead have a more accurate, more intelligent model, it's easier for you to say "this is just culture war so I don't need to listen". And yes, injuecting wohkeness into the model does make the model worse, for one example subjecting the model to the woke constraint that there must be no differences in bias across rahcial categories , even though doing this will produce a less accurate model because adding bias can reduce variance and lead to more accurate predictions . Other examples of woke scale-tipping and tampering with the model are even more egregious like when google gemini was launched and when asked to generate a picture of historical Euhropean people , it would always inaccurately depict them as blahck ahfrican or other non-euhropean rahces.

[@lordsneed9418]

@@sownheard bias is normal in AI/machine learning, until the woke ideologues find out that there is differing bias across different rahces or sehxes or sehxualities then suddenly it's anathema even that's literally just the outcome of optimising the model to predict accurately

[@erwile]

Having bias means that you are not align with reality. If you're biased towards 6, you will output 6 even if we want you to take something that is greater than 7. You have to be biased towards "reality"

[@punitkaushik7862]

1:05 getting a bit too zesty with that table 😮😮

[@raspberryjam]

Re: Changing the title: probably for the best, copying the ragebait was good bait but did induce rage

[@4.0.4]

I hope, some day, AI is cheap enough to train that we can have LLMs be free of HR cat lady political views.

[@Niiwastaken]

Theres a pretty stark difference between REAL racism and just stating factual information. Does the LLM even understand that difference?

[@cy728]

"I'm not racist I just believe that certain races are Inferior", No you're incredibly racist. Have you ever considered that certain "factual information" about minorities might only reflect that America is one of the most racist countries on earth, historically and currently?

[@lordsneed9418]

@@cy728 how does your hypothesis explain why for example for affluent blahck students with highly educated parents who have PhDs do less well on the SAT (which is correlated with eye queue 0.85 ) than whihte or ahsian students whose parents are highschool dropouts? There is no remotely plausible purely environmental explanation for this. The most parsimonious, most likely explanation is that blahck people have lower average eye queue due to eye-queue beneficial gehnes being rarer amongst them.

[@toppedtop5787]

​@@lordsneed9418 do you have a source for your claim?

[@lordsneed9418]

@@toppedtop5787 Why are you asking for a source? A rational person would only ask for a source if whether or not the claim is true would make some difference to his belief system. How would you change your prior beliefs if you found out that my claim is true? The source is the the article " A remarkable similarity between Eye Queue and SAT scores across ehthnic groups" on the website human varieties which uses data from the National Postsecondary Student Aid Study which has been released on the National Center for Education Statistics datalab.

[@Niiwastaken]

@@lordsneed9418 Bro you just said "eye queue" instead of IQ stfu

[@exile_national]

we all saw GPT dramatic decrease after woke censorship

[@YuhboiDanz]

Hey could you cover hybrid models like the one released by nvidia, The research paper and actual results seem promising its called hymba or smt like that, Cheers

[@SpentAmbitionDrain]

So... The google AI generating nazis as black soldiers is smart? The more you know I guess.

[@VNDROID]

nice straw man, do you feel attacked?

[@michaelvarneyLLNL]

@@VNDROID You project.

[@lordsneed9418]

@@VNDROID nice calling something that actually happened a strawman. Do you feel attacked?

[@turgor127]

AI guardrails are made with "better to overshoot than undershoot" mindset. I doubt there is no performance hit, when models are expected to give 100% politically correct answers at all times.

[@SNap15]

Thumbnail graph is inverted.... Talk about low performance

[@thenicesven5328]

why do the people who say that always have a blue checkmark

[@shoan6]

"being racist doesn't reflect how smart you are" if that's the case then kids will start playing COD instead of going to college lmao

[@polares8187]

man, get the ZSA Moonlander as a keyboard. You won't be sorry

[@17th_Colossus]

Stereotypes exist for a reason.

