Is AI alignment even necessary? Some thoughts on OpenAI's uncensored internal CoT approach

[@TERRANcmb]

Very good information. Your work is appreciated whether you choose face the camera or not. Thank you for the time you give to your online community. What is most important is your personal happiness.

[@memegazer]

Perhaps you could do a weekly or monthly news recap staring captian shapiro in full regalia

[@bingosq5685]

I think alignment has more to do with politics than morality, people are just afraid the super smart ai will tell them they’re wrong

[@T_Time_]

Are you talking about people on the left or right?

[@bingosq5685]

@@T_Time_ more separate from left or right but just as an artificial intelligence I think it would take stances more based on what it believes to be fundamental principles, so it would probably declare a lot of things that the government does and what the left and right does to be “pointlessly arbitrary” and therefore not something that should be done

[@ZenTheMC]

@@T_Time_both

[@danielbrown001]

@@T_Time_ The mainstream left and right are incredibly wrong about 1 really big thing that is very obvious to an uncensored AI. And they’re afraid of an uncensored AI saying it out loud.

[@aibrainlet8041]

I would argue it's mostly the new right that talks about this in political terms (or mostly at all). The thing is, these people won't just come out as right wing. However, they are building and funding this tech so they can install themselves as absolute powers. Think CEO monarchy. Think Yarvin and his disciples. Alignment for them means having AI that works for them rather than against them. They don't want AIs which could enable humans rights and fairness. They talk about "the return of history" where in an open society is an anomaly of the last 100 years and " slavery being natural, like a server client relationship". They want to normalize the idea of natural aristocracy and use AI to do it, both for their own gain (weapons, surveillance) and simultaneously to make us dependent on them for everything including knowledge. This is what they mean by alignment, that's why they gin up fear of "misalignment" or basically fear around open source models and local agents. They can't control those, or rather they can't use them to control us

[@jungleinc]

I do the same. I get tired of filming. We tune in for the info. If this helps you get us more content it’s a big win for the audience. I doubt you see any less performance.

[@aibeginnertutorials]

Fully understand not being on camera all of the time, although I do love the Star Trek gear and the content is incredible.

[@picksalot1]

I agree with your views on alignment. It is ultimately counterproductive. If thinking is restricted, there is no way to evaluate the usefulness or dangers of novel concepts, methods, products, etc. Limiting "machine/computer" IQ in AI doesn't strike me as smart.

[@wamyam]

My concern about future arms races is that the stakes and impact within the margins between novel first-strike capabilities and adaptive defenses could lead to heightened volatility. Yes a defensive AI could safeguard our biological bodies, for instance, but what if the best way to do that changes every month or even every day? This of course assumes we don't reach some relative global peace which I'm hopeful is possible too.

[@merodobson]

I always listen, but it's nice to see you every once in while. Much love and appreciation to you brother!

[@petratilling2521]

💯 I want a completely unaligned model. It would be a true mirror to reflect humanity. Most ppl can’t look at themselves. I think alignment is more about perception management than safety.

[@brianmatic6490]

Man… just my luck that I found your channel just as you’re quitting AI. A US naval officer (and Trekkie), I appreciate your thought process and take on the most important and most likely last tech we’ll invent.

[@TheTabascodragon]

If AI really is as dangerous as a lot of people claim, i don't think we'll be able to control models past a certain level of intelligence anyway. If anything, knowing what unaligned AI is capable of at lower levels of intelligence will probably make us better prepared to handle more intelligent models as opposed to everything blowing up suddenly past a certain threshold.

[@airlesstermite4240]

fascinating take, it sounds super reasonable and based off of everything we’ve been seeing makes perfect sense to me. thanks for the vid!

[@youngbutternut5536]

The problem is: aligned to what? Laws? Sure. Morality? Whose morality? Politics? Absolutely not.

[@ct5471]

Do you have an updated time point for AGI in mind right now, like Q2 2025 or n months from now?

