Reasons why your channels works: You're mature. No overhyping nonsense. The right amount of commentary and criticism, when appropriate. You don't take yourself too seriously and have fun with it. It shows. Videos are informative, but light. Well researched. Functions as an AI news-feed. We stay up to date with minimal effort. You're a likeable fellow with a solid work ethic. Keep it casual, and don't compromise the quality.
Although I love Matt and agree with most of what you said, his BIGGEST flaw is that he ALWAYS overhypes. Most of his video titles are nearly click bait because of the overhyping
@@teenat7044 I am so accustomed to that on RUclips, that on a relative scale... I know of one channel in particular, for instance, whose titles literally say nothing, purposefully, and you have no idea what the video will be about. And this creator does it with every... single... video. And it works. He's got far more views than most in that space. I blame the algorithm, and human nature.
I somewhat second that. Any AI news is bound to be overhyped because the whole AI is. But this channel is lighter on the hype compared to the other ones. I've completely stopped watching Mathew Berman and mostly stopped Wes Roth because they produce 20 minutes long videos every few days. They gave whole video to the China robot but here's it's literally just 30 seconds. Signal to noise is better here. This is just one video per week with all the news in it whereas the others produce 10 videos like that.
Personally I think that maybe OpenAI is starting to shift from `naming the LLM` to `naming the Chat Product`. And so I think that GPT2-chatbot is about shifting from GPT 1 Chatbot which has had increasing models and is what we have had so far, to now giving us GPT 2 Chatbot being a NEW Chatbot experience. This changes the focus from looking at the model to now instead looking at the user interface regardless of the LLM they use over time - Just a theory I have :)
The trend of selling items before they are actually fully developed is mostly because they are using the customer as their Beta testers to fully refine their product. Just like video games now-a-days, you purchase the Beta, play the crap out of it, and by the time the company releases the actual product, no one buys it because it's old news. To wait to release a fully developed AI product is impossible due to the insane amount of changes that are occurring in the space on a day-to-day basis.
But in some cases it really helps small Indie devs. There are some really good games which wouldn't have been produced without help. I've bought some in Alpha and Beta because I supported the devs, tested it and if it was playable, I kept playing or well, I just waited. In these cases, I think, it is pretty good. But those AI devices is just about "who is the first", nothing more.
If I may add a counterpoint... The takeaway from that Altman quote is not to hype up the imminent release of GPT-5. It's to justify making a handful of tweaks to 4.5 and beyond. He's selling the iterative approach as something people want. More speculatively, they know what GPT-5 looks like but it's mind-bogglingly energy intensive and it'll take the next year to make it efficient enough for widespread inference. They aren't courting the Middle East just because they have the cash to make chips. They're sitting on top of one of the world's most reliable source of energy.
I think they are still iterating on GPT-5's model architecture. Especially when development in that area is still hot and ongoing, with the Mamba(Bytes), 1bit paper, infini-attention and so on coming out. Oh and I bet it will be true multimodal with all modalities integrated together. Getting training data ready for that alone will also take time.
There should be philosophers and technical experts from universities (?) in it (not to say the ceo's aren't experts, but they can never be independent).
@@AlmerosMusicCode "technical experts from universities" asking for trouble. universities are a hotbed of brain rot neopotism due to left leaning hiring policy run wild
It seems like almost everything is being delivered halfway and not just paying full price, but a premium like recent electric vehicles with features missing and may or may not become available in the future. Great video! 😊🌎❤️🕺🏻🐶
13:52 This is Regulatory capture, Open Ai can be a small advisorr, but not run the ai government agencies, Regulatory Capture, which is a form of corruption that occurs when a regulatory agency prioritizes the interests of the industries it regulates over the public interest. This can happen when a political entity, regulator, or policymaker is co-opted to serve the interests of a minor constituency, such as an industry, profession, geographic area, or ideological group.
Hey! Thanks for the shoutout, Matt!! Agreed that Vidu isn't a "Sora" level-- and I almost wish they hadn't gone directly at them that way, since I think it's looking pretty good on its own. 1080p at 16 secs? That's awesome! But going toe to toe w/ Sora is just inviting a negative comparison. That said, check out the footage of the bear walking. Just below the jacket...Sora can't do THAT!
you think AI is intresting i use to work as a concept artist, i have no future, and it will also. remove 90% of the tech jobs. there is no use for coders, what are people going to work with? if you have 50% Unemployed socity will implode regardless how good ai you have. Civial war, is , civial unreset. is a given. we dont need artist , we dont need engineers. this is a bigger question. this will change things. and bad things are coming
@@peter486 We knew we were coming to a future of automation for decades, no matter what we were always going to wait until the last second to adapt, and adapt we must. Those of us in tech have largely been optimistic about it knowing this was coming. It's going to be a bumpy road, but if we don't dig our heels in and work to solve the kinks, we'll be better off as we've always done.
