Who else found this interview fascinating?

One reason I'm good at describing what I want, in English, is the years I've spent programming--learning how to formalize and decompose problems. People will still need those skills for a while, though there may be other ways of developing them.

There will be no programmers programming, but there will be the need for the programmers. Have you talked to a non-technical person talking about tech? They don't know what they want. The value of the programmers will be their logic. It's how to break down big goals into smaller, specific ones to get what the team wants. So programmers will still be needed.

Writing code is the final product of a much longer and more complex process. The most difficult aspect of ‘programming’ is not writing code, it is the interpretation of requirements and the design of structures to meet those requirements. The requirements are normally presented by users in an imprecise, fragmented, and internally contradictory manner. Almost all end-users are incapable of supplying a workable specification (they are task focused people, not software architects). The most likely effect of the LLMs will be the production of code snippets that ‘programmers’ can use as components when building bigger structures.

Nice call outs. It's important to stay critical in the face of something that is this exciting. Appreciate the breakdown. Great work.

I didnt think i'd see an android interviewing Sam Altman this early in the AI revolution. Truly spectacular stuff.

Great Video and thanks for showing the Interview. I don't really like to decrease the usability of the UI. However, having an AI that can perform any number of actions that the UI is capable of, or advanced program / app interop /interaction will be helpful. I believe that we shouldn't want to decrease the UI so much that we rely on the LLM / AI.

People like to interact and learn in a wide variety of ways; visual, auditory, tactile... The way we prefer to interact with AI would likely be just as varied, and some people do enjoy their screen time while others prefer to hear it through their Ray-Bans. Some would rather not have any direct interaction at all while others long for that direct neural interface. There's plenty of room for all kinds of modes of interaction.

The problem with "reputable" sources is that is essentially an appeal to authority which is chosen out of a confirmation bias. The AI needs to be sophisticated enough to take in a broad range of human perspectives and be capable of even generating perspectives synthetically in order to actually produce reliable output.

Ok, it looks like, that there will be no AGI in 2024, but likely some improved models. Well, that's totally fine! Having models, that solve more complex tasks increases the number of possible use-cases. I would love to see models this year (from whatever company), that are able to break down high-level tasks into details and then go back into high-level when required. That "focus on the big-picture mode" seems not to be possible with current models, yet, but is super important to solve more complex tasks.

Hopefully, AGI will be better at distinguishing between ethics and politics than our "representatives" are.

User interface might be augmented but it will never disappear. We need to be shown all the options. If we are limited to do only what we think we can do our ability will shrink a lot. We need visual offering of various option.

I really hope graphic user interfaces will not go away. Beyond the simple beauty of it, in many cases it is much swifter to interact using to tap, slide, zoom, swipe, etc., than using voice.

How in the F is NYT a reputable source

We appreciate your break down of this interview with the balanced insights!

The sooner the addition by a inference model is integrated with some sort of reasoning the better, a lot of noise will be added by news and a stable core of ground triuths could help filter that.

22:00 I am 100% against armies and war but for defensive and rescui operations that would be an amazing tool

