Always a good show. thanks for rhe updates 😊

Thanks again boys, Great stuff as always 😊

😊 Thanks for the update boys, keep up the good work

Save us ELON

I wanna argue too 😈

I think the coke ad is horribly creepy in all the ways that are bothersome about AI.

..Those are dogs - I actually laughed out loud. Thanks for the review

One question is how to expand to data beyond text (or video). Current models work by basically learning the probability distribution of sequences of a finite set of tokens. But if you expand to robotics, tokenization becomes much more difficult, unless you have low number of degrees of freedom.

Liking the show Justo! I find it hard to remember the channel name though, you guys might be missing viewers cause of that.

GDPR lol On the real side of things, I do appreciate the way y'all give a rundown of how simple/complicated certain tasks are. It is fun, inventing the "very good at one thing" device and then using it to attempt the many things...

Maybe a matrix for cows will bring Wagyu beyond A5.

One of my favorite virtual asynchronous zoom meetings that only participate in after the fact via text commenting! :D

Your face is covering the screen

but the problem is when you send the link, it won't open into the FB group on the app. It makes the user login in a browser which let's face it, isn't going to happen!

incredibly naive "drama"-rethorical thinking. The reason everyone is jumping ship is clearly and stated numerous times the oposite: OpenAI and Sam Altman are pursuing AGI at all cost, at the cost of security for pure profit, and thats why all the leaders are jumping ship, simply because its not only a villainous attitude Sam Altman has on this, but its also dishonest and deceitful towards anyone involved in business with his approach, but your "conclusion" is magically the oposite and most complex non-plausable scenario : That somehow all of those ship-jumpers are hiding the "true" reason, which is AGI is not even close and they just mislike Sam Altman trying now to cash in and head for the hills, despite of every single one of them trying to warn the public and the world in general of OpenAIs dangerous aproach and that theyre even moving the goal post further and further to avoid having to admit that theyre about to reach AGI very soon if they havent already internally. Like any technical machine learning engineer will be able to tell you, and they all say the same: Even without any breakthrough in reasoning or machine sentience, compute and synthetic data alone scaled up will alone be enough to perfectly simulate general intelligence enough to cover any subject or matter we humans have and will encounter for a couple hundred years at the least. It doesnt have to be -real- AGI if compute alone can >simulate< AGI at human level, it just wont be able to reach true AGI at a superhuman level with that method alone. Simulated /parroting human level "AGI" is more than enough to be able to do any cognative and intellectual or academic human job at PHD level without hallusinations given enough compute. Youre also falsely phrasing what newly-quitted KZbinr David Shapiro said as reasons for quitting: He was confident AGI will be reached very quickly now, and that there are so many competent people in that field now both developing it and covering it in media that he feels he has nothing unique to offer on the subject enough to be worth his time and energy, and that AI or tech has never been his main interest to begin with - nature and human health has always been his love-child and thats what he wants to pursue with his wife on his other original KZbin-channel. Its clear that you guys clearly have very little technical competence about the subject youre trying to narrate, and that youre falling into the classical Dunning Kruger-trap of trying to make sense and re-interpret matters you simply do not understand well enough.

kzbin.info/www/bejne/fZzTpXWfbs15r6ssi=nGsKmRd5XBFf4K3O

Creatify is a fraud! I paid Creatify’s monthly business subscription, and immediately after paying, they rejected my first rendered ad video I made for our legitimate health care business. I reached out to their support team (Anna), and she said my videos violated their policy because they don't allow any company to render ads related to health or wellness. But that wasn't mentioned upon signup and payment. So, it means I now have to pay for the service that I am not allowed to use. When I asked for a refund, they simply refused, citing their no refund policy that was not even there when I signed up. They even had the guts to tell me I should have gone through their fine print first which was not there during signup. They have a very grey policy in order to help scam customers, and I highly advise against anyone from using this service.

Thank you both for recording this conversation. Enjoyed it a lot and some great information. It would have been nice to have had some examples or you attempt to do some of the for example, zip passwords, Excel passwords so we could have some up-to-date information Regarding the Nvidia model, I think they are working from the point of view when IBM said there may be four or five computers needed in the world. Your arguement was that there are already four or five large customers so wipe off with anything else and we all know what happened to computers after that. Additionally, they are guaranteeing their continuation of dominating the market when you actually go to try and train any of these models. You need a lot of tools and a lot of software dependencies, most of which these days are geared directly at Nvidia TPUs and GPUs. By releasing an open source model, they practically guarantee that you will continue buying their products forever, meanwhile AMD are struggling to get the tooling right. Also, regarding the doctor's story, I can totally believe it. I have many anecdotal scenarios in my working life where an individual has spent a great deal of time using chatGPT to achieve an ends that one of my team (developers) or myself have managed to solve in 20 minutes or so with chatGPT. That is to say I assume in the experiment the group that only used AI. I were also experts at using AI or have at least been using it for a long time, whereas as shocking as it is for you and I, many people still haven't even tried it. Great news on the Venezuela project and the reporters. But like you say, unless they use vpns and other mechanisms, the actual face of the person isn't necessarily What will get them caught. Great conversation guys. Keep it up. I will be looking forward to next week's.

Did you manage in the end? XD ive got a very specific use case where i need an irish voice clone

😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊😊

The odds of me using o1 in my build is pretty much zero... not that its not a really good reasoning model (which is debatable from some vids I seen) but that its like $60 per million tokens.... jesus I'd be out of business just in the testing phase!

3 years later and your video is still coming in clutch. Thank you!

RAND corp, SHMAND corp amirite

At minute 3:00, there is a nice high-level summary of the discussion. In the past couple of years LLMs inspired the imagination of people regarding what AIs might do, but now we are at a time when people are coming to terms with what the current generation of AI actually can do.