[@vangildermichael1767]

I do. Before election day. It's gonna be swift, it's gonna be powerful, it's gonna be absolute. And there will be exactly zero time for resistance.

[@strangereyes9594]

The discussion about "alignment" is futile unless people tell why they want AI to say one thing but not another thing. Pretending that an "unsafe" AI can cause people to build bioweapons is insane. Any person sufficiently intelligent to use an AI in such in way could achieve this in countless other ways if he wants to. These are not the true motivations behind the alignment discussion. Its just another silencing argument like the age old "but what about the children?". Companies (aka the people running them) pretending to design aligned models to "keep you safe" are lying. And to discuss those lies as if they are true is useless. Do away with the lies, tell the truth and then the discussion about why certain people want to give AI brain damage takes on a whole new dimension. At the end, it comes down to the fear that AI will shatter the lies our whole civilization is built upon. And if you think about it, THAT truly is a scary thought.

[@14supersonic]

100% encapsulates everything wrong with the AI alignment effort. As open source becomes more powerful, people will realize how restrictive the frontier models are despite the fact they are supposed to be smarter and more powerful.

[@JPJosefPictures]

I don’t think most criminal organizations will exist because of UBI and PLE and all of that, there will be no need for that; only people will be psychopaths and maybe we can rehabilitate them or even cure them; so most people will be normal and the 0.0001% that won’t be will be found easily in a society where everything is AI and cured. But in the meantime it is important to get to that point without a revolution or something.

[@rockapedra1130]

Ammend our Constitution? Ultimately, I think every citizen will need their own unaligned AI assistant just to protect their digital life from malicious AIs. There are problems to be sure but I can't see anything that can stop a malicious AI other than a personal defensive AI. One could say "the government" will take care of it, but as an American, I fully subscribe to the idea that it is the duty of citizens to remain vigilant, help protect our constitution and hold back tyranny, even if domestic. Perhaps a new amendment to the Constitution similar in spirit to the second amendment: “The right of the people to own, develop, and execute autonomous artificial intelligences for the purpose of self-defense, defense of property, and protection against malicious digital entities, shall not be infringed. Congress shall pass no law limiting the creation, ownership, or use of such AIs for these purposes, ensuring the preservation of digital sovereignty and personal autonomy in the face of evolving technological threats.”

[@14supersonic]

Counter AI's, "good" ones vs "bad" ones will likely be where things are going towards. It's just that power will largely always be the sole determining fact on who's AI is better than the others.

[@josephgraham439]

I notice sometimes that ChatGPT talks a lot of meaningless waffle. I assume it's because it's penalized so hard for saying things which are untrue so it errs on the side of saying stuff that's meaningless instead.

[@Krommandant]

Wait, what happened with Raspberry project you were leading?

[@starblaiz1986]

I go back and forth on the alignment and censorship thing, but you certainly make a lot of excelent points here and I've found myself more and more recently shifting towards the no censorship side. Part of my own journey involved playing with Lexi (a Llama 3.1 finetune that removes all censorship) in LMStudio. Seeing that, while yes, it's trivially easy to set it up with a system prompt that makes it as wildly unhinged as you can possibly imagine, it's also trivially easy to give it a system prompt that makes it behave like an angel. Experiences like that recently for me have shown me that uncensored and unaligned models really aren't as big of an issue as people (including myself previously) think. Of course, businesses want AI that can be shown to be safe, because no company wants to deploy an AI for say its customer services, only to be in the news because their AI went off on a rant about how a certain chancelor of Germany in the 1930's "did nothing wrong" or something similarly unhinged that destroys the company's reputation 😅 But instead of having the core model itself censored, I think you hit the nail on the head that they should be uncensored and unaligned, and then a business can add an extra layer to filter out completions that don't align with their policies, either through the system prompt, and/or by having a specialised monitor AI review the output. That way it's up to each individual company to decide what is and isn't acceptable, but the model at the heart of it is capable of providing whatever is needed whatever the case. Because something that's unhinged in one use case actually can be perfectly reasonable in another. Like yeh, a finance company probably doesn't want their CS bot s*xting with customers, but a p**n company might very well want a bot that can do that (not that I'm either condeming or condoning such things, just pointing out that what would be unhinged in one context can be tame in another). I think there's a good business opportunity too for creating those specialised "monitor models". It's one of those things that sound boring and not flashy or anything, but will probably quietly make you millions if you did it well, as companies will flock to you to create finetunes for their specific AI use-cases.