🎯 Key Takeaways for quick navigation: 00:00 🚀 *Introduction to Sam Altman's interview revelations on GPT-5* - Sam Altman discusses the unexpected success and utility of ChatGPT and GPT models. - The technology is more integrated into daily life than anticipated. 01:24 📈 *Progress and expectations for AI and GPT-5* - Improvement in AI intelligence and problem-solving capabilities. - Expectation of GPT-5 to provide significant advancements over previous iterations. 03:25 🧑‍💻 *Impact of developers on AI advancements* - Cross impact of model enhancements and an influx of developers. - Sam Altman predicts model performance as the primary driver of AI progress for the year. 05:31 📊 *Incorporating real-time data and overcoming current AI limitations* - AI will improve access to specific data and real-time interaction. - Large language models are expected to improve in inference and contextual understanding. 06:39 💻 *AI as an operating system* - Shift to using AI in everyday computer operations. - Conversation-based operating systems are the envisioned future for human-computer interaction. 09:25 🔬 *AI's role in accelerating scientific discovery* - Large language models expected to contribute to scientific research and discovery. - Sam Altman sees significant potential in AI aiding the advancement of science despite near-term limitations. 11:16 📰 *Content licensing and AI training* - Discussion on the importance of trusted content in AI responses. - OpenAI's approach to training on copyrighted versus public domain and licensed content. 15:09 🤔 *The role of data in AI training* - The contradiction between needing extensive amounts of data versus higher quality data for training AI. - Future models may rely more on reasoning with less data. 17:16 🗳️ *OpenAI's approach to election security* - OpenAI's proactive initiatives for election security. - Recognition of election security as a significant concern, with adaptability and partnerships as key strategies. 19:52 🤝 *OpenAI's workforce efficiency and impact* - Quality of work over quantity of personnel in tackling AI challenges. - Sam Altman emphasizes the capabilities of a small but specialized team at OpenAI. 21:14 🛡️ *OpenAI's policy changes for military use* - OpenAI updated policies to allow military use cases not involving lethal decisions. - Ensuring alignment with the company's values and supportive stance on liberal democracy. 25:08 🌐 *GPT's adaptability to diverse global values* - GPT models to reflect users' values, potentially customized by individual preference. - OpenAI envisions a balance between global standards and regional or personal customizations. 29:48 🔍 *Sam Altman on the reflection of values in GPT responses* - AI models may vary responses based on cultural, country-specific, or individual values. - Commitment to upholding essential ethical principles regardless of regional differences. Made with HARPA AI

Voice Interface: no way is voice going to take over GUIs as the interface of choice, imagine how annoying a train full of people chatting to their devices for simple navigation through facebook. Not to mention the privacy concerns.

There is always more knowledge, and new ideas with new perspectives, outside the walls of a company than inside. Point being, going to other entities and offering the opportunity to provide input on the _very_ important topic of elections is actually a sound decision. It's essentially research conducted by SME. Secondly, OpenAI appropriately sought global input. That indicates there is an effort to protect more than just US elections but all elections. The input received wouldn't have been just input to help OpenAI for the States but also elections in the 3rd parties areas of knowledge. That was a very impressive decision.

Great video as always Matthew, I’d love this presented in a podcast format.

He’s also saying, they want to only return answers from approved places. I understand the good intentions behind this. Unfortunately, there is also a huge negative to doing this. Our (even recent) history is filled with the majority being wrong…

@matt , why do you do a video about how do you use ChatGPT for your work and in your personal life ?

30:00 I think the point is GPT-5/6/AGI, or whatever it will be called, will be customed to the user, i.e. a personal assistant. So not specifically tailored to a country but to the person.

18:30 What Sam said was they need a huge amount of data but the models will concentrate on a smaller amount of _quality_ data.

The real Question is: "What is copyrighted information?" I've used the same basic looping formulae for 50 years. What of the "Copyright" of all the intro to programming 101 books that include it? Consider this: If: Times publishes an article, and I read it - Am I allowed to form my opinion based on the article? - does my opinion now belong to Times? - am I allowed to say what Times said? What does the copyright protect? - The Idea? - The Reporting? - The Opinion presented? - The actual sequence of words published? - The enfluence on readers? - The money Times wants to make on all of the above? What does the copyright protect?

That subtle punch to the board 😅

Love the direction of your channel. Keep it up.

Regarding the callout of synthetic data, I would like to interject that in general, synthetic data does not get counted towards amount of data used in the LLM space. For example, a thing we already had mid last year was WizardLM, which if I read the paper correctly, took a dataset, but instead of training on it directly had a capable language model generate a bunch of different variations of a piece of input data. In that sense, the model wasn't finetuned on the original dataset at all. Also, this is just part of what is done for any sort of deep learning called data augmentation, which uses clever techniques to get more mileage out of some amount of data. With all of this, what people really talk about when they mean training data is the human-generated data / input dataset raw data, since any sophisticated training loop can generate arbitrary amounts of more variations of the data and as such the amount of synthetic data isn't meaningful to talk about unless you are trying to make statements about training time estimates.

