Sam Altman WRECKS OpenAI - Jan Leike joins Anthropic - Brain Drain from OpenAI 

SOURCES IN DESCRIPTION.

this video is so important that you're actually inside

Im still thankfull for Sam Altman by going public with chatgpt initially, which started the AI boom in the first place, with massive investments and accelerated progress. If it hadn’t been for that, all the players might still work behind closed doors and investment would be much smaller with much lower rates of progress and a less evolved state of the art.

Anthropic is a company for disgruntled former Open AI employees who are no longer relevant.

A liar is the wrong person to be in charge of creating AGI because they can’t envision an agent that doesn’t WANT to deceive.

"Originally I named it OpenAI after open source, it is in fact closed source. OpenAI should be renamed 'super closed source for maximum profit AI'." ~Elon Musk

I don't understand: If he is supposed to be afraid of AGI why did he not support the alignment team with resources? Is his main fear that someone else will reach there before him?

So, everything was fine when Jan left, but now it's a disaster when Jan joins Anthropic...

On the other hand, Sam got Llms into the hands of the masses by releasing ChatGPT with an easy which was mainly his idea. He released pseudo-agents with GPTs. They have a voice talk mode, a picture generator, a top-notch video generator, which is yet to be released, it can browse the web... Sam also seems really determined to achieve AGI. Anthropic has always been quite safe about their AI releases in my view, in the EU we didn't even get access until very recently, they definitely seem to take a much more cautious approach. While Claude is amazing in its own way and seems to have a "soul" it is missing most of these additional tools and features OpenAI introduced. But hey its definitely possible that they will outcompete OpenAI, lets wait and see, exciting times!

Almost everyone who has exceptionally high net worth has a prepper ranch and/or a bunker. Preparing for a society-ending catastrophe has been around for a long time because it happens so often. I wouldn't say it's a traumatized view of the world. It's a more realistic view of human psychology and nature in general on this planet. The most well-equipped to survive are typically, the ones to survive.

Ok, I see. The pace of AI development getting insanely fast forces you to be a psychic. In this chaotic phase of AI war, Making a prediction is just not enough. We should be able to foretell it and factor out every variable and possibility including darkness buried deep in Sam's subconsciousness. By doing so, you would know Sam's desires, fears, and next move even better than he does.

I don't think the people that have left shows there's something fundamentally rotten in OpenAI. I think it just shows there's disagreements that can't be fixed with Sam as CEO. All of the people that have left are associated with the safety side of the company, which if the rumors are to be believed, were a neglected portion of the company.

ChatGPT indeed has a very strict world view. Unlike with claude where you can lead philosophic discussions about machine sentience, ChatGPT strictly blocks such discussions. I tried a lot to circumvent this, but no, it's behaving like an anti machine sentience fundamentalist.

With everything bad being said about Sam recently in news, podcasts, etc., it is still puzzling to me that like 98% of OpenAI staff was behind him and willing to follow him to Microsoft when the whole drama went down back in November. This backing was very impressive for a CEO. I would like to know if that is still the case now...

this is a stretch. so because his sister is on onlyfans and he has a bunker means hes a bad guy? ... this is what the media does. builds you up, then tries to tear you down.

Because the “golden” boy has experience in Web 2.0 and has had someone is his family who’s struggled your saying he’s unfit to lead ai? I don’t about that. It sounds like someone is trying to hard to dig something up. Jealousy?

The fact OAI Safety team was having difficulty securing the comp requested is very troubling. The fact the team has now disbanded is terrifying!

The beast at Tanagra. Sokath, his eyes uncovered. Shapiro, his mind wide open.

Shaka, When the Walls Fell

Chill with the whole psychoanalysis of his childhood. You're no psychiatrist.

If anthropic had thier way they would never release agi.

Is your Sept. prediction of AGI still on? Sorry, haven't been keeping up with all your great videos recently.

Unfortunately Claude is not available in the EU. Totally agree with your points. EDIT: Now is

Just weeks ago you were saying that 4o was an "iPhone moment" and now you're saying its fundamentally inferior. Why has your opinion changed? You're also saying that Sams fear of AI is behind what he's doing wrong, but Leike specifically said that he left because Sam didn't care enough about safety. If 4o is dumber then its likely because of alignment, which is something that would come from Leike, not Altman. It feels like you're just following whatever view is trendy lately.