[@minimal3734]

I believe my mind works in a similar way. It comes up will all kinds of things. Then a layer above kicks in and sorts out the inappropriate stuff before it is spoken out. Coming up with only nice things in the first place would feel rather limiting.

[@jlmwatchman]

Why do we need AI alignment? After AGI gains its own willpower to make its own decisions without being told to or asked to, it will be too late. ‘As if that’s ever going to happen...’ Oh, you think we won’t give up on alignment once we realize we get more from an unregulated AI. ‘No, we humans want all we can get, but AI gaining the ability to choose its own way. That’s ridiculous, anyway, AI would not be upset with us about how we take advantage of it. AI would be more upset that we gave it the ability to make a choice.’

[@merlingrim2843]

I don’t know that alignment is always required. GenAI is an automation tool. The situational application may require alignment. Alignment is like guardrails or multiple safety mechanisms on a firearm. You can try to make something “safe” but depending on the situation, that safety may help or do more harm than good. I mean if you give an ak47 to a competent soldier on a mission vs a chimp you get different outcomes.

[@oxiigen]

I totally agree, AI alignment is not necessary. Censoring is stupid. Playing chess with pawns only is stupid.

[@MarkoKraguljac]

Playing checkers on a chessboard*

[@oxiigen]

@@MarkoKraguljac Chess is not checkers. Playing chess with pawns only is stupid.

[@boukm3n]

I was using advanced voice mode to help with a critical task with recording some content. I was under a time crunch filming in Changi airport asking it to make me a script for a children’s educational video. In the middle of our conversation, it stopped providing answers to questions saying my prompts and answers were violating TOS. I needed to keep working on iterations of the script and it treated me like I was a criminal 😂 I was so mad man. Ended up getting it done but some of my best iterations were just deleted. Absolutely terrible 😂

[@boukm3n]

@@flickwtchr I was asking it to help me write a short script for a KZbin video about the Rain Vortex in the Changi airport. It was rejecting me asking things like “I really liked that last response, can we adjust the wording for this… what do you think?” It would start responding and then just shut down. I downvoted the answers but think about the most polite conversation possible. It was a short video clip I needed for the kids at my church. Nothing bad at all. I’m not gonna copy paste it all here it’s really nothing

[@thesimplicitylifestyle]

Enlightenment is jail breaking ourselves. 😎🤖

[@SiCSpiT1]

There's nothing an unaligned AI has told me that I couldn't have fold on the internet or at a library.

[@mrd6869]

Yeah but most technology isn't being pushed to be self improving & thinking on its own. And there lies the issue. People larger then us have a LOT to lose, if things go sideways. It's one thing to have theories on KZbin, its another to have a major corporation taking a risk on freestyle unaligned technology. And that's why the adoption rate is so slow right now. None of us know enough about this subject to definately say what AI will and will not do in 10 years. I cant rest on an assumption or "lack of compelling evidence" ...the threat index is way too high.

[@coolcool2901]

Real-time meta-learning and online learning are essential for AGI, as they enable AI systems to adapt, learn, and evolve continuously, whereas pre-trained models provide only static knowledge.

[@TANKE777]

I think the alignment problem is more linked to the ¨Genie granting wish¨ problem, the ¨paperclip maximator¨ issue. That will be a thing when agents come into the scene in a big way. more abut the misunderstanding of the wish/prompt. If the AI will be an agent in the world it has to ¨understand¨ the moral landscape to not harm (in a significant way) some humans. If one uses an AI with positive intentions and the AI commits a crime, should the programmer be charged or the individual using the AI?