There will always be need for programmers, especially as a system becomes more complex. I think what people are seeing now is that programmers simply arent being paid that much, because there are so many of them in India and China. This doesnt mean that there is, or will be, less of a demand for programmers.

Ina Fried (she), formerly Ian Fried (him), is an American journalist for Axios

BTW, in regard to the NYTs law suit against Open AI there is a third option where Chat GPT can redact certain material and then one can pay to un redact it, say ten cents or whatever and that goes to the owner of that material. That way the material can be in the model and accessible but not free. Thus the owner of the material might opt not to opt out but cash in instead.

thanks Matthew for your very informative videos to keep us updated on the latest AI news. In relation to your point "OpenAI's approach to election security", the answer is 'they outsource it because OpenAI has no clue how to deal with this topic internally'. OpenAI team is way over their head in managing chatGPT technology. Which has been a surprise even for its creators and computer scientists in general.

When an LLM trains on public information, isn't that information not regurgitated verbatim? It is heavily paraphrased ue to the nature of the training. The training is just approximating the data and outputting what it thinks the prompt is asking. It's like my friend telling me about a story he saw in the NYTimes. In this respect, the LLM is an entity just passing on information interpreted by the black box of the algorithm. On the topic of programmers. I believe there will still be a need for programmers, but in the computer science level where the language is developed rather than the program itself. The LLM, currently, can only use the languages that exists, but programmers are needed to make new languages, or make the existing ones more efficient. I'm sure LLMs will help in that respect, but human intervention is needed. Another place where programmers are needed is in the MCU world. Programs are intrisictly linked to the physical circuitry surrounding the MCU, and if an MCU changes its internal circuitry, human programmers will need to update libraries, or lower level programming so the LLMs can accurately provide the code. High level programmers will be no more in my opinion.

i don't thing he contradicted himself when talking about amount of data, i think he was referring to data distillation. yes lots of data in its raw form from anywhere you can get it, but you then distill that down to a very small fraction of that data to actually be used for the model.

You get a fresh cut my guy? Looks good. Thanks for the excellent and consistent content

great! - thank you for the update.

Ilya is “unassigned”. You can find him on the roof of open AI with other unassigned employees.

A.i. is simply a translator. Programmers used to translate english to code. Now machines understand english.

A few few minutes into the video you mentioned pre-ordering something? Is there some new device that is available for pre-order? What is it and do you have a link please?

Matt, I'd like to know what you think about the relevance of some of the questions starting at about the 20 minute mark. Open AI can take steps to make their models harder to abuse in bad actor situations, but once a certain level of difficulty is reached, won't bad actors just use another one of the hundreds of models to get the job done? Like if a serial killer can't buy a gun or even a knife, he can just head down to a hardware shop and pick up a wrench or something.

This is interesting. I don't believe programmers are going away anytime soon. The whole reason that engineers are needed has nothing to do with writing code. It has to do with reasoning out peoples ideas, figuring out how they are best implemented for computers, and finally writing code. Perhaps some day the AI's will hit those first two, but for now, they're very much stuck in the realm of "script kiddies".

I couldn’t answer that interviewer with a straight face. Every time dude talks, I’d just giggle a little 😂

Star Trek Scotty going back in time tries talking to the mouse as a microphone LOL. I SO wanted to see that in real life - sounds like its time!

I also agree that user interfaces will need to be re-imagined. Most likely interfaces that are extremely dynamic.They will be graphical but it won't be a static user interface that is the same for everyone.

Which one is Sam Altman?