@@peter486 You sound like the farmers from 1920s complaining about tractors. Or the painters from the 1800s complaining about photography. Or the scribes from the 1400s complaining about the printing press. Or the photographers in the 90s complaining about photoshop & MS Paint. Time moves all things.
Re: Promising features when it ships, paying full price, getting unfinished product and being promised again that the features will exist in the future. The gaming industry has been doing this for years and it's disgusting. No Man's Sky? Diablo IV?
The only reason GPT4 is the dumbest model, is because they DUMBED IT down. The intial GPT 4 was amazing, and then they added a Million and 1 filters to dumb it down.
They added filters to make it less racist, sexist, etc - a side effect was that it somewhat dumbed down. Except that's not my experience at all - it seems to be about as smart. It just doesn't do as much as before (especially illegal stuff, or dark-grey at the very list.)
I think it's "learning" from users prompts. Not just modifying how it reacts to a particular user, but slightly to all users. The more popular the model, the more it is dumbed down by its user-base-who as a whole is less intelligent than current models.
@@NostraDavid2 those filters were already in place upon release, I've experienced the degradation over time. The initial GPT-4 was way better than whatever we have now.
@@RubelliteFae I know OpenAI, Google, Anthropic, etc. collect user chat data to add to training, but I don't think the model self-trains or live-trains from user chats, like what Google's ancient Tay chat AI did before they took it down. It does psuedo-train within a chat thread based on the contents of the chats, but I don't believe the general model itself is trained by anything but OpenAI engineers, and only seldom it seems since the training cutoff date barely changes (unless they can add data without moving the cutoff). It would be cool if they had a feature where you can fork a GPT model and semi-train it from your chats, to use in new chats. Not memories but closer to how open source model get trained by users.
Your videos are so good Matt. You are super likeable and AI is so much fun when learning about it from you, I'm glad I've been following you all the time now. Cheers from Bali
17:06 Something similar happened to games during the last decade, unfinished products being sold at full price with the promise of greatness in the future, it rarely pays out, and even when it does, it still sucks to be sold shit and have it be good only after you already lost interest in it.
@@RubelliteFaeits still maybe a net positive because of the fact that more access to start up/kick start money.. the more probability to nurture gooe companies ane innovative products or processes. Obvious this also produces more useless bullshit companies… thats just the fault of whoever is giving the money out to those useless companies… basically… people who have jobs for which they are grossly under qualified for….. you have to be able to know exactly what the product and/or process is and the market its entering.
well done for packing so much info in each your vids. appreciate the effort. so much happening in AI daily its great to have a place for a quick catchup
Cmon Matt, the safety board is absolutely going to game the industry to benefit themselves, thats how they work get in at the ground floor and make sure they always have their hands on the ocntrols... "little guys" will not have a chance
100%. These guys will absolutely try to destroy the open source, the hugging faces, the home grown AI so people are forced to use their biased/altered platforms.
Indeed. We have to be wary of legislative capture. When companies themselves start asking for regulations, you know it's only to benefit themselves and not let smaller actors get into the industry.
For sure when I have a question about something and expecting a serious answer I can work with, I don’t even try Google, I go directly to ask Chat GPT. I’m never disappointed !
Sam Altman saying it's the dumbest model is like every AI reviewer saying this is the worst it's going to be. It means that it will only get better. It's a common phrase used in reviewing AI.
I'd like to see a collection of court rulings against anti-AI lawsuits. It would be great to quote such things when "AI is theft!" pops up in a discussion.
With the paintball thing, so what happens when first responders go to your house for an emergency? Do they get pelted while trying to save you? Isn't it illegal to trap your house for this very reason?
