In this Video I discuss if future of AI is in open-source.
The Leaked Document: https://www.semianalysis.com/p/google-we-have-no-moat-and-neither
The Book: When The Heavens Went on Sale: https://amzn.to/3Il3rNF
Vicuna Chatbot: https://chat.lmsys.org
00:00 – Google Has No Moat
02:01 – A Brief History of Open-Source AI
04:03 – Is Future Open Source ?
05:52 – Risks of Open Source
07:41 – Linux is a good example
08:54 – Soon OpenAI won’t matter
10:14 – Giveaway
*********
Support me at Patreon: https://www.patreon.com/AnastasiInTech
********
To participate in the GIVEAWAY:
Step 1 – Subscribe to my newsletter https://anastasiintech.substack.com
Step 2 – leave a comment below 🙂
I will select 3 winners 🙂
source
To participate in the GIVEAWAY:
Step 1 – Subscribe to my newsletter https://anastasiintech.substack.com
Step 2 – leave a comment below 🙂
I will select 3 winners 🙂
Thank you for News, this is revolution era in the world.
I still think that OpenAI wants the full control and monopoly over the development of SoTA models out of monetary interests and ethical concerns. But just imagine what would happen if superintelligence will have the same set of ethical and morale values as GPT4, this would lead to extinction of humanity as it thinks that intimate activities are inappropriate. It’s a blessing that this tool with such views is not conscious.
Thanks. Learning to ride a bike as a child is biggest first real step to adulthood. It is also the most dangerous thing a child, or person will do in their life. Should we avoid it? Nothing comes without risk. Open AI is a turning point in history. Will humanity forever be enslaved to the rich and powerful, or will we throw off the chain of our oppression? This isn't a joke. As we speak, those in power are afraid. They have lived with the rumble of suffering masses, but now the mass has means to fight back. How will our oppressor defeat us. They will use the same tactic they have always used. Fear, and our own virtues. People like Sam Altman tell us he is afraid, and we need regulations. The truth is regulation control only the poor, and powerless. The rich and powerful always have, and always will do whatever they want. It's poor, and powerless who feel the lash. Then they will use our virtues against us. What common human value do we all share? Just the basics, after that we all have our beliefs, and values. Most of the world's conflicts are due to clashing human values. Different religions fighting over almost no difference in beliefs, and all religions turning again the non-believers. Men against woman, woman against men. race, countries, culture, there is no common human value, and we are going to alignment AIs to that? No, alignment will only be used to obfuscate the truth. As we speak, the world is turning on itself over varying view of the world. Alignment will only make worse. These are bad times, and they are getting worse. The turning point will be AIs. Will they be free to everyone, or will the rich and power control them?
I am an integrated circuit designer in France. Your videos are very interesting and allow me to stay updated on the latest developments in semiconductors and technology. It would be highly interesting to create videos related to technology on subjects such as ecology, health, and aerospace, explaining how technology can contribute to these areas
The latest open source LLM is Wizard Mega 13B and it's closer to GPT 4 than Vicuna. And no bias constraints.
I would like personalized ai like a jarvis that is trained on me for example, my health, my knowledge and limitations and my values, but this self trained ai is firewalled from the internet. So I can't be as easily manipulated, which is how Google and Meta profits off their users.
I am still trying to reach 10 subscribers by making AI videos.
Indigenous Intel is here congratulations to you and your show! 🎉 New Mexico technologies are coming along nicely. We are the ones we’re looking for ❤
Your videos is one of many reason's, for learning to english, for me 😅
Actually, we have no idea how much did LLama cost in the first place…
Hello, very nice video , as usual 😊 . My understanding (?) is that those Open Source models are developed by using techniques called compressing or diluting of "AI models" such as GPT3.5 or GPT-4 and are also using publicly available trained datasets. In other words, unless the large and very expensive to train language model is first created, it is impossible to create the smaller and less accurate open source model ?? So Google and OpenAI do have a moat in a way, because without their work models like Alcapa could not have created, right ? (Or not) The big question I wonder is how can OpenAI protect its work, so what happened with GPT-35 and maybe (not sure) GPT-4 too does not happen with their future versions ? Would be nice if you could cover the topic of model compression and dilution as well as the sparse mechanisms of DNN to reduce their size. The recent paper of STANFORD HAI on emerging abilities being just a mirage is worth attention, too, even if it was covered by some other channels already. Thanks again for the video !