If they did train future GPTs on 'everything' they could get their hands on, then the honest and fair thing to do would be to give people a slice of the pie in terms of the money that this GPT is going to create for an ever increasing glocal AI economy... This could be done with micro transactions such that user input into chatGPT is paid for, because after all, all human ideas, everything we have ever written down and every invention all start in the mind of an individual. We are giving OpenAI all this input data for free with no credit no royalties, and this is a vast amount of data that could be used to make the AI smarter... So yea, its time they started sharing the AI wealth with its users, its an ever growing pie/cake, so they can do this.

@Mattew: I assume that visual interfaces will stay, because there are lots of flows where it's easier and faster for a user to use mouse or keyboard, then to speak. Speach is analogue single-channel means of comms.

Programming and Software Engineering are 2 completely different things. Programmers have existed even before modern computers were invented. Programmers aren’t going anywhere. But the barrier for software engineers will be reduced to "basically anyone can do it". At this point, SWE is basically getting closer and closer to Prompt Engineers

Question: why is it wrong to train on copyrighted material which is provided at no charge?

BRO YOU CLICKBAITED ME. 0 NEWS ON GPT5

Great video but one very important remark: It‘s highly unlikely that GUIs will be entirely replaced by language-controlled interfaces. Most of the tasks and actions we perform on computers nowadays are implicit and direct, they don‘t require the "cumbersome" and indirect process of forming language in order to reach point B. More likely will LLMs be leveraged to tailor the selection of controls and possible CTAs based on context.

i agree 100% with everything Sam said, re copyright law. i was a print reporter for half a decade and the law is almost certainly on 'fair use' side of things.

Programmers aren't going away, though their roles will shift. What is going away in large part is today's programming paradigm. But today's programmers can and will adapt.

at 16:40 when you say sam contradicted himself about needing less data, but that he still intends to gather more, did you consider the fact that larger models generally need more data, and that his meaning could be that for a GPT-4 size model, the amount of data required to get the same result will be less, and that they are planning to continue to gather data while also increasing the efficiency of data consumption with the intent to scale a model to a much bigger size? There is also the possibility that he intends to collect different modalities of data, and as a result of this multimodal learning the proportion of data required overall is less?

I am very excited about these advancements! What an exciting time! :)

In regard to the "middle ground" of military applications, building chat bots to train operators and maintainers could likely hugely benefit from AI. For example one might make a Chat F-16 to train maintainers to keep complex systems like the F-16 flying by training maintainers as well as assisting them in their job to maintain them. That said, the nature of warfare will likely dictate how AI is used in that if say China and Russia use AI for something, the US will be temped to do likewise to keep up or get ahead. Thus national Security will always be a huge driving factor in regard to new technology.

WHY NOBODY TALKS ABOUT THIS? when customs GPTs was released the model was able to reply with Files and documents. I tried and gave me a fully formatted Word Document with images. Within hours this functionallity was removed. What are they holding? Is AGI already invented?

I would definitely like to see you pursue the issue of AI and election integrity. It would be very interesting to have a deep look into these organizations that OpenAI is handing the study of election integrity TO. Since the Democrats have tried to redefine democracy as 'all the policies, practices and perspectives of radical Democrat leftists', the idea of protecting and preserving democracy has quite an Orwellian meaning when these people talk about it. True journalists out there will get to the heart of how much bias, political & corporate influence and leftist assumptions pervade any of these and other organizations that are working on election 'integrity'.

I cant see voice chat replacing spreadsheets totally for example. There were reasons we took that route originally.

don't be afraid, the mind type a programmer has is not being replaced. It's still needed to build systems through the AI new interface.

Is that Sam Bankman-Fried with makeup & glasses on interviewing Sam Altman?

I think the same mistake with sql is happening, that was supposed to be something everyone could use and understand. human language is multi interpretable, I personally find it easier to explain a computer something in computer language then to explain a human something in human language. Maybe it is just me but AI is still slowing me down in every possible way I tried to use it. I might give it another go with the next break trough, which chatgpt 5, or whatever they are going to call it, is not going to be.