In the RUclips video "AI News: Mysterious AI Shocked Everyone (Then Vanished)," the host discusses the sudden appearance of a new AI language model named "gpt2 chatbot" on a website called chatbot Arena, which outperformed other state-of-the-art models and left the origin of the model a mystery. OpenAI's Sam Altman confirmed it was not a new version of GPT, but the model caused a stir in the AI community due to its impressive performance. OpenAI also announced the rollout of the memory feature in ChatGPT and rumors of a new search engine. The video also touches upon ongoing legal disputes between OpenAI and newspapers over copyright infringement claims, the potential launch of OpenAI's search engine, and Apple's reported talks with OpenAI and Google about incorporating AI technology into future smartphones. Additionally, the video discusses recent advancements in AI, including GitHub's GitHub Co-pilot workspace, and criticisms of new AI products like the Rabbit R1 for being incomplete upon release. The speaker also shares their excitement for new AI tools and the legal implications of AI-generated content, and announces a new collaborative RUclips channel called AI Community
The latest version of GPT from Open Ai seems to have gone backwards, lagging and when using for development purposes it seems to take 2 steps forward and 1 step backwards
I think GPT2 is going to be GPT-4 but MUCH smaller and I think the model will be released to the public for personal use. Just like all their competitors and just like Elon Musk pressured them to do. I’ve not seen that factor brought up anywhere in conversations about this.
Agreed. My prediction too, I think it'll be trained on optimised output from GPT4 (I've been doing this with Agents and it's surprisingly easy, and incredibly effective), and similarly with Claude Opus and Anthropic API - that works even better IMO.
I knew AI was growing way to fast, but what I saw on the News flipped me out. AI was flying a Fighter Jet against a 20+ year professional Fighter pilot and had a mock-up fight where the AI pilot out maneuver the 20+ year professional Fighter pilot. It's getting really serious now, but that's all. I also saw in the News that Sam's Club is now using AI to scan customers receipts, so there goes more jobs to AI. Don't get me wrong, I use AI everyday, but I think AI is getting to big for it's Microchips!
This is the inevitable outcome of AI; it will be capable of doing virtually anything, outmoding the necessity of human work. The question is, will "they" allow society to transition to a new way of life, or will we all just end up in the streets, poor and destitute...?
@@Aedaeum “they” are slow to regulate new things. This might be different if it affects jobs in a meaningful way, but either way I’m planning on a 20-30 year painful transition, during which it will pay to have money and assets, regardless of where we are with UBI/etc.
I'm on the MidJourney $30 plan and have had access to their site for some time now. I can't get it to do face swap though, and it's just equally as censored, so I have no reason to use their web site. I just use a private room in the discord, it's much easier with more features.
They needed to do something like this, to keep the story going. Hence, it won't be much longer before people realize how limited ai is really going to be.
I would love it if on one of your podcasts you did a deep dive into how the post phone AI world might look. A lot of things I try to imagine about it that are still kind of fuzzy.
Hi Matt, I'm wondering if perhaps it's just a re-trained version of GPT-4 that OpenAI decided to put out there to test it out, incognito. As far as Apple, it makes sense for them to not only shop around for AI capabilities, but for terms and cost, as well.
Judges should not so easily shut down these lawsuits about using copyrighted material for AI because it will affect millions, if not billions of people’s jobs, and the first part of that will be very harmfull and painfull, before it gets better.
Great AI news channel. But I feel the topics are generally about LLM and Generative AI. Would be great if you can expand to other industries like breakthroughs in medical AI.
18:33 no but he did say that it doesn't only apply to ai tech and that it's a trend. He even gave Tesla as an example. And at least le apple vision pro is usable and multifunctional at many level. It's not like a product that entirely depend on a model that is not meant to support that type of function
My thougt on gpt2 is that they think of the old gpt-2, gpt-3.5, gpt-4 as subversions of the model thats called chat gpt, and as sam said during an interview the next model will be a whole new level so it would make sense that they think of it as the gpt2. Speaking in version numbers it would be the step from 1.2.0, 1.3.5, 1.4.0 to 2.0.0
Every time I made an app for Apple and I had a coming soon feature down in, and Apple found it, in the review they would deny my app. So two-faced. Cracks on the surface, that's what we're seeing.
AI, and AGI for that matter, really have to be based on just video/audio with no text. That's what we want AI to do. But we can label everything to begin with and training with that works, that is, makes AI. But we have to remove the text labeling and get AI/ML to just operate on observing everything with video/audio and learning everything that way. And we know there's tons of data and more to be gathered. That's the limit they rarely mention making datasets. But we've got whole cities with cameras everywhere now all over the world.
Any idea why the image generation tool that was built into MS Bing / Co-pilot is gone now and when or if it will be back? I used it as recently as about 3 weeks ago, but its been down for at least 1 week.
General question for the audience, as I know that people who watch this channel are probably smart. When can we expect to see visual avatars that we have in an app that can teach us, order stuff, and generally be there 24/7?