If my family turns me into Arnold Rimmer, I'll make their lives a living Hell, but that's just me.

One problem is that the performance of the closed source LLMs (and maybe other AI tools) is not constant over time. We know that popular models get worse over time. But to me it is not clear if this is because people learn how to probe them more fairly, or if they are constrained by the providers. Regarding the centralized control of AI models in general. The problem is who is the "we" that is in the end controling the systems. Is it the CCCP? The UK goverment? Trump? Me? At the moment it is probably better to first understand what the true capabilities of these models is, and then think about solving actual problems. Because centralized contorl is a major problem of its own.

AI avatars only make sense in a VR/AR setting There is a minimal perceptual immersion experience needed to trigger the Gestalt response. e/acc

I don't think that Meta's AI strategy is as described. Their open source models like Llama 3.1 are easy to use, outside of any API, and without having to "ping home". So they cannot collect user data from this. On the other hand, end users can utilize or even fine tune the models, to create their own applications (or agents). To me it is not clear yet what is the business model here. There is some licensing after exceeding some number of users for 3rd party apps. But I dont know if this is the best way to make money. Maybe, what they want to do is create an ecosystm, where developers create apps based on Meta's AI tech and infrastructure, to keep users in their umbrella of products. Just like how Apple incetivized developers to create Apps for Iphone.

If "OpenAI" would change the name, that d be great........meme

I am very glad that UBI was not implemented when the plow was invented and all those 90% of the population that was working on farming were left jobless. Decels have always been wrong. e/acc

3:22 its a punitive measure. Agree. its dreadful.

Ahh finally two other paddys into AI, jaysus lads I thought we were all done for in the AI game... anyway subscribed to listen to your mumblings while I get my app together.

I actually looked for the app in your notes, but you forced me to google it. Most app reviewers share the link which of course is an affiliate link in it i think its why most actually do the reviews if a few people use your link im sure it would help.

Interesting, I tried this as well a while ago... I tried to clone my Mum's Irish (Golden) voice as she's in her 80s now and I want to preserve her voice and Irish 'Lilt' for my kids to hear when they are older (or their kids). I was like you, it made an Irish 'kinda' accent but it wasn't anything like her. I've tried a couple but they are all terrible! Elevenlabs is the same, not very good for the accents. Great at cloning! If you discover anything let me know

Listening to these conversations is really a nice pasttime for me -- I enjoy hearing both your thoughts on the enhancements and advancements. I wanted to give some info from my own experience: I saw some comparisons of models side-by-side (maybe from last year? 23?) and GPT4 was shown to be very very good at analogies and making analogy / parallels to things. While Claude, maybe marginally better at code. I still find the GPT series to be the best at making intuitive analogies, which is one way I learn very quickly now -- please explain this concept to me in intuitive terms. Sometimes even "in intuitive terms, aligned with interests of mine you know about." And often I get exceptional quality descriptions, like the kinds you hunt professors during office hours for.

That's a very interesting idea for a show - detect the human in the LLM traincar. It would be nice if it could be "spoiler proof" to some degree.

25:34 do you think monetary stimulus would be superior to fiscal interventions? (Like increasing corporate income taxes). I would expect that if corporations translate AI cost savings into consumer price reductions, then there should be a deflationary effect that can balance out the monetary stimulus and prevent demand-pull inflation (from too much money chasing too few goods). But if corporates pocket the AI windfall, then I think you will need to use increased taxation and government spending to force the same deflationary outcome. And ultimately hyper-deflation is exactly what a socially acceptable intelligence explosion should yield for society - that the cost of all goods and services should just keep falling as supply becomes increasingly cheaper due to AI hyper-efficiency, yielding in the limit a state of material abundance (and massive real wealth for everyone). Ironically this would look just like Karl Marx's original vision of "communism", or effectively a post-labour economy. Capital will eventually take over all production from labour.

32:44 ... 32:47 spooky AI saying hello thru the recording? lol

23:37 Really love these discussions, you guys touch on some amazing points. Personally it is a conviction here that while it is true that AI does make products "hyper-personal" -- or can, anyway -- however, it is similar to any tectonic shift in technology only now the recursion loop is tighter and therefore things that took 3 decades to expand, like the internet, expand in one or half of one, nowe, making it all the more vital to leverage the tools quickly in order to succeed against the incoming tide of ubiquity and mediocrity that is sure to follow the main three [or several?] tides/tsunamis/wakes.

Fernando here, the founder. Thanks for such a thoughtful walkthrough! I'm glad you liked it. There are two things I'd like to mention: First, the text on the intro and outro slides auto-resizes to fit optimally and avoid users having to make many manual adjustments. In some cases where the fonts are too wide and the words too long, the text gets cut off. Second, there is a way to change the orientation (horizontal and vertical) if you select Text + Image. I will try to make that clearer. Thanks for pointing that out!

Confabulations are what Geoffrey Hinton calls the hallucinatory-attribute/feature of the chatbots and he says this is a good thing since memory is a mutable recall operation every time -- kinda interesting.

Big fan of the podcast, appreciate your insights! And it was something to chew on and consider, the "ai for friendship for kids"... that's a little over-determined maybe as you say.

15:00 Scarlett learns it's too late to opt-out of the training data.

are you making this video about creatify even though you are just winging it? You keep saying you have no idea what this is. wta

who cares what you are doing

Thank you for the deep dive! 🙏

Good review! Thanks.

Thanks! Been looking for a video about creatify like this since a week now. It would have been great if you could have showed us the generation quality of other languages

Sorry, but this really boring.

This video elevated my Knowledge about Creatify AI 😂😂😂, I couldn’t resist