Just wanted to point something out: I do not believe he contradicted himself (15:53) - it's just a misunderstanding. Regarding their models using more data for training and their models will 'needing' less data: In other words, [for now] the team will continue to supply their models more data for trainings, but the models will [eventually] require less over time as its reasoning skills improve. A good analogy would be giving a tourist directions to the beach from downtown. Sure, at the beginning they will need to rely on your instructions (more dependency on more data) to get there. But, eventually, they will be able to pick up clues on how to get their without your instructions (less dependency on more data) -- like seeing and/or smelling the ocean, observing the public with beach items migrating to a general area, etc.

I'm having a blast asking about language learning with Bard , and I hope synthetic data will be a lot of fun .

🎯 Key Takeaways for quick navigation: 00:00 🚀 *Overview of GPT-5 Updates* - Sam Altman reflects on the unexpected success of ChatGPT and GPT-4. - Technology surpasses expectations, proving more useful and integrable in daily life. - Launching future models with a different perspective based on the success of current ones. 02:05 🤖 *Expectations for GPT-5 in 2024* - GPT-5's focus on generalized intelligence improvement. - The anticipation of GPT-5's significant advancements in problem-solving. - Emphasis on the continuous overall capability growth of the model. 03:25 🔄 *Gains from Models vs. Developer Contributions* - Expected gains from both more powerful models and increased developer involvement. - Recognition of advanced models as a crucial step forward in AI evolution. - Emphasis on integrating AI into workflows with a dual focus on research and productization. 05:04 🔍 *Improvements in AI Capabilities* - Anticipation for improvements in accessing specific data for more relevant context. - Expectations for addressing current issues like slow voice processing in AI. - Emphasis on the importance of real-time information retrieval for large language models. 07:08 🌐 *Shifting Computer Interaction Paradigm* - Insight into the envisioned shift in using computers through language-based interfaces. - Speculation on a future where workflows happen inside language models. - Prediction of a gradual transformation in computer interaction over the next few years. 09:38 🚀 *Future Work Environment with AI Assistance* - Expectation of increased usage of AI assistants for daily work. - Speculation on a transition from occasional use to constant integration into workflows. - Recognition that certain job classes, like programming, are already heavily reliant on AI. 10:50 🧪 *AI's Role in Scientific Discovery* - Belief in AI's potential to significantly accelerate scientific discovery. - The prospect of large language models autonomously making new scientific breakthroughs. - Acknowledgment that this might not be achievable within the current year. 11:57 ⚖️ *OpenAI's Approach to Data Licensing* - Differentiating between data for training and displaying content at inference time. - The significance of licensing deals for real-time information from trusted sources. - Emphasis on the goal to display high-quality content rather than using data solely for training. 14:13 📜 *Copyright Challenges and Data Usage* - Challenges in respecting opt-outs and avoiding copyrighted content regurgitation. - OpenAI's efforts to minimize reliance on copyrighted material for training. - Acknowledgment of legal complexities and the evolving nature of data usage. 15:37 📈 *Increasing Model Intelligence and Data Dependency* - Sam Altman's assertion that models will need less data as they become smarter. - The importance of a small amount of super high-quality data for improved reasoning. - Contradiction in stating both an increasing need for data and a decreasing dependency. 17:16 🔐 *Securing Democracy and Ethical Considerations* - OpenAI's efforts in securing democracy and collaborating with external groups. - Addressing concerns about the potential misuse of AI technology in political campaigns. - Acknowledging the importance of ethical considerations in the use of AI for democratic processes. 17:31 🌐 *OpenAI's Approach to Election Security* - OpenAI's blog post on Democratic inputs to AI grant program. - Outsourcing election security tools to multiple global teams. - Acknowledgment of the importance of staying ahead in addressing new challenges in election security. 19:37 🔍 *OpenAI's Focus on Election Security and AI Influence* - Sam Altman's historical concern about election security and AI influence. - Limited personnel dedicated to election-related work at OpenAI. - Acknowledgment of the potential misuse of AI in political campaigns. 21:00 💼 *AI Models' Efficiency vs. Workforce Size* - Discussion on the misconception of workforce size as a metric for AI model capability. - Emphasis on the productivity and impact of a few highlyskilled researchers. - Comparison of OpenAI's achievements with a smaller team against larger companies. 22:22 🛡️ *OpenAI's Policy Change Regarding Military Use* - OpenAI's decision to allow military use of its models. - The importance of avoiding blanket restrictions and considering legitimate use cases. - Sam Altman's support for liberal democracy while addressing military application concerns. 25:50 🌍 *Global Standards for AI Models* - Anticipation of the need for global standards in AI usage. - GPT customization for individual users based on their values and preferences. - Recognition of challenges in aligning AI models with diverse cultural and national values. 28:12 🌐 *OpenAI's Stand on Government Requests* - Discussion on potential conflicts with government requests. - OpenAI's commitment to avoid actions conflicting with fundamental principles. - Balancing respect for diverse values while ensuring ethical AI use. 30:02 🔥 *Clarification on Sam Altman's Activities* - Clarification on fundraising activities for AI chip efforts. - Dismissing misrepresentations about external fundraising activities. - Highlighting the strategic importance of chip efforts to OpenAI. 32:48 🌐 *Sam Altman's Investment Perspective* - Clarification on Sam Altman's investment role and ongoing obligations. - Support for startups with aligned goals while being cautious about proximity to OpenAI. - Shifting focus from personal investments to strategic endeavors. 34:39 🚀 *Future Considerations for ChatGPT* - Sam Altman's advice on using ChatGPT internally for increased productivity. - Encouraging companies to focus on building efficiency within their workflows. - Recognition of the critical role of infrastructure investments in the AI domain. Made with HARPA AI