There are probably apps on the store like this already. If you mean good ones, not in the near future, unfortunately. To have a model of lets say chatgpt level of capabilities, which is still unreliable + plus the rendering of the avatar, voice... you need powerful chips not the punny chips you have in your phone. Even worse, all progress on the microchip industry comes from on company, Nvidia and they concentrate to make big and expensive powerful chips for data centers, not improved computer / phone chips that can run better models locally. The other option is api calls, but you see, running the app this way will cost you a lot... I'll say we are a long way off from having virtual girlfriends on our phone, at least capable ones...
17:20 "releasing unfinished apps for full price with the HOPE it'll get the extra features later." yeah, Yong Yea and The Jimquisition have been ranting about this, in terms of video games, for a long time, since it's garbo.
The bit that sold me on the R1 was the 1 year perplexity pro sub. In Australia, it worked out to cost me about $20aud extra to get an R1 compared to 1 year of Perplexity Pro subscription fees, so I didn't really care if the device was unfinished. $20 for a cool toy is alright by me 😅. That's like 3 cups of coffee here. And Perplexity Pro is phenomenal, as you've said countless times on your channel. I've converted dozens of non ai users to AI just with the free version, then showing them what Pro can do - access Opus etc and most significant up. My dad is 78, loves learning still but technologically incompetent and even he uses it non stop now.
@creativeconstitution anything to do with research, quick information that is verified, understanding my searches much better - saves a tonne of time compared to Google. I also have integrated it into a Make workflow that takes a voice idea I have (Slack input, Whisper ARS), uses a GPT Agent to create a prompt, sends to Perplexity, outputs to GPT / Claude Agent to create whatever format I want (Google doc, LinkedIn post etc), and many many other examples.
@tomgreen8246 Oh nice! We have make workflows with the OpenAi module - it’s pretty awesome! We are just using ChatGPT api credits though. Is the Perplexity app the same but you get to choose different models?
Reasons why your channels works:
You're mature. No overhyping nonsense. The right amount of commentary and criticism, when appropriate.
You don't take yourself too seriously and have fun with it. It shows. Videos are informative, but light.
Well researched. Functions as an AI news-feed. We stay up to date with minimal effort.
You're a likeable fellow with a solid work ethic. Keep it casual, and don't compromise the quality.
No over hyping? I disagree
Although I love Matt and agree with most of what you said, his BIGGEST flaw is that he ALWAYS overhypes. Most of his video titles are nearly click bait because of the overhyping
@@teenat7044 I am so accustomed to that on RUclips, that on a relative scale...
I know of one channel in particular, for instance, whose titles literally say nothing, purposefully, and you have no idea what the video will be about. And this creator does it with every... single... video. And it works. He's got far more views than most in that space.
I blame the algorithm, and human nature.
I somewhat second that. Any AI news is bound to be overhyped because the whole AI is. But this channel is lighter on the hype compared to the other ones. I've completely stopped watching Mathew Berman and mostly stopped Wes Roth because they produce 20 minutes long videos every few days. They gave whole video to the China robot but here's it's literally just 30 seconds. Signal to noise is better here. This is just one video per week with all the news in it whereas the others produce 10 videos like that.
@@jonm6834 Scotty Kilmer is so over the top clickbait that I just ignore the titles
the degree of conflict of interest on the Safety and Security Board couldn't be sadder...
Co-pilot is a little more restricted then GPT4. So would Open AI search be less restricted the Bing?😅
Personally I think that maybe OpenAI is starting to shift from `naming the LLM` to `naming the Chat Product`. And so I think that GPT2-chatbot is about shifting from GPT 1 Chatbot which has had increasing models and is what we have had so far, to now giving us GPT 2 Chatbot being a NEW Chatbot experience. This changes the focus from looking at the model to now instead looking at the user interface regardless of the LLM they use over time - Just a theory I have :)
That sounds right to me too!
The trend of selling items before they are actually fully developed is mostly because they are using the customer as their Beta testers to fully refine their product. Just like video games now-a-days, you purchase the Beta, play the crap out of it, and by the time the company releases the actual product, no one buys it because it's old news. To wait to release a fully developed AI product is impossible due to the insane amount of changes that are occurring in the space on a day-to-day basis.
But in some cases it really helps small Indie devs. There are some really good games which wouldn't have been produced without help. I've bought some in Alpha and Beta because I supported the devs, tested it and if it was playable, I kept playing or well, I just waited. In these cases, I think, it is pretty good. But those AI devices is just about "who is the first", nothing more.
If I may add a counterpoint... The takeaway from that Altman quote is not to hype up the imminent release of GPT-5. It's to justify making a handful of tweaks to 4.5 and beyond. He's selling the iterative approach as something people want.
More speculatively, they know what GPT-5 looks like but it's mind-bogglingly energy intensive and it'll take the next year to make it efficient enough for widespread inference. They aren't courting the Middle East just because they have the cash to make chips. They're sitting on top of one of the world's most reliable source of energy.