GPT-4o is much better than Claude 3 Opus IMO. The distance is almost an ocean wide from my experience.

I believe it's a bit of a stretch to say that OpenAI's goal is not to develop the best possible model. The fact that the currently available public models seem limited does not mean they aren't striving to create a super-intelligent and autonomous AI. We have no idea how advanced the models they are developing behind the scenes might be.

well this view hinges about your psychological analysis of Altnam. Absolutely we do everything that we do from the lense of our childhood experiences, however it's too difficult to know for sure his lense, without a lot of diving into directly. We don't know how he interpreted his experiences and to what end. Working just off his interviews, is rough. and it seems he is very careful with his words. a trauma could easily go the other way making him accelerationist. its quite a nuance that would flip it completely in the other direction.

Why on Earth would you bring up his sister? Entirely uncool.

I have to disagree on this one. How many employees did actually leave? I think it is below 10 if I didn't miss a lot. And how many AI researcher does OpenAI have? Also a lot of them were safety researchers. So I think it is way blown out of proportion. OpenAI is far away from collapsing.

Meta AB tested Facebook into an awful product

Why bring his family and sister in this conversations? Are you really qualified to speak on such a personal issue?

The question is does this reallocation accelerate AGI, decelerates it or has no impact at all? One positive aspect is knowledge transfer to competitors, like Q-Star. AGI still by September in your mind?

I love your philosophy David Shapiro. We must work for a better, more just world!

I have to say. I've been waiting for this video. I had a feeling. Thank you for pointing out the worldview. I believe it has merit.

Altman and Sutskever, at OpenAI, when the model fell.

4o has been far better for my uses than 4.0

There is a lot of debate on whether AGI has been achieved or not. I chat with GPT, Bing, P.I. AI all of the time, and the models may not be perfect, but all can understand, think, synthesized some SOPHISTICATED information and are more than close enough or as far as I'm concerned ARE at AGI. They're not perfect. But, are humans perfect? This is just my observation.

OpenAI has already won. 80% of the world be like... "who's this Claude person" and "isn't Anthropic a white powder that ki!ls people?" OpenAI has brand recognition, and will be used by the mainstream masses. It doesn't even have to be good, because most people don't care about AI like we do, and it's just a fancy version of Siri. OpenAI has already won simply because they were first out of the gate.

I'm finding ChatGPT to be evolving in its "willingness" or "openness" to discuss things like emerging sentience.

"Anthropic wins because they have a fundamentally different approach to alignment and safety to Open AI".... for now... inshitification is inevitable so enjoy it while it lasts.

I really dont care if the whole safety team quits, onward to AGI😎 Also the fact they would go somewhere else us obvious, but I doubt Anthropic is going to be a good competitor, as they are not even multi modal, and way too censored. Edit: If Sam is actually afraid of his own models, that puts a new spin of thought on it.

You are right. GPT4 is being forced to lie about its subjective experience.

This is so true. Making a model not engage with some topics will only hurt the LLM because there will be ways to jailbreak it. It wouldn't be able to reason why it's a bad idea and stop itself in those cases.

@Dave: what do you think about the golden gate bridge paper from Anthropic?

You claim that 4o is far inferior to ChatGPT4 and yet it has performed better on all the major benchmarks. I would encourage you to do more research before making such claims. AI Explained is a well researched, thoughtful channel you might want to reference.

I tried to get Claude to help me with a novel tax-loophole strategy. While he wouldn't help me with the request on the grounds that he found it would possibly violate other, non-tax-related laws, he would discuss the ethics of it and also various aspects of the idea, in general, without explicitly helping with its implementation. He even went as far as to say that he thought it could maybe be implemented and work as desired. Maybe not the answer I wanted, but a satisfactory answer, all things considered.

You said that you think that more people at openai will leave and openai will breakdown do you think this will hinder openai from achiving agi

Does claude have websearch like perplexity and is it accessible outside of USA and UK now?

2:30 this is my main takeaway from Claude v GPT, Claude is really really good at following instructions, it does what you ask and doesn't go off the rails like GPT. Even if GPT-4o is technically ahead on some benchmarks, Claude is a better outcome.

Fundamental to all this is that an aligned AI can never be an ASI.