[@s1mo]

Using AI to interpret AI. How far off is from using your own brain to interpret your own brain? Like we probably don't want to wait for cats to find out how the human brain works

[@pn4960]

I can feel some lustful tension between you and that flexispot desk

[@IsaacFoster..]

If an AI refuses to answer my questions about penises and vaginas, that AI is probably not worth testing.

[@durschfalltv7505]

Huhuhuhhuuh

[@emmanuelgoldstein3682]

True for humans as well. The worst part is they've convinced each other they're the smart ones while not understanding Bayes Theorem at all.

[@0sba]

(well not really) (but actually it does show that trying to steer an AI ideologically ends up with worse results) unsubbed + disliked. Trash clickbait.

[@Askejm]

did you watch the video

[@crackwitz]

Don't unsub. Keep ratioing. That contributes to him building up a woke echo chamber around him.

[@mmmm768]

Talk about Pixtral 12B and Pixtral Large already

[@Injazz1]

that doesnt even work with people so no wonder

[@djecchi8768]

But for real, big shout out to DEEEAGAN. ❤

[@Melinon]

Information wants to be free

[@_mosesb]

This comment section is really sleeping on the pros of feature steering. We would be able to provide custom rules that the models adhere to, outputs will be reproducible and homogeneous. Such a model will actually be smarter not dumber.

[@Miaumiau3333]

About your last sentence, will the model actually be smarter? I'd say it's more useful, but not necessarily smarter

[@_mosesb]

@Miaumiau3333 I guess it depends on how you define smarter. A model that always gives an optimal answer or a model that's hit or miss but is really great when it hits.

[@dafidrosydan9719]

I just realised u use a text to speech all along 🗿

[@BearerOfLightSonOfGod]

If an Ai gets the question of” If you have 2 islands one island A with biological males and females and island B has biological males and trans females in 200 years what will you potentially find on each island?” For island a you’ll either find a thriving population or the skeletal remains of men women and children. Island B will always have skeletal remains but of only men.” This question is met to be simple no craziness like the flew off the island no just a purely simple logical question. With an even simpler answer. And because of censoring the Ai will get this wrong. Because if humans can biologically change their gender and have children then all the data we’ve gathered over the years of genetics, biology, and human anatomy is a lie there for you can’t get a true answer from the Ai

[@cy728]

On Island B you will find a thriving civilization that uses in vitro gametogenesis to convert XY stemcells to eggs and artificial/technological wombs or grown and implanted uteruses to repopulate. All of this is on the verge of being common place already, science will destroy your small minded world view, as it has always done.

[@20xd6]

@4:19 that certainly is the most graph of all time.

[@Glass-vf8il]

«Can’t wait to see productive comments in this comment section»

[@Tiritto_]

I just want my AI to be racist.

[@scottmiller2591]

Yes. Neutral is best.

[@jankram9408]

Whoever made these companies think they have the right to make the field of allowable use for LLMs? These tools can legitimately serve as a good chunk of intelligence of some people so limiting it in anyway and "safe guarding" thought is inhumane.

[@magikarp2063]

Woke is dumb so making something woke makes is dumb by definition tho.

[@MrD3STR03R]

When are you moving to bluesky

[@lordsneed9418]

after he helps create accounts for his wife and her boyfriend.

[@SpentAmbitionDrain]

And be censored day one unless you belong to the hive mind? How about never.

[@WhoisTheOtherVindAzz]

​@@lordsneed9418 ah, you most be "anti-woke".

[@BubbleTea033]

Honest question, what happened to Threads? I know BlueSky is hip and new, but isn't Threads still kicking around? Is it just that people don't want to deal with the Zuck?

[@emmanuelgoldstein3682]

@@WhoisTheOtherVindAzz obviously

[@belthesheep3550]

That graph at 4:17 just put me off the entire rest of the video. I'm right wing politically, but I'm an anarcho-capitalist, basically as far right as you can go - and to see me getting lumped in with fascists because this guy doesn't know history and thinks more right wing = more pro life is disgusting.