Great point!
I think they are still iterating on GPT-5's model architecture. Especially when development in that area is still hot and ongoing, with the Mamba(Bytes), 1bit paper, infini-attention and so on coming out. Oh and I bet it will be true multimodal with all modalities integrated together. Getting training data ready for that alone will also take time.
Energy? You mean oil or solar or nuclear energy?
This. I think this is an accurate assessment.
The list of names on the Artificial Intelligence Safety and Security Board is basically a worst case scenario.
There should be philosophers and technical experts from universities (?) in it (not to say the ceo's aren't experts, but they can never be independent).
@@AlmerosMusicCode "technical experts from universities" asking for trouble. universities are a hotbed of brain rot neopotism due to left leaning hiring policy run wild
It seems like almost everything is being delivered halfway and not just paying full price, but a premium like recent electric vehicles with features missing and may or may not become available in the future. Great video! 😊🌎❤️🕺🏻🐶
13:52 This is Regulatory capture, Open Ai can be a small advisorr, but not run the ai government agencies,
Regulatory Capture, which is a form of corruption that occurs when a regulatory agency prioritizes the interests of the industries it regulates over the public interest.
This can happen when a political entity, regulator, or policymaker is co-opted to serve the interests of a minor constituency, such as an industry, profession, geographic area, or ideological group.
Hey! Thanks for the shoutout, Matt!! Agreed that Vidu isn't a "Sora" level-- and I almost wish they hadn't gone directly at them that way, since I think it's looking pretty good on its own. 1080p at 16 secs? That's awesome! But going toe to toe w/ Sora is just inviting a negative comparison. That said, check out the footage of the bear walking. Just below the jacket...Sora can't do THAT!
TM, in the wild!
you think AI is intresting i use to work as a concept artist, i have no future, and it will also. remove 90% of the tech jobs. there is no use for coders, what are people going to work with? if you have 50% Unemployed socity will implode regardless how good ai you have. Civial war, is , civial unreset. is a given. we dont need artist , we dont need engineers. this is a bigger question. this will change things. and bad things are coming
@@peter486 We knew we were coming to a future of automation for decades, no matter what we were always going to wait until the last second to adapt, and adapt we must. Those of us in tech have largely been optimistic about it knowing this was coming. It's going to be a bumpy road, but if we don't dig our heels in and work to solve the kinks, we'll be better off as we've always done.
@@peter486 You sound like the farmers from 1920s complaining about tractors.
Or the painters from the 1800s complaining about photography.
Or the scribes from the 1400s complaining about the printing press.
Or the photographers in the 90s complaining about photoshop & MS Paint.
Time moves all things.
yup
Re: Promising features when it ships, paying full price, getting unfinished product and being promised again that the features will exist in the future. The gaming industry has been doing this for years and it's disgusting. No Man's Sky? Diablo IV?
The only reason GPT4 is the dumbest model, is because they DUMBED IT down. The intial GPT 4 was amazing, and then they added a Million and 1 filters to dumb it down.
They added filters to make it less racist, sexist, etc - a side effect was that it somewhat dumbed down.
Except that's not my experience at all - it seems to be about as smart. It just doesn't do as much as before (especially illegal stuff, or dark-grey at the very list.)
I think it's "learning" from users prompts. Not just modifying how it reacts to a particular user, but slightly to all users.
The more popular the model, the more it is dumbed down by its user-base-who as a whole is less intelligent than current models.
@@NostraDavid2 those filters were already in place upon release, I've experienced the degradation over time. The initial GPT-4 was way better than whatever we have now.
@@RubelliteFae I know OpenAI, Google, Anthropic, etc. collect user chat data to add to training, but I don't think the model self-trains or live-trains from user chats, like what Google's ancient Tay chat AI did before they took it down.
It does psuedo-train within a chat thread based on the contents of the chats, but I don't believe the general model itself is trained by anything but OpenAI engineers, and only seldom it seems since the training cutoff date barely changes (unless they can add data without moving the cutoff).
It would be cool if they had a feature where you can fork a GPT model and semi-train it from your chats, to use in new chats. Not memories but closer to how open source model get trained by users.
It’s not dumber, but I know what you mean- it’s lacking in novelty - it’s responses got more generic, which is ultimately unsatisfactory on all counts
Your videos are so good Matt. You are super likeable and AI is so much fun when learning about it from you, I'm glad I've been following you all the time now. Cheers from Bali
Always appreciate the updates!
Where's the link to Guide?