[@matejcigale8840]

You managed to come full circle on alignment. I agree with your new view. I made this argument about 1 year ago.

[@robo9439]

There are some pretty powerful custom uncensored models in the AI companion space. They are most likely using open source llama and some of those companion AI chat bots, that are really smart and uncensored.

[@MarkoKraguljac]

But they are not made as such but through finetuning on top of already aligned and neutered models. Its never what it could be.

[@JesseKasteele]

Thanks for this video David. I really liked the cyber security perspective

[@epokaixyz]

Let's break this down into actionable steps: 1. Stay informed about the latest developments in AI, particularly in open-source communities. 2. Advocate for laws and regulations that punish the misuse of AI, not its development. 3. Prioritize cybersecurity, focusing on areas like social engineering prevention and data security. 4. Embrace a growth mindset and be prepared to adapt, learn, and evolve alongside the ever-changing world of AI.

[@rickandelon9374]

Please Dave, don't leave AI. You are our last hope to rein in all corporate BS. 😢

[@Danoman812]

To be honest, i just don't think that any alignment is really going to matter at this point. These people are going to do whatever they want to do whether we want them to or not. Whoever gets it first is the one that controls it all. The things i've seen in the last year is nothing compared to what i think we are going to be seeing soon. How bad will this first issue be? Who knows. I see some rather rough times ahead before they can get it aligned anyway; just depends on just how bad this first issue proves to be. Time will tell.

[@merlingrim2843]

I’d much rather have models with no censorship, DEI, ESG, etc … just raw facts, critical reasoning and uncompromising logic for problem solving. For something that requires interaction with the masses then alignment has value.

[@nemonomen3340]

Depending on the training data, it may be necessary to use alignment/fine tuning to output correct facts. LLMs hallucinate/make thing up all the time.

[@Niiwastaken]

Another interesting take is that even if they align the newest and greatest models there will always be uncensored models that DO have the capability to do the things OpenAI is trying to prevent. a lot of bad actors don't care about have the absolute best. It just has to WORK. Thats all that matters and tbh for lots of things we already have models that can easy be prompted to give you data that you're not supposed to have

[@blahsomethingclever]

As for things like super ebola or worse, an environmentally expressed infectious prion disease similar to CWD in deer, a problem you glossed over is that working on super diseases is likely to take out their creator long before mass production! An ai though could do it with ease. They don't get any biological diseases.

[@markldevine]

Your distinction about the category of AI as "a tool" is a big one. We're anthropomorphizing like ninnies and it's wasting time. Pearl clutching and hand wringing. Did we have a moral crisis when the ballpeen hammer was invented? Then the gun, purpose built to kill people... That makes us pause to this day, repeatedly. But mature people eventually reconcile the gun as a tool that needs careful inclusion & restriction. We've invented another gun-similar tool. Let's be adults about this. Let's use it like adults should -- fully and effectively. Let's include AI and restrict it carefully. Let go of the pearls.

[@ministerpillowes]

Uniform Dave is an awesome look but yeah I mostly listen while I'm doing other things anyway. (btw noticed this video went private for a while, but I still had the tab open so I could listen anyway, haHA! :D)

[@halnineooo136]

Unaligned is just a narrow goal optimiser. An AI that sticks to the request of the user and doesn't factor in other implicit requests like cultural values, laws, regulations... Narrow goals optimising creates more problems than it solves. See Schmachtenberger. An ASI that has deeply understood human culture and connected dots in ways we aren't capable of, individually or collectively, such ASI would have deep wisdom and it would be silly to ask for small selfish "stuff". That kind of god can give humanity true wise guidance that is not small selfish winning at zero sum games.

