Claude 3 — THE END OF CHATGPT IS COMING. (Shocking AI News)

the last week was hot in the world of AI claw 3 stable diffusion 3 new AI announcements and updates open AI acquisition and more so let's not waste any more time and get straight to business anthropic just announced claw three and they claim it to be more powerful than chbt what is it all about the official blog post gives a lot of information about the new model that I think we should check out first cloth 3 comes in three different versions depending on how smart and Powerful the model is cloth 3 haiku Sonet and Opus Haiku is the least powerful of the bunch while Opus is the true Powerhouse

developers say that this division into three models allows users to select the optimal balance of intelligence speed and cost for their application here is how the graph looks we can see that claw 3 Hau is many times cheaper to run than Sonet while delivering over half of the intelligence and judging by this graph the Opus is even more expensive than Sonet while also getting around 50 to 60% % higher performance but this graph doesn't really tell us stories so let's look at the comparison chart of all three claw models Chad gbt variants and two Gemini variants Opus outperforms its fears on most of the common evaluation benchmarks for AI systems including undergraduate level expert knowledge MML graduate level expert reasoning GP QA basic mathematics not saying that and more it exhibits near human levels of comprehension Flint on complex tasks leading the frontier of general intelligence all claw fre models show increased capabilities and Analysis and forecasting Nuance content creation cenation conversing in non-english languages like Spanish Japanese and French what I find interesting is that even the basic claw free Haiku outperforms both gbt 3. 5 and Gemini Pro sonit seems to be somewhere in the middle between gbt 3. 5 and gbt 4 Gemini Ultra and Gemini Pro and in many tasks it even manages to outperform the most advanced models from Google and open AI of course all the test conducted can be called cherry-picking and may not fully

paint a picture of how powerful the CLA 3 is another area where CLA 3 saw a significant jump was in terms of vision capabilities the chart shows that Claude 3 Opus in many ways is better than gbt 4 vision and in some cases is very similar to Gemini Ultra that was proven to be really good and image recognition also the third generation of Claud makes fewer refusals the Claud free models show more Nuance understanding of requests recognize real harm and refuse to answer harmless prompts much less often developers also claim to improve the accuracy of the responses when I tried cla a while back I cannot say that I noticed that many incorrect responses there were some but not very common I think they mostly refer to Enterprise cases where the type of requests is different and the amount of those requests is much higher than anything a regular person could ever reach Opus and Sonet are already available to use as API in 159 countries Hau will be available soon but that isn't the only interesting announcement that happened stability AI has announced stable diffusion 3 in early preview and the newly published research paper promises some seriously interesting stuff first stable diffusion 3 uses mmdi architecture which stands for multimodal diffusion Transformer this architecture is similar to the one used by Sora previous versions of stable diffusion and most current image generation AIS use a diffusion model large language models for text generation like gbt use a Transformer architecture diffusion models perform well at creating detail in small regions but are poor at generating the overall layout of an image but Transformers are good at layout but poor at Crea in detail being able to combine the two models lets us take advantage of the best bars of both architectures and now let's quickly look at example images from the analment focusing on limitations for example one of the persistent issues for all AI image generators is the text generation even though stable diffusion 3 is much better at this than it was before we can still

see imperfections that a person wouldn't allow to happen these imperfections come down to spacing of letters keeping them properly sized and so on for example the main image with the text stable diffusion 3 has a couple of these issues like letters A and B becoming one letter and variable distance between letters and the image of a Bus shows the size inconsistency on the dreamon sign all the letters have a different size also on the same image we can see that stable diffusion still isn't perfect when working with shadows the Shadow from the bus Falls in One Direction while the Shadow from the houses fall in a different direction however the developers in their research paper put a huge emphasis on performance and human feedback we have compared output images from stable Fusion 3 with various other open models as well as closed Source systems such as Dolly 3 moury V6 and idag V1 during these tests human evaluators were provided with example outputs from each model and as to select the best results based on how closely the model outputs follow the context of the prompt it was given how well text was rendered based on the prompt and which image is of a higher aesthetic quality we have found that stable diffusion 3 is equal to where outperforms current state of the art text to image generation systems in all of the above areas all of this sounds really promising but unfortunately stab Fusion 3 isn't publicly available yet but as soon as it lands we will make a video about it so you better subscribe to the channel another