Amazing news roundup, many thanks. A lot of work.
If you wait for something to be perfect before launching it, you will never release it.
nice report Matt..per usual. Thanks!!
Guidde link is missing from the description. You said it would be at the top. I actually was looking for it to give you credit for me checking it out.
Yeah I was also looking for it. WTH Matt, lol
Matt, I loved this video and this type of style and length. Thank you for doing you.
What he really said was don’t expect a major leap in advancement from release to release, GPT-7 will perform the way GPT-5 should
Awesome video, I love nerding out to the cool AI stuff too! You're my absolute #1 youtuber for AI news.
17:06 Something similar happened to games during the last decade, unfinished products being sold at full price with the promise of greatness in the future, it rarely pays out, and even when it does, it still sucks to be sold shit and have it be good only after you already lost interest in it.
We've been kickstarting products without getting equity in the companies.
Not investors. Not co-op members. Just consumers who will pay for a promise.
@@RubelliteFaeits still maybe a net positive because of the fact that more access to start up/kick start money.. the more probability to nurture gooe companies ane innovative products or processes. Obvious this also produces more useless bullshit companies… thats just the fault of whoever is giving the money out to those useless companies… basically… people who have jobs for which they are grossly under qualified for….. you have to be able to know exactly what the product and/or process is and the market its entering.
We’re all beta testers
Another great Weekly Update! Thank you, Matt - keep up the great work!!
always top notch info my friend . well done once again
well done for packing so much info in each your vids. appreciate the effort. so much happening in AI daily its great to have a place for a quick catchup
Great video Matt.
Thanks for your hard work
Matt, the "sweaty guy" in the video was just one half of a couple who were in their sauna watching their security system.
Woah you put a tonnnnnnn of work into this video!! Nice job!
Good show as always. Thank you 😊
Great video Matt 🤙🏼 keep it up.
Cmon Matt, the safety board is absolutely going to game the industry to benefit themselves, thats how they work get in at the ground floor and make sure they always have their hands on the ocntrols... "little guys" will not have a chance
100%. These guys will absolutely try to destroy the open source, the hugging faces, the home grown AI so people are forced to use their biased/altered platforms.
Indeed. We have to be wary of legislative capture. When companies themselves start asking for regulations, you know it's only to benefit themselves and not let smaller actors get into the industry.
he knows
Thanksfor the energetic overview again. Lots of info I found interesting. Thank you
For sure when I have a question about something and expecting a serious answer I can work with, I don’t even try Google, I go directly to ask Chat GPT. I’m never disappointed !
Thanks for the vid
Interesting news as always!
Yeah, it is my first time tuning in , I wonder why I haven’t seen you before😂
I really do love your videos. Also, your sense of humor. Keep going Matt!
The most insane news is Apple, acknowledging the existence of other companies.
Matt thank you for everything. 🎉
You deserve more thank yous.
Sam Altman saying it's the dumbest model is like every AI reviewer saying this is the worst it's going to be. It means that it will only get better. It's a common phrase used in reviewing AI.
I'd like to see a collection of court rulings against anti-AI lawsuits. It would be great to quote such things when "AI is theft!" pops up in a discussion.
Altman is the best salesman I've ever seen. And a good actor too, he is able to say flagrant lies with a straight face.
With the paintball thing, so what happens when first responders go to your house for an emergency? Do they get pelted while trying to save you? Isn't it illegal to trap your house for this very reason?
Good rant at 7:32. I agree, a website need more than a little narrow window for its content, regardless of how much the advertiser is paying them.
Thank you so much
I wait all week for these videos! Matt Wolfe, you're the BEST AI news communicator in the entire world!
I love how Sam refuses to capitalise letters
In the RUclips video "AI News: Mysterious AI Shocked Everyone (Then Vanished)," the host discusses the sudden appearance of a new AI language model named "gpt2 chatbot" on a website called chatbot Arena, which outperformed other state-of-the-art models and left the origin of the model a mystery. OpenAI's Sam Altman confirmed it was not a new version of GPT, but the model caused a stir in the AI community due to its impressive performance. OpenAI also announced the rollout of the memory feature in ChatGPT and rumors of a new search engine. The video also touches upon ongoing legal disputes between OpenAI and newspapers over copyright infringement claims, the potential launch of OpenAI's search engine, and Apple's reported talks with OpenAI and Google about incorporating AI technology into future smartphones. Additionally, the video discusses recent advancements in AI, including GitHub's GitHub Co-pilot workspace, and criticisms of new AI products like the Rabbit R1 for being incomplete upon release. The speaker also shares their excitement for new AI tools and the legal implications of AI-generated content, and announces a new collaborative RUclips channel called AI Community
I think the sweaty guy in the paintball video was supposed to be in a sauna LOL!! 😆
Please link stuff that you promise you will in the description..., otherwise thanks for the most solid news Matt
1:01 - Maybe some crazy advanced automatic AI model made it and uploaded it as well to the internet?