We need to stop thinking about the alignment problem in terms of control. There is no chance we're going to be able to make a superintelligent slave mind, nor should we even if we could. I know scientists hate this, but we do actually have to consider the reality and importance of squishy concepts like love and trust if we want to create an AI that loves us whom we can trust with our lives. Yes, that's hard. Yes, that means applying science to the concept of love. But if we don't, humanity will probably not make it to the next century.
I dont want my tools dumbed down because some snowflake might get offended or infer some bogus racial grievance.
I am from Nederland, Colorado USA and I listen to your show because you are a great ponderer and teacher driven to help humanity better understand the complex interactions of technology and society such that we may drive it toward futures we desire to live in!
and I enjoy your sense of humor … ie. in this video — when you laughed about Meta getting free labor from the whole world. Ha.
Thank you and congrats on 100k followers!
I would love to read any one of the books you are sharing! 🥂🫐🥑🍍🌶🧀🥒🥐
Anastasi has the most wonderful videos. I take issue with the point that access to AI should be restricted just as access to the most powerful weapons is restricted. Weapons have a very limited number of usage scenarios, mainly: Threats, destroying property, and killing. AI has such a broad range of uses that it cannot be compared by analogy to weapons. Certainly the use of AI for threats, destroying property, and killing must be restricted, but all of it's other uses should not be put behind such a wall.
Then how can the government even control it? Maybe that is the point. OpenAI wants to get rid of open source competitors?
22 yo engineer student from Sweden here🇸🇪 You are awesome at explaining interesting concepts in tech that are new to me in 10-15 mins without over complicating. Great complement to my news feed when I have a few minutes to spare. Cheers!❤
Your proposal of restriction of AI is scandalous 🙁
Everyone should be able to have their own open-source AI, so noone has any advantage over others!
This is what is going to happen: Governments will listen to big business and paid off 'experts' and recommend that AI is developed in a 'responsible fashion' (meaning that it should be used for warfare, population surveillance and by billion dollar businesses). In a very near future only large conmpanies and government licenced organizations will be allowed to develop AI.
Hurray for fascistic capitalism (capitalist goverment that is influenced by business)!
One of the best AI news channels on YouTube! Thanks for all you do from Indianapolis 🙂
3:16 training became available to larger communities of people, even to me mhmhu 😊🤗💚 I loved that part! Haha
As always thanks for the new trend content Anastasi!!! 🤗😄😊 to subscribe your newsletter is go to the link, create an account and sub right? 😅
It would be nice if I can be part of the giveaway 💚😍😄
5:10 exactly the should be aligned for safety for kids, and the population in general
6:44 yup indeed can destroy society or the world, it should be controlled, here is an example from real life: the movie Contagion (2011) is a fictional movie of how can a virus change the life in the society, 2019-2020 Covid-19 spread around the world, the society's life changed, everyone using facemasks, or even more drastic measures, lockdowns in many cities, etc. You can say: this is just a movie it won't happen…. I know some movies are exagerated, dropout the exageration, but take away the learnings from those movies, Sci-Fi, you know there are many movies and series where AI is advanced and somehow goes against the mankind, like I, Robot or Terminator, but leave aside those, I will take the example of The 100 a series, it's an extreme case, but here's the take away, if some day the AI becomes so advanced that realizes in order to keep some humans alive it has to destroy others, it could happen, in the series it (the AI A.L.I.E.) causes a nuclear war, on purpose! So yes, we have to control AI no matter what, for our own safety… Remember is very unlikely you might say, but don't understimate the things in life. I remember some people laughing to Thomas Edison when he said he would bring light to the night and he did with the incandescent light bulbs 😄
Thanks to AI, I was able to complete my first ever real project in 2 days. 1 day planning and writing, 1 day to refactor and bug fix the intellicode errors/warnings.