People are assuming the NORAD / NSA doesn't already have their own proprietary models. Why would the US Government rely on a company with a sliver of the budget? We've had missile targeting lasers for years. We've had AI for years. There's no way they haven't developed their own systems. It's actually quite easy if you have the money.

Hi, me again :) Another feedback regarding 17:47... Here are my 2 cents: I think it is a great idea to reach out to other teams from around the world for this. Technically speaking, they are only creating the AI - but they probably 'shouldn't' also have too much power in governing AI, in general (you know, because of conflicting interests and biases). Varying perspectives and the reduction of bias are imperative for creating ideas/tools for governing AI - especially when the goal is to make rules more morally fair or safer or whatever.

It seems everyone says Dávos in English. But it actually is Davós. Something I had to look up myself, is that they even say it with / f /.

Great video!

I can't wait until we elect AI systems to run government entities such as medium sized cities, etc. I am working on getting this implemented for my local city.

16:21 It’s not entirely a contradiction. Various projects are making progress achieving the same results with as a given project with less data while more data is still beneficial. It’s like improving the efficiency of an engine and the quality of the fuel on one hand while securing a greater supply of the fuel on the other.

Hi Matt, another great video, thank you! Two thoughts: 1. They probably hired 10 outside teams to get breadth and depth of perspective and to diversify the efforts and results? 2. It wasn't in your video, but I watched the whole interview, and Sam at the end said CEO's should be thinking about internal productivity, and specifically "...thinking about how can I let me software developers get access to a fine-tuned model of GPT-4 on our own code base...." My ears perked up! Our company has a decently large code base (mobile/web marketplace platform) - is that a doable thing now, having a model fine-tuned on a large code base? Thanks again!

WHY THAT DUDE LOOKIN LIKE VELMA ASKIN ABOUT AI

Very fascinating. I remember getting shot down on Reddit last year when I said AI will be writing its own OS for any device on the fly in the near future. I guess Sam is not quite saying the same thing, but it’s pretty close!