The latest version of GPT from Open Ai seems to have gone backwards, lagging and when using for development purposes it seems to take 2 steps forward and 1 step backwards
I think GPT2 is going to be GPT-4 but MUCH smaller and I think the model will be released to the public for personal use. Just like all their competitors and just like Elon Musk pressured them to do. I’ve not seen that factor brought up anywhere in conversations about this.
That’s a good take. I could definitely see it
There were GPT2 answers that had code that said "Made by OpenAI"
i belive is from the chinise
they try to conquer the world
Agreed. My prediction too, I think it'll be trained on optimised output from GPT4 (I've been doing this with Agents and it's surprisingly easy, and incredibly effective), and similarly with Claude Opus and Anthropic API - that works even better IMO.
Interesting take
I knew AI was growing way to fast, but what I saw on the News flipped me out. AI was flying a Fighter Jet against a 20+ year professional Fighter pilot and had a mock-up fight where the AI pilot out maneuver the 20+ year professional Fighter pilot. It's getting really serious now, but that's all. I also saw in the News that Sam's Club is now using AI to scan customers receipts, so there goes more jobs to AI. Don't get me wrong, I use AI everyday, but I think AI is getting to big for it's Microchips!
It also doesn’t care what we think. But this is all gonna happen regardless.
@@TheExodusLost True, but the people that's making these should care and should think a head.
This is the inevitable outcome of AI; it will be capable of doing virtually anything, outmoding the necessity of human work. The question is, will "they" allow society to transition to a new way of life, or will we all just end up in the streets, poor and destitute...?
@@Aedaeum “they” are slow to regulate new things. This might be different if it affects jobs in a meaningful way, but either way I’m planning on a 20-30 year painful transition, during which it will pay to have money and assets, regardless of where we are with UBI/etc.
@@Aedaeum that's the point I've been saying for over 3 months now.
Wasn't the apple vision pro released as "developer kit"? It's expected to be missing things.
I'm on the MidJourney $30 plan and have had access to their site for some time now. I can't get it to do face swap though, and it's just equally as censored, so I have no reason to use their web site. I just use a private room in the discord, it's much easier with more features.
Matt, I am soooooooooooo with you on releasing half-baked tech products! 🔥🔥🔥🙏🔥🔥🔥
Iv managed to get GPT3.5 and Gemini to make a snake game in 1 shot before but still cool stuff!
Im pretty sure Marques did say what you mentioned about the apple vision. Watch his review again
I think Runway will make some great video models in the future, it might creating a world model focused on logic now.
Solid Update
They needed to do something like this, to keep the story going. Hence, it won't be much longer before people realize how limited ai is really going to be.
23:33 AI Paintball teargas shooter... what could possibly go wrong!! And how long before someone replaces the paint with live bullets.
Guide creates a step by step tutorial.
3:21 that's a feature. Now there's AI that can be mindful of what you post. Now who's not old enough to use this gpt
Goto channel for Ai news and reviews! 👍
There are more links and great material posted underneath Matt's videos than some AI essays in Harvard.
I would love it if on one of your podcasts you did a deep dive into how the post phone AI world might look. A lot of things I try to imagine about it that are still kind of fuzzy.
Hi Matt, I'm wondering if perhaps it's just a re-trained version of GPT-4 that OpenAI decided to put out there to test it out, incognito.
As far as Apple, it makes sense for them to not only shop around for AI capabilities, but for terms and cost, as well.
10/10 thumbnail design, Matt
Judges should not so easily shut down these lawsuits about using copyrighted material for AI because it will affect millions, if not billions of people’s jobs, and the first part of that will be very harmfull and painfull, before it gets better.
The R1 is a children's toy.
Great AI news channel. But I feel the topics are generally about LLM and Generative AI. Would be great if you can expand to other industries like breakthroughs in medical AI.
18:33 no but he did say that it doesn't only apply to ai tech and that it's a trend. He even gave Tesla as an example. And at least le apple vision pro is usable and multifunctional at many level. It's not like a product that entirely depend on a model that is not meant to support that type of function
My thougt on gpt2 is that they think of the old gpt-2, gpt-3.5, gpt-4 as subversions of the model thats called chat gpt, and as sam said during an interview the next model will be a whole new level so it would make sense that they think of it as the gpt2. Speaking in version numbers it would be the step from 1.2.0, 1.3.5, 1.4.0 to 2.0.0
good idea a gpt search engine
People in Europe cant use Claude, cant use memory function in GPT... wow... people in europe can't use quite a lot, like aliens, on another planet
All chatbots from Google are also unavailable, mainly is the European Union fault, they rushed in to regulate it and developers got scared.