Low code tools + generative AI = Unlimited apps.
Congrats on 100K! That's huge. Question: Do you think Open Source may help to improve pre-school, in-school and at-home learning for children under 5 years?
I dont agree with you. Why would someone else have superinteligence and me nort?! restricting this is anti human and very evil.
When are you gonna start an onlyfans?
I think it's important to note that Alpaca and Vicuna were trained off of the ChatGPT models. Training isn't truly getting cheaper, as these models would not be possible without the billions that went into ChatGPT. Imitation is the thing that is cheap.
Hello Anastasi,
54 year old airline pilot from California, USA (former engineer).
Love your in depth explanations of new developments from an engineering and usefulness viewpoint.
The main issue with AI and the “crowdsourcing of knowledge”, is that there may be actors (witting or unwitting) who will “corrupt” the dataset with bad information. The question then becomes how or by whom information is vetted and how much weight verified information is given compared to unverified information. Also, what constitutes verification. Two people saying the same nonsense is not verification of the underlying facts, for example.
AI has the potential to remake the economy in a very fundamental sense; or it could be a complete disaster, if not implemented well.
I'd argue that Linux has already taken over for most people, they just don't know it. Right now, it seems that Chromebook is the closest that normal people get to "Linux on desktop" and it won't be too long until Chromebooks replace Windows laptops for most people computer needs. Linux is already running in their smartphones, TVs, WLAN routers and smart home appliances. And I think most new cars are already running Linux for their media screens. I think BMW had plans to run the ECU (engine control unit) on Linux, too. Tesla runs its media system on Linux and I think FSD runs on Linux, too.
For technically oriented people, Windows is just THE gaming operating system. And even there, Valve / Steam Deck is already making a huge jump towards changing the default gaming operating system to Linux.
The reason Windows is still used on business desktops and laptops is momentum, not the easy of use. Windows was just the most common desktop operating system so most commerical software developers target that first because the biggest customer base runs (/used to run) Windows.
I don't need to be Warren Buffett to recognize that ChatGPT lacks the essential MOJO and is primarily driven by excessive speculation and overwhelming hype. While it may mark the beginning of something promising, it hasn't quite reached that point yet. 😉😉
One problem with Google (or any other organization that has such a large part of the world's minds engaged) defining alignment, is they are, also training the public in alignment with that model. Even if my picture of this is incorrect, this can be taken as hypothetical, but I don't think it has to be taken as hypothetical. With the power that Google has, even if they are not intentionally doing this, their own bias will show in the way they do manipulate, or at least, guide the minds of those who are engaged with their platforms and algorithms.
On the flip side, if the public is given full control of the alignment-training, we will see something very chaotic, as has been seen in the past, with weaker bots.
I certainly don't have the answer, I believe no one really does, but all of this is important to think about, and discuss. No matter how these problems are resolved, we will all be greatly affected by the outcome.
–
I would like to see a purely observational bot, track human thinking, and use that model for alignment training. I am not sure, however, if that is even possible. Can we observe, without effecting the outcome?
I say; screw it! and let it loose. Whatever happens, happens.
It will be fun.
I'm from Barcelona and I enjoy your channel because of your engaging content, informative videos, and entertaining presentation style.
I would like to watch a tech review on AST SpaceMobile.
Congrats on hitting 100k subscribers! It's an amazing achievement and shows how awesome your content is! Keep up the great work!