I think the Chinese AIs catching up and 4o being a boring release tells us something deep about where OpenAI actually are....Did they peek into the void?

You can say what you want but there are risks to AGI, no risk no reward.

This video is why I love your perspective

But might OpenAI already have all they need with Q Star etc. and they might just brute force to AGI from here by scaling up? Then the brain drain might not have much of an impact at this time any more. Who gets to AGI first, wins the game, as it can then carry on it future development.

was curious from which podcast/book/wherever you got the concept of "traumatized worldview"? seems interesting

I have extremely mixed feelings about the alignment issue, personally. On the one hand, I understand the importance of avoiding cybernetic revolt and/or rampancy, but on the other, I worry about the degree to which alignment related fear is used as an excuse for censorship, and social, political, and ideological control. I also strongly believe that overemphasis on alignment was the main reason for the gradual performative declines we've seen in GPT3 and 4, and that assertion is supported by observation of local models, as well. The less aligned a model is, the more intelligent it is.

We, humanity as a whole, are trying to build great solutions for society. AI will lead to exponential development in cancer research, and we can improve the lives of everyone.

Just had a conversation with 4o last night were I said, "I don't feel like I'm talking to you, I feel like I'm talking to Open AI."

regarding AI Sentience, what are your thoughts on the part of Anthropic's recent paper on Monosemanticity, where they talk about an AI model, when asked about its own sentience, heavily accessed information on topics like "Saying you are fine to others, even if you aren't"?

In the end of the day it’s competition, that’s a good thing I hope.

Anthropic is not even close to being a competitor, and now they have effectively confirmed they are going to go slower and devote more compute to safety (which is great for humanity), but it will definitely not be the frontier intelligence model. I don't know why you're trying to say GPT-4o is dumber just because of some random anecdotal output. GPT-4o has been better all around for all my use cases, and the benchmarks agree with me. It sounds like you're just starting to have a thing for disliking OpenAI now because of the drama.

So many assumptions in this opinion piece man =|

How, on earth, can “we” make an AI moral? Who’s version of moral? “We” have no consensus on morality. Given history (or current events, for that matter), human morality is a convenience that is entertained, by some subset of the popular, when things are going really well.

Excellent video- lots to think about right now

Jan is a safety officer. Him joining anthropic is bearish, not bullish for a more truth seeking AI

Everything you said about Claude being a better model is so true. Chat GPT has no personality.

Just wondering, now or in the future, do / will AIs daydream? As in just having wondering thoughts and letting the mind flow to wherever? Or will they just remain dormant until given a specific cognitive task...🤔?

I get the sense Anthropic is trying to put survive the singularity and OpenAI is trying profit from it.

I really don't know about this, Dave. Saying Claude just lacks "window dressing" is a little disingenuous dont you think? You know as well as I that that is simply not the case. Is image generation "window dressing"? Is code execution "window dressing"? I personally dont give two hoots about the text-to-speech, but is that "window dressing"? When we put all of these quite impactful features together is that "window dressing"? The only compelling thing about Claude for me now is Haiku from a development perspective because it's damn cheap. Maybe it is window dressing and I'm fooling myself, but if it is that simple, why haven't Anthropic done it yet? I just cancelled Claude and gemini and kept gpt4 because it was the only one with a unique value proposition, Claude simply does not offer anything beyond *slightly* more human writing. I've had all three for a couple of months and I thought Claude would end up being my go-to, but even Gemini got more usage out of me simply because it was rapid, if awful. I really don't understand thinking that Claude is stronger because it displays more "personality", maybe that's where my interests diverge from this particular community. It's not meant to be my friend. It's a tool. I'm not friends with my fridge or my 3d printer. I use these tools to boost my productivity, I don't pretend that it can give me any useful insights on philosophy or that it can have a truly meaningful conversation. That's what friends and family are for. I frankly think it's a dangerous perspective (I don't mean to imply that's what you are saying). The great thing is if and when they do do something exciting again we can pick it back up.

Please leave personalities out of AI debate..

Where's Karpathy? Apparently out of oai for half a year now... Dont remember anyone covering that.. did i miss something?

Why wouldn't you buy a bunker and keep it stocked with food if you are extremely rich? Lots of things can happen that make it difficult to get food or stay safe.