announcement that just happened was co-pilot for finance from Microsoft this is a very nice niche product optimized for people working in finance with huge spreadsheets from what I see on the website it seems eily similar to those wallframe logins for Chad gbt and it will be integrated into the Excel Outlook and teams and should help with routine tasks and data analysis for example it can access and analyze data from various sources including Financial systems like Dynamics 365 and sap as well as emails and spreadsheets it also can automate repetitive tasks such as variance analysis Collections and invoice matching it can generate insights and recommendations and even suggest actions Based on data this all sounds really great especially for people working the corporate jobs in the financial area doing less manual work being more efficient making more informed decisions doesn't sound a little dystopian to you we made a video about jobs that may soon disappear and seems like Finance professionals should be added to that list now there's also been a very interesting Chad gbt arrival revealed mistol large was created by a French startup mistro and Promises to be very similar to gbt 4 according to the company Mr Large ranks second in the top llm Global list with metrics and accuracy only outperformed by Chad bt4 the developers highlight a couple key capabilities it's natively Flint in English French Spanish German and Italian with a Nuance understanding of grammar and cultural context it has a 32k token context window precisely follows the instructions and allows for custom moderation policies and it's even natively capable a function calling but let's look at the comparison chart and

see whether it is as good as the developers claim it to be and judging by the numbers mistal is pretty close to CAD bt4 in common sense reasoning and knowledge and when it comes to multilingual capacities mol outperforms mexal and Lama 2 by a noticeable margin even in terms of math and coding it is similar to gbt 4 and Gemini Ultra the developers are also releasing a small and more optimized model focusing on latency and cost and apparently this model is also a tbit faster than its competitors I think it is great that European startups are catching up to the big players in that industry we need more news like that even though this startup doesn't have that much money or capacities as open AI or Google its partnership with Microsoft is quite promising and could lead to interesting developments that are different from one is doing it would be really fun to make a huge comparison of the most powerful AI Chad Bots don't you think the world of music is also moving forward with Adobe previewing Cutting Edge generative AI tools for crafting and editing custom audio the new project titled music gen AI Control according to the post allows musicians to generate music from text Broms and even have full control over the audio with Project Music gen AI Control generative AI becomes your co-creator it helps people crack music for their projects whether they are

broadcasters or podcasters or anyone else who needs audio that's just the right mood tone and length the AI will then create an initial tune based on the users's prompt which they can also edit using text Adobe set users can for instance edit the generate music's intensity extend the length of the music clip or create a repeatable look I must say the video they made doesn't really show anything visually but it's fun to watch also the tool doesn't even have the interface yet it's all done through the console commands this is very different from commercially available AI music generators we've tested before and some of them even give good results for example music effects created by Google is a formidable opponent to adobe's Creation so it might hint at what the finished product will look like music effect generates tracks from a text prompt and lets users control each part of the prompt through drop- down menus I cannot say that this is some crazy flexibility but it's already enough to create some background music for a short video we've also tested beatoven and it was quite good this was an interesting week for sure and one of the releases was Amazon's rofus even though the initial unveil happened in February only now it has gotten to a point where people were actually able to

test it and use it and well it's definitely something here is what Amazon says about it Rufus is an expert shopping assistant trained on Amazon's product catalog and information from across the web to answer customer questions and shopping needs products and comparisons make recommendations based on this context and facilitate product Discovery in the same Amazon shopping experience customers use regularly but when the Washington Post did their testing they found that Rufus is not quite there to be called an expert Amazon's chatbot doesn't deliver on the promise of finding the best products for your needs or getting you started on a new hobby depending on how I phrase the question the Amazon bot several times offered basic suggestions that I could find in a how to article and didn't recommend specific products still when the Amazon bot responded to my questions I usually couldn't tell why the suggested products were considered the right ones for me or I didn't feel I could trust the Chad bot's recommendations to me this all is quite fun huge corporation with all the money in the world rolling out an underdeveloped and untested feature just for the sake of it we get to laugh and the users get frustrated it's definitely not something we want AI to do luckily ruus is still under development so hopefully in a few months it will eventually become somewhat decent but not every company is as bad at AI as Amazon Wix a website building platform has just released their own AI Builder and it's pretty cool creating a site is as easy as hitting the create with AI button and chatting with the bot about what you're looking for in your website

the questions are quite simple and take only a couple of seconds to answer basically it's like chatting with a human designer then you wait a bit and get a full-fledged website the promo materials in my opinion are bit exaggerating the real capabilities of the AI in past videos I've tested dozens of different AI website Builders and none of them were able to generate something like that usually the generate websites are quite crude and simple in their nature without complicated animations or anything like that Wix also can change themes for the website Each of which has a different color palette and overall look but if you want you can just type in more requirements for the website and will think of something even the layout can be changed with a simple prompt I like that very much this promotional video really paints a very pretty picture but if we take a look at the website we'll see that the real world results are a bit more modest still nice but modest don't expect anything Super Creative or Advanced this is pretty much the level of a junior designer but faster and cheaper and Wix is a very well-known and reputable platform for website building so I have a lot of trust in this Ai and believe that it will get better and better with every update there's also been a small update to the GPT store the new update allows users of 30 party gpts to rate them using a drop- down menu that lists a new Option send feedback open AI showed how it's done and consensus a pretty good research assistant we should make a video about it anyway to leave a review we need to click on the GPT name and select send feedback the menu will then appear allowing us to leave a one to five star rating and a text box to send an