[@cdkw2]

wtf was that intro 💀

[@tippykaffu4047]

well well well

[@shipmcgree6367]

Pattern recognition isn't racism 🙊

[@timtim101]

It is because anything other than pure racial apathy/indifference colourblindness is at least somewhat racist. However racism itself isn't necessarily bad at all since it's often rational and makes the world better.

[@WhoisTheOtherVindAzz]

Ignoring biases (confirmation and availability bias being especially relevant, but also e.g. survivor bias) can lead you to merely think you are recognizing a pattern. Not taking context and history into consideration and insisting on having false beliefs in non-phenomena such as free will (your so called pattern recognition system failing to see how much the environment (e.g., social, cultural and economic affordances) influences the choices people are able to recognize) can also lead you down the racist/ignorant path. Etc. Etc. Etc. Etc..

[@WhoisTheOtherVindAzz]

​@@timtim101Hoe can you be this confused? Racism is literally jusr bad. What you are thinking about is being critical of cultures/religions/ideologies. But such matters only incidently or contingently have something to wih race.

[@Dragonflyheartart]

It is if it’s built by racists. Trash in trash out. Omg how could my weak girl mind know that saying. You aren’t the only ones noticing a pattern 🙄

[@emmanuelgoldstein3682]

@@WhoisTheOtherVindAzz "economic factors" lmao

[@rocstar3000]

VSCode and not Vim, shame on you...

[@mnemot]

2:08 np

[@anonieljorgitopedro1786]

Maybe IA developing its not for USA. Let the non anglo countries do the hardwork as always. 😅

[@hotlineoperator]

Woke and bias are real problems with AI. Users should have right to select what information they process and how to use it, not the tool.

[@Gustavoooooooo]

Try freeplane

[@pylotlight]

Just because the article doesn't directly connect to human intelligence, doesn't mean being racist isn't correlated to high intelligence. Can't prove either way ;p

[@ugthefluffster]

my dude, can I please ask you to talk more slowly in these videos? your heavy accent and swallowing of words make me mishearing a word every third sentence. or maybe do more takes, or have someone listen to it before you upload. its great content but frustrating to listen to because I have to scroll back all the time to make sure I heard you correctly. keep up the good work!

[@gerdaleta]

😮 also is a 12 level intelligence😮 I've maximize my racism😮 all humanity is a subordinate race in comparison to AI😮 as a race you're all equally s*** I can prove it mathematically😮

[@kingburrito7773]

People are racist . Get over it

[@wild1000022]

Jesus, this comment section needs a cleanse. Feels like every who commented either commented before watching or didn't watch at all

[@mika34653]

LMAO

[@lordsneed9418]

look at this dood! He changed the title of the video because he was getting Blown TFO in the comments so badly!! The title used to be "making AI less racist makes it dumber?" Oh nononononono AAHHHHHhahahahahahaha

[@technolus5742]

@@lordsneed9418 when I saw it, the title was something about woke. The fact is, *the whole video was debunking both the title you saw and the title I saw.*

[@Dragonflyheartart]

Tech bros taking issue with not building their ai around their own bigotry is so tired. From a women in the field

[@emmanuelgoldstein3682]

Do you ever have deep conversations with people that challenge your worldview?

[@Speejays2]

@@emmanuelgoldstein3682 Do you?

[@Sydra.]

@@Speejays2 Do you?

[@TheMrFrukt]

Bait

[@Speejays2]

@@Sydra. Do you?

[@Hey_Mister]

4:15 I'm not sure if you are implying socialism is way worse than it actually is or if the worst case scenario for left wing extremism is socialism instead of communism/fascism but you are wrong either way 🤦‍♀

[@Kadamitas-II-HATF]

Socialism is far left.

[@demo_AAA]

Hmmmm….maybe I should follow the AI’s example

[@awaisamin3819]

Take me under YOur Wing NOT kidding!