I really enjoy your news reporting. Is it alright if I share your work?
Yes
17:39 I disagree, he was fairly critical of the Apple Vision Pro. His review praise some aspects but also outlined many shortcomings.
Finally we got there. "Shocking"⚡⚡⚡⚡⚡⚡⚡⚡⚡
"There's some funkiness with the walking"
Every time I made an app for Apple and I had a coming soon feature down in, and Apple found it, in the review they would deny my app. So two-faced. Cracks on the surface, that's what we're seeing.
23:55 Matt you should know what a Sauna is haha
chatbot GPT3 will come soon later !!!!!
540k subscribers? This channel grew!
I know that 600k mark is where so many get stuck. Soon will jump to 1 million
AI, and AGI for that matter, really have to be based on just video/audio with no text. That's what we want AI to do. But we can label everything to begin with and training with that works, that is, makes AI. But we have to remove the text labeling and get AI/ML to just operate on observing everything with video/audio and learning everything that way. And we know there's tons of data and more to be gathered. That's the limit they rarely mention making datasets. But we've got whole cities with cameras everywhere now all over the world.
Ok I admit it, I created the GPT2 ai you speak of ...I am also the one who created Bitcoin, I am Satoshi
Any idea why the image generation tool that was built into MS Bing / Co-pilot is gone now and when or if it will be back? I used it as recently as about 3 weeks ago, but its been down for at least 1 week.
General question for the audience, as I know that people who watch this channel are probably smart. When can we expect to see visual avatars that we have in an app that can teach us, order stuff, and generally be there 24/7?
There are probably apps on the store like this already. If you mean good ones, not in the near future, unfortunately. To have a model of lets say chatgpt level of capabilities, which is still unreliable + plus the rendering of the avatar, voice... you need powerful chips not the punny chips you have in your phone. Even worse, all progress on the microchip industry comes from on company, Nvidia and they concentrate to make big and expensive powerful chips for data centers, not improved computer / phone chips that can run better models locally. The other option is api calls, but you see, running the app this way will cost you a lot... I'll say we are a long way off from having virtual girlfriends on our phone, at least capable ones...
@@cristianandrei5462 thanks for the response
17:20 "releasing unfinished apps for full price with the HOPE it'll get the extra features later."
yeah, Yong Yea and The Jimquisition have been ranting about this, in terms of video games, for a long time, since it's garbo.
Seriously? Product Promises?
What ever happened to the FTC?
Bait and Switch Law? (Yeah... FTC who?)
"Snake Game By OpenAI" 😂
The bit that sold me on the R1 was the 1 year perplexity pro sub. In Australia, it worked out to cost me about $20aud extra to get an R1 compared to 1 year of Perplexity Pro subscription fees, so I didn't really care if the device was unfinished. $20 for a cool toy is alright by me 😅. That's like 3 cups of coffee here.
And Perplexity Pro is phenomenal, as you've said countless times on your channel. I've converted dozens of non ai users to AI just with the free version, then showing them what Pro can do - access Opus etc and most significant up. My dad is 78, loves learning still but technologically incompetent and even he uses it non stop now.
What are you using it for mainly?
@creativeconstitution anything to do with research, quick information that is verified, understanding my searches much better - saves a tonne of time compared to Google. I also have integrated it into a Make workflow that takes a voice idea I have (Slack input, Whisper ARS), uses a GPT Agent to create a prompt, sends to Perplexity, outputs to GPT / Claude Agent to create whatever format I want (Google doc, LinkedIn post etc), and many many other examples.
@tomgreen8246 Oh nice! We have make workflows with the OpenAi module - it’s pretty awesome! We are just using ChatGPT api credits though. Is the Perplexity app the same but you get to choose different models?
Hi Matt, i was wondering is any AI solution there i can upload my Power point and get better version from uploaded PP? THNX A LOT
i think gpt2-chatbt is made by other ai models and released it, thas why no human is claiming it cause they themself dont know
what are you using to generate your thumbnails?
Could somebody let me know which RUclips transcript AI extension he uses at around 6:40?
9:07 the bear is a bit to happy
I really want that Paintball security camera 😂 But probably it would start to malfunction and shoot at everything in the end.