optional private email to that gbts Creator there's also a checkbox option where we can leave our own address visible or hidden a little privacy trick I know this feature seems small and insignificant but you won't believe how many interest in gbts I've tested that needed some fine-tuning and adjustment and not all of them had an option to contact their creator in any way now when you encounter a custom gbt and have problems with it or have ideas on how to improve things you can send all your suggestions directly to the person who created this gbt also this rating will probably be used in the store indicating user satisfaction with certain gbts again this is a small update but an important one but open a felt quite talkative this week and released a blog post about its relationships with Elon Musk and Tesla and you won't believe what's inside I'm not going to read you the whole thing but instead I'll will give you a shorter version what happened between Elon Musk and open Ai and how Tesla was going to acquire open AI please know these events transpired a few years back and were published only now following elon's lawsuit so here's the story open AI is all about making sure artificial general intelligence AGI is good for everyone they realized early on that building AGI would be a massive money sync far beyond initial estimates Elon was one of the big names behind their funding suggesting they announc a

whopping $1 billion commitment to kick things off his pitched in less than 45 million himself with over 90 million coming from other supporters at first the team thought $100 million might be enough but Elon pushed for a bigger figure to not look weak next to Tech giants like Google by 2017 it became Crystal Clear reaching AGI wasn't just about smart ideas it needed serious computing power and consequently a ton of cash discussions about switching to a for-profit model came up because the nonprofit route wasn't going to cover the massive costs Elon had big Ideas here too like mering opening AI with Tesla and taking full control himself Elon suggested making open AI part of Tesla Hinton at a unique opportunity for open a to ride on Tesla's Financial coils he even mentioned the idea in an email that open AI could attach to Tesla as its Cash Cow which could give open AI the Financial Muscle it needed to compete with the likes of Google despite these discussions the idea of merging with Tesla were any deal that would give someone complete control over open AI didn't sit right and that's a good thing open AI held firm which eventually led to Elon pardon ways to start his own AI company and later soon open AI as it began making significant strides without him this is pretty much the whole story and to me it seems like something absolutely crazy right now but let's fantasize for a while do you think it would have been beneficial for AI if Elon Musk was in charge of the most influential AI company of the last decade personally I didn't think so Elon without a doubt is a very talented businessman sometimes his decisions aren't all that logical just think about him buying out Twitter for an absurd amount of money you can see it for yourself on Twitter even in the 4 you feed right no Twitter is uh we're roughly break even at this point depending on how things go if Cent Trends continue I think we could be profitable think of all those funny features in Teslas like fart and sounds and fart on demand is what I like to

have ai is a serious thing and you cannot approach it lightly it has to be guided by a person who understands not only the positive sides but also possible negatives of the technology and it surely has to be a person who directly participates in the development with the number of companies that Elon Musk owns at least a part of it I don't think that open AI would have been able to achieve such Heights as I said Elon is working on his own Ai and I honestly can't wait to test it out myself another AI related thing that happened this week was the release of the M3 MacBook Air by Apple Apple's press release for the new M3 MacBook Air includes a bold claim calling the machine the world's best consumer laptop for AI the claim comes ahead of WWDC 2024 during which apple is expected to announce a bunch of new AI features across all its platforms with the transition to Apple silicon every Mac is a great platform for AI leveraging this incredible AI performance Mac delivers intelligent features that enhance productivity and creativity I must say apple is playing safe here they're not directly saying anything about AI itself and mostly mention apps that already have ai in them but here's one interesting bit MacBook Air can also run optimized AI models including large language models and diffusion models for image generation locally with great performance if you're not used to reading Apple's press releases there's one interesting detail here Apple didn't mention artificial intelligence at all in the press release for the M3 MacBook Pro last October while Apple generally mentions things like machine learning and press releases this is a very rare mention of AI specifically and by the way we just made a great video where we did a deep dive into all the leaks and rumors regarding Apple's upcoming AI features because those are going to be fire the world of AI is moving fast and the flow of news just keeps coming we will keep you updated so be sure to subscribe to the channel thanks for watching and see you soon