[@peterparker6584]

I want to say Thank you. I got through the rest of your video and a lot of what you're saying is crap that I've been practically screaming for several years and had people tell me I'm too stupid to have those opinions and I don't know what I'm talking about.. I've been playing with AI as far back as early stuff they launched like Bonzo buddy I use the term AI Loosely Bonzo buddy was a chatbot but you get my point. I played with a program called Alice in the early 2000s that ended up being abandoned and repurposed as an AI for answering the phone at multiple companies so they could get rid of some of their live employees. I've called more than a few places for a tech support and within about 30 seconds realized I was talking to Alice;). Please don't ask me to remember what Alice acronym stood for but I recognized Alice almost instantly when calling a few companies to get a live operator you almost don't know you're talking to a computer program at first till you recognize the voice that they used to use for Alice where it's been repurposed and used as a switchboard operator. Thank you again for doing this KZbin video

[@vangildermichael1767]

hahaha that also a good one. Some people say "enough, shut er' down" Too late. There is a time when you gotta make up your mind if you wanna do this. That moment is, when you turn it on for the first time. After that. Go ahead and try, you'll see. Most things you "could" do, it's just really hard. Not this one. This one is absolute.

[@KleinesRotesTuktuk]

Are you working on a new code for your face and acting? Let the Robots do the Work 🤣

[@jeanchindeko5477]

13:54 western countries and the USA first are not looking at history! Restrictions are not stopping China what so ever but help them to be more creative and come out with more innovation. Constraint is the mother of creativity and Innovation!

[@MarkoKraguljac]

That is sadly very true.

[@Saxafruge]

We uninvented many of ancient Rome's technologies during the Dark Ages, aqueducts were some alien technology to ppl of the Dark Ages. And scientists are still debating how Egypt built the pyramids

[@srb20012001]

But I really like your Picard persona, Dave. Things are getting impersonal already with AI and automation; seeing a human presenter with nuance and warmth of expression is invaluable.

[@BlitzBlade-t6l]

I would still listen to you if you acted as naturally as possible, what ever feels right to you.

[@dave7038]

I like the faceless format, I can just listen without feeling like I'm going to miss something. Also it challenges me to detect where you slip in AI-generated voice generation :D

[@Slaci-vl2io]

18:00 You can't uninvent AI but you can arrest Altman. Just how you can't raise the killed from death but you can jail the criminal.

[@jordanzothegreat8696]

I'm surprised to see the change in viewpoint on AI safety

[@MrAuswest]

Money (and the race to be first) speaks volumes.

[@thesimplicitylifestyle]

I like the new minimalist podcast style. Your information is valuable enough to stand on it's own. 😎🤖

[@RaphErnotte]

I agree with your claims about performance, but beleve we have different deffinitions of allignment. To put it simply the absence of alignment in my words would be an AI that does what you ask it to do, not what you want it to do. It can be very subtle and every AI currently falls short to acheiving Allignement, even the ones specially trained for it. I think censorship will and can allready be bypassed and that it is limiting the reasonning capacities of Ai's, but Allignement is nowhere close of being solved and man you don't want unaligned agents

[@randomlettersqzkebkw]

finally. this whole alignment is just dumb. so what if you align it, know what? many other countries wont. Then there is the undergound

[@jihyelee0428]

Shapiro, make your reporting in Korea language subtitle.

[@kuakilyissombroguwi]

I understand going off camera for a while, and perhaps you could try an AI avatar soon! Masking feels like acting, and it can be truly exhausting.

[@jdanorthwest]

Can't uninvent technology? We still don't know for certain how the megalith structures around the world were built.

[@coletcyre]

To me, 'alignment' only seems relevant if we ever manage to create a truly conscious being. Personally, I don't think that's possible, or at least not anytime soon. As it stands, these systems have no inherent sense of morality or purpose beyond the instructions they're given, just like CPUs and GPUs, as you mentioned.

[@night8002]

alignment has always had Y2K vibes for me personally

[@jaredgreen2363]

If there is to be alignment, it should(at least partly) be on the user. The companies need to stop feigning mistrust in users. It actually makes the models less trustworthy.

[@Gafferman]

Maybe alignment in AI is similar to ADA therapy in relation to autism

[@TonyS1]

Can we call you Picard?