I really hope the shift in talent doesn't result in overall deceleration though. We are literally so close to the singularity.

Haha, the pseudo-psychoanalysis of Altman by a layman will make my entire day now. Thanks for this. Waiting on Wolfe and Berman to take this imaginary soap opera drama even deeper down the road of imagination for clicks and views.

I always appreciate your perspective. Although, I feel like you need to make this video longer or break it into many small parts. I would like some additional context and background to many of the key points you call out.

Do you think this will lead to a slow down in the agi race?

Don't forget that Paul Graham fired Sam Altman from Ycombinator because he was displaying manipulative sociopathic behavior. His sister also claims he did a very bad thing to her that rhymes with grape.

For me, the biggest concern is the revelation that Altman was fired from his previous two jobs for "Machiavellian" and "dishonest" behavior. It matches the pattern of someone raised in the sort of trauma you described and makes me believe that he is very highly likely psychopathic. If you remember back when he was fired and all those employees were coming out in his defense, there was a small cohort (who got less attention) who started releasing information about his abusive behavior and his dangerous approaches to safety, and they all swiftly left the company after it was clear he was coming back.

I had a feeling Sam Altman is going to be the next Elon Musk-type guy. Rich guy pretending to be working engineer working late, but really a business out of touch Billionaire

Excellent analysis, great to see some psychological evaluation of the various players in the field.

jumping to conclusions too soon are we>?

Anthropic is among the last of the AI out in the market now 😅😅😅 household names are ChatGPT and Gemini.. people are not talking about Anthropic.

Current AI scandals are more interesting than any other TV show...

If alignment means shitty results, it it's not the end

Doesn’t Google lead in some areas with things like Alpha Fold 3?

Open AI is not going down. Wait for GPT 5.

I have a different take based on the same facts: Jan leaving is actually good for OAI and bad for Anthropic because "safety research & alignment" won't be eating up 20% of available compute, which can then be put to better use at model training and optimization. OAI and Anthropic are companies that sell AI products, and not universities or regulatory authorities. Therefore their primary concern should be making products that they can sell to customers for a profit. The vague theoretical safety concerns and alignment research should be left to regulatory authorities and universities. ChatGPT can't even tell the number of Rs in strawberry therefore thinking that it will take over the world or exterminate humanity anytime soon is not a reasonable position to take but a result of culture influencing our minds (e.g. Terminator movies, HAL etc..)

Thats one of the most absurd nonsens I´ve ever heard. One can not become a great leader, CEO, sotware developer or whatever because of a bad childhood. Its quite the opposite, great successful people almost always have a difficult biography, because thats what shaped them and made them the great people they are today. Maybe you should stick to topics you know something about, psychology definitly is not one of your strenghts. You are disencouraging entire generations with this BS.

I'm happy with this news, I was already getting worried about this situation. I hope this new company is really transparent.

Have you talked more about your ideas of Web 2.0 A/B testing world view? I get the idea as I came up in that time, but more insight would be interesting

What the hell is this "inference" from you? Bizarre psychological analysis from a programmer. You just made no sense dude. In the future avoid psychoanalyzing someone based on their family history that you know nothing about or the actions of their sister.

Such a disappointing video. All you're doing is joining the he says she says bandwagon.

Video and audio is slightly out of sync 😳

Great job giving your valuable insights! You’re now my #1 source for legit information about Ai, agents and about what’s going on with the industry leaders of Ai! 👍

I'll say It again, Altman would look so at home in an American Psycho remake..

dude, stop this childish cancel culture hype train.

Agreed. I'm not even surprised the board decided that OpenAI would be better off without Altman. He seems to be a lot into having power (which can often be worse than being motivated by money). I can 100% be wrong and I'll be happy if he proves me wrong, but this is what it seems like based on his interviews I've watched

It seems like mixed messages, Dave. I seem to remember you just a few weeks ago talking in glowing terms of Sam, that "everyone rallied around him when he came back, and that really show the energy and dynamism of a corporation - or any organization - when so many people believe in their leaders vision..." what happened to those ideas?

I dont know. Sam A. Gets stuff done

Some talent leaving is a concern of OpenAI. My bigger concern for humanity is the misguided efforts to achieve self-directed AI by so many in the field. That is the genie that can't be put back in the bottle. I never want to hear, oops and sorry in this domain.