That's a great question. When are not "great questions" asked!

What is the thing which is interviewing Sam Altman?

Where do facts live in an LLM if we can't copy the truth from other sources? There can be no ownership of factual data and information. Hands off the private, but public is all fair game for TRUTH AND FREEDOM.

Based on your recommendation I looked into what a Rabbit is. After watching the video I've just ordered one. OMG I cant wait. Revolutionary.

I recently started to learn(?) to ask ChatGPT to code to put some ideas into an executable program. Still practicing. During the process, I wish I understand coding while talking with it since it's much quicker for a programmer to review its code and point out what might be wrong or to find out where to modify. Maybe after 10 years, the coding ability of AI is perfect and we just to need to ask for what we want. Then maybe, we don't need programmer to code.

What the hell is that frankenstein posing the questions?!? 😂 looks like AI made a wrong turn 🤣🤣

That interviewers voice is really suss

Could you please add a link to the device that you mentioned which you ordered? Thanks.

I'm a programmer, been doing it for like 10 years now. Understanding how a computer works is more important than knowing how to program imo. Programming languages are just a tool to make computers do a thing, there is art to it, but at the end of the day it's a tool. But the writing is on the wall, i'm learning AI in my free time, building AI applications etc.

Interviewer was a major distraction.

The danger of AI isn't that it's too smart, it's that it's too dumb but we treat it like it's not. Like delegating your job to a parrot.

OpenAI has trained on sources that they really shouldn't have, because their lawyers are doing their level best to avoid disclosing the data they trained on in multiple lawsuits, even though they've been ordered to. He's probably right that their models have already trained on enough (copyrighted and licensed) data that they don't really need any specific data anymore, but it couldn't have happened without their initial wholesale ignoring of copyright.

Seeing how fast AI is evolving and how stagnant everything else is remaining, I'm wondering if we are in for a Her situation.

That interviewer is hard to take seriously.

The man interviewing Sam is wearing meta glasses and is currently recording Sam. I think that’s why Sam isn’t making direct eye contact😂

Even better than taking to AI, imagine it just monitors your activity on a private workstation and it infers what you want and looks after your well-being - half of the time you won't even be commanding it at all?

I believe there's no issue with using copyrighted material for training purposes, especially given that everyone draws inspiration from others' work routinely. If the output is rephrased or resembles the original, I don't see any harm in it. This is a common practice. Just to add, in the context of treating prompting as an art form, it should be the user, rather than OpenAI, who bears responsibility for how the tool is used.

AGI Waifu when?

What th is interviewing him? What is it?

I see the world moving into a new era of what I would call "Information Colonization." In the future, AI will be so trusted that the average person will not question any information it outputs. As such, the content used to train the AI will define future perspectives of politics, history, medicine and other discipline. With Sam Altman impliedly referring to a bias in the content used to train AI, some sources like the US media outlets are going to greatly dictate what AI outputs when it gets real-time. Places like Africa that will not have the chance to be included in the real-time content generation bandwagon then will have their stories told by third parties. This happened during colonization, especially in the writing of History in books. This time, it happens because Africa and other third-world regions do not see the value of participating proactively in driving AI

We will need a visual interface. Audio on voice command are OK but we need visual data. All we build is mostly visual. AI will help. But voice only AI is not the future.

What in fuck sake is that person? We have a revolution in tech but have devolution as society.

I like these videos, but this is way too long and the most we can get here are tid-bits. So if i/us can get more clipped form of this, it would be much more ideal.

The reason he is outsourcing election security might be that he does not want to dilute his internal resources from what they are currently doing.

Re: programmers, I think the whole thing will merge with everything else and code will become a way to get work done rather than a thing you build once to do the same thing over and over. So creating libraries will be more important than UIs. And your interactions will become training data.