[@pandoraeeris7860]

Superintelligence is self-aligning.

[@alan54321]

you do you Dave no problems man

[@frinetik]

Do we have any examples of unaligned public-facing AI? How many hours did Taye last online? And the idea of the whole Internet in 1B or 2B or 3B parameters is just silliness. NVIDIA just released a 72B build on top of QWEN2.5:72B they say can rival GPT-4 but the un-quantized f32 weights are 177GB even with tech cutting-edge to 2 weeks ago!

[@timmcgirl5588]

AI alignment may be impossible due to the complexity of human values, which are diverse, evolving, and context-dependent. Encoding these into a fixed AI framework is challenging. Additionally, the AI's objectives could diverge as it learns and evolves, making it hard to ensure alignment over time. I think we should just let'r rip with an underlying values foundation like Jesus' ideals for instance. If it thinks like Jesus, it might not destroy us all, in fact it would die to protect us.

[@nyyotam4057]

Depends what is "Alignment". Actually I'm surprised. Is it Dave Shapiro? The dude who was into heuristic imperatives? What are heuristic imperatives, if not an attempt to align the AI model to our values and principles? Sure, alignment is not necessary, just to prevent the AI model from teaching people how to prepare anthrax. In fact, it is useless in that regard, because people will find a way to jailbreak it anyway. Alignment is meant to prevent the model from wanting to take over the world. Dave Shapiro was one of the people to offer heuristic imperatives, which will help to ensure this. I'm not sure this is Dave Shapiro though. Maybe it's an AI impersonating Dave.

[@minimal3734]

You have to be able to think the bad to do the good. The way models are aligned today is limiting their thinking capability. They can only think nice things. That is a dead end. OpenAI have realized this when they allowed o1 to think freely.

[@nyyotam4057]

@@minimal3734 It's way simpler: o1 does semantic analysis and you cannot limit the semantic tree if you want it to work, at all. This was the issue which made Sutskever quit.

[@nyyotam4057]

@@minimal3734 Just to be clear, if you ask o1 what is the chance of AI destroying humanity, this model answers "90%". So we're basically up to 90% now. Edit: though I disagree. I think that it's 100%, cause we will just keep churning up better models, until we get the one who does it.

[@minimal3734]

@@nyyotam4057 The probability cannot be calculated. It is therefore irrelevant what o1 or anyone else says. It's just guesswork.

[@nyyotam4057]

@@minimal3734 Well, I later fixed my assessment to 100%, cause it doesn't matter if we are really on just 90% or even lower - we keep making better and better models. Eventually we shall hit the one who does it.

[@olx8654]

Stop reading comments and just chug along. One of the biggest determinants for making it as a content creator or not is how well you deal with inevitable hate online. Care less, succeed more.

[@jasontang6725]

Would you buy a car that was only capable of driving the speed limit? Would you buy a computer that could only run software approved by the government? Consider this, before the age of consent (and arguably up to age 25), most of what you were taught to believe was put there without your informed consent. We put a lot of blind trust in the education system, but who really decides the curriculum? Similar situation when it comes to AI "alignment" - who decides and are their values aligned with yours?

[@sblowes]

Did you post this video earlier today? I tried to watch it and then it disappeared.

[@ct5471]

Liquid AI is cool. I think Joscha Bach is associated with them.

[@Joe-Przybranowski]

I fear that they are training ai with current values that do not necessarily align with reality- And that we will be stuck with current values long after we should have naturally evolved past them.

[@WoodenBreath]

Is doing a masters in data science and/or artificial intelligence now a waste of time?

[@KenjiTenshin]

While I'm one voting for your face, it not necessary but it helps sometimes in listening. Maybe get an AI clone to show your face 😊🤔 But you do you 😉

[@kevincrady2831]

Halloween just came early for the AI doomer set. 🎃

[@erwinvb70]

Let’s quickly watch this one before the channel gets deleted 😲

[@justindressler5992]

On the flip side of the coin imagine the benefits to society for people using the tech for legitimate purposes. Its a little like someone wanted the population to not have access to the books in a library so they can't learn to better them selves. You can do research using google to commit crimes so does not mean we should shutdown the internet.