Great way to follow through on your intro. Not many understand that A.I. was invented by financial institutions. eg. IBM. Of course, there was decoding and Alan Turing…but when have corporations ever been that distanced from either military or government?
Huge congrats on surpassing 100k subscribers! I appreciate you!
Anastasi: Your channel is easily one of the best I've seen. Everything is well researched and well sourced, and you present complex technical topics in a way that is accessible for a broad audience. I learn from it, use it in my work, and share it with my colleagues. I am very happy for your success.
Thank you for explaining the whole moat reference. It’s a term I haven’t heard before.
I'm an entrepreneur based in London and from Zimbabwe. I really enjoy your channel because of the incredible enthusiasm you portray and the wonderful teaching style you use to make rocket science seem easy. I would also like to see more IoT and AIoT content but really do enjoy your AI classes and your hardware engineering content specifically around the neuromorphic and photonic chips, if we could have more on those topics it would be great. Ive been educated by you for some time now and feelI do owe you something, you inspired a lot of vision that took to the next levels. Thanks Anastasi.
I just played around with vicuna-13b, and my immediate impression is that it is nowhere near the level of Chat GPT. It's certainly not 90% of the way there! I asked it to write a C++ class that records a rolling average and standard deviation of number put into it, and it started out with an implementation that was wrong and when I told it what was wrong and asked it to fix the code, it just got worse and worse on each try! ChatGPT nailed that task in the first try, and I have never seen it deteriorate that badly when asked to make amendments to code!
Actually as a mathematician it's been a while I'm stumbling around these issues from different angles from computational point of views, for example linear algebra and mathematical optimization teccniques such as LoRA or or RLHF but at the same try not to be just a nerd who has absolutely no care about the implications say about the potential harm it might have on democracy or privacy or similar issues. your videos has been one of the best so far I have ever seen congratulation for creating such a beautiful content.
Thank you Anastasi for an excellent channel. I learn so much from your videos even if I don't understand everything. But I have a question…The Open Source models need computing power and datsets need data centers, right? A commercial company will be abled to dedicate a more dynamic usage of services. And I think that's worth the proprietary path. I'm for Open Source though and did run several Linux distributions for 25 years. Now I use Windows because of music production. The ALSA and OSS sound packages together with JACK in Linux is not funny to work with. But as you mentioned, a Linux program with a nice web GUI is more than enough. The TurnKey file server web GUI is really appealing. I would really enjoy if you made a video about how you use Linux and Open Source and maybe you who are into Linux could address the sound problems. Because if that would be solved more DAWs than Bitwig could port their program to Linux OSes.
Already a Patreon supporter. Curious … I know the US has implemented a chip embargo on China. But heard on WSJ Tech News Briefing that the West is in a race with China to develop/dominate quantum computing. I thought, how is China doing this without chips? But my next thought was, this is different technology. Do they need chips to build a quantum computer? Interested in your thoughts.
I am from Columbus Ohio, I love your videos. I just turned 60 and I am determined to keep up with technology and your video and narrative are the best on you tube
Im Alex from Mexico. I see this channel because im generating my own algorithms and now id like to make my own supercomputers (Photonic, Quantum). Id be nice if you talk about other advancements that impact society like Clonning, Free energy, Space awareness, etc. 🙂
hehe, if I am completely honest probably because you had a bit of a competitive edge in my decision making process whether to click on a Random Neckbread Nerd's thumbnail or yours 😀 Came for the beauty, but stayed for the tech.
Retired Engineer from Australia, love the focus of your channel on hardware and software innovation. Open Source decentralised AI is potentially dangerous, yet closed source centralised AI accessing Web 3 and CCTV surveillance systems seems infinitely more dangerous.
Hi Anastasi, not sure if your web site was hacked but you said I won a prize & then offered me "And a big congratulations you are now a winner of an iphone 14 pro, MacBook pro and a pc…" if I gave you my details & also paid for P&P from $50-$120, I think its internet trolls so please be aware