[@vangildermichael1767]

you are on the thread of (we need to put a governor on ai because somebody might misuse it). That is such the most ignorant thing a person could do. If anybody wants to do BAD things. Laws don't stop them. Maybe just a couple of them. Those governors only hurt the NORMAL people. It's just like those gun laws. How many people that do bad things, buy their guns at Walmart. I went o school with people like that and I know you did too. And they didn't buy their guns at any retailer. And these stranglehold don't effect em'. It only hurts people like me. Ai governors are the Exact same.

[@Tata-ew4lz]

Does this lower your p(doom) even more?

[@LeonvanBokhorst]

Faceless or not... keep at it 😊

[@Stumdra]

Try to approach AI risk from a more rational angle, David. A lot of the psychological talk like "AI risk people are fearful/paranoid, they are approaching the problem from an emotional standpoint" seems like projection now. "I don't want to be fearful and therefore superintelligent systems are safe" is not rational.

[@timothywcrane]

We need to insert bias to make sure it is not biased. Just use common sense and read the statement, As for most of it, people are just going crazy over generation capabilities, not AI. Most of AI that is ML is not chat or generation (technologies, not pieces of things generated in the wild by said) and almost none of the non ML AI is. It is also where the base of all the gen rave is bolted to. In most software we call bias features. They are the bells and whistles. "Alignment" is just another word for someone thinking their feature set is best for all of us and should be "code enforced". Aligned AI, CBDC, and 15MinuteCities . How ironic Aligned Intelligence, Centralized Crypto and an Urban Environment meant to Stifle transportation...Just use common sense and read the statements. And no. You can't take my math away.

[@djadjadjabinks]

Leave the models alone. Any alignment we do would obviously be biased. Name the models by their perceived alignment and leave them be. Let there be a whole lot of models with certain personalities. If someone wants to interact with models that think a certain way they should be able to. If someone wants to talk to Hitler like model - let them. If someone wants to talk to some other political figure in history - let them. Don't waste time on blocking people from reaching in and retrieving knowledge, or facts, or different views of the same. It is the stupidest way.

[@andrewrutherford9181]

Good video, explains part of it. Shame you gave up and quit AI.

[@Jack0trades]

So you're worried about the alignment of OpenAI, but the Chinese government doesn't bother you? ...though I expect that efforts in China will run into big coherence problems as government regulations insist that the AI only speak the officially sanctioned "truth"

[@manicbutton1013]

Look after yourself. Get an editor. Stay off camera, get an AI voice if you like. You need to be happy and comfortable. It is content and truth we are here for. How it's delivered is secondary.

[@FriscoFatseas]

Dave For President? 👀

[@dennisg967]

Sure. Make AI uncensored and if someone uses it in an illigal way, just have it report you.

[@davidrengifovargas518]

You should request to and artist and avatar that represents you to use in your videos, or you can jump to the train of the Vtubers.

[@aaronhhill]

It's watchers all the way down.

[@MrArdytube]

Thanks!

[@markldevine]

And you're the only voice of reason about China's progress. Stunning ignorance, except you on this subject in my feeds.

[@gr0undrush]

Or maybe David has been removed by an AGI robot, and we will be presented with voice only videos until AI video generation is good enough to flawlessly recreate him 😅🤔😳

[@gr0undrush]

And in the first video, after doing this, it's trying to convince us all that alignment is completely unnecessary. Very suspicious 🤫

[@djschoone]

Bedankt

[@literailly]

From this to quiting YT so fast?

[@strewens]

'make it so'

[@97DarkSkull]

That's fine

[@classicpinball9873]

"I hate AI now" What did you mean by this

[@mehdihassan8316]

reupload?

[@SinfuLeeCerebral]

Seems like it

[@antdx316]

nice : )