Microsoft New AI Search, New Mistral Large 2 Model, Amazon Prime is AI Now & More!

Microsoft has launched Bing Generative AI Search for better answers, and Amazon revamped Prime Video with AI upgrades. Plus Mistral AI's new model is creating a lot of hype and tech stocks are taking a hit, especially in AI. Stay tuned to the end of this video as we dive into all these recent AI developments. Alright, so Microsoft has rolled out a new feature in Bing called Bing Generative Search. What's cool about it is that it combines Bing's search results with the power of large and small language models, or LLMs and SLMs. This means it can understand your search query better and pull information from millions of sources on the web to give you instant summaries and relevant info. For example, if you type what is a spaghetti Western? Bing will give you a rundown of the film genre's history and origin and even highlight some of the most famous movies. It'll also show you links and sources where it got the information from. Pretty handy, right? Jordy Rybus, who's in charge of Copilot and Bing, announced this launch and called it a meaningful step forward in AI driven search. It's Microsoft's direct response to Google's AI Overviews, which launched in the US back in May. Google's feature aims to give users quick answers without making them sift through tons of articles, and Bing Generative Search does the same thing. Now let's talk about the hiccups Google faced with their AI overviews. They had some major fails, like suggesting mixing non toxic glue with cheese to keep it from sliding off a pizza and giving a recipe for mustard gas when someone asked how to clean a washing machine. Crazy, right? Other AI search engines like Arc Search and genspark also had their own issues. Arc Search confidently told a reporter that cut off toes would grow back, and genspark listed weapons to use for harmful activities. Yikes. Microsoft is trying to avoid these kinds of problems. Another big issue is the lack of attribution. Some AI engines like Perplexity, got into trouble for plagiarizing articles without giving credit. Microsoft wants to make sure they don't fall into that trap. There's also concern about how AI search could impact website traffic. Google's AI overviews might hurt around 25% of publishers traffic because people get the info they need from the summary and don't click through to the actual site. This is a huge deal for content creators and smaller websites, but Rebas is confident that Bing Generative Search won't have this problem. He says they're monitoring how it affects traffic and claims that it's maintaining the number of clicks to websites. However, he didn't provide any hard data, just some early data that's not published yet. So we'll have to keep an eye on that essentially. Bing Generative Search is here and it looks like they've learned from others mistakes. It'll be interesting to see how it compares to Google's AI overviews. What do you guys think? Drop your thoughts in the comments below. Alright now, Mistral AI has just revealed their latest large language model and it's called Mistral Large 2 or ML2 for short. So let's break down what makes this model a serious contender against the big names like OpenAI, Anthropic and Meta. Alright, here's the deal. Mistral AI's new ML2 model has 123 billion parameters, and it's pretty impressive. This announcement comes right after Meta launched their new release 405 billion parameter llama3 model, which has a massive context window of 128,000 tokens and supports eight languages. ML2 also features a 128,000 token context window and supports dozens of languages plus over 80 coding languages. This language diversity is a big deal, since many open models usually only support English. Mistral is continuing to push the boundaries here. Now let's talk benchmarks. If we believe Mistral's numbers, ML2 competes closely with OpenAI's GPT4, Anthropic's Claude 3.5 Sonnet and Meta's Llama 3.1405 B across various language coding and math tests. For example, in the massive multitask language understanding benchmark, ML2 scored 84%. To put that in perspective, Metaslama 3.1 scored 88.6%, GPT4 scored 88.7% and Claude 3.5 Sonnet scored 88.3%. Human experts typically score around 89.8% on this test. Here's the kicker. ML2 achieves this with way fewer resources. It's less than a third the size of Meta's biggest model and about 1/14 the size of GPT4. This makes it super efficient and attractive for commercial applications. The 123 billion parameter model needs about 246 GB of memory at full precision. While that's still too big for a single GPU, it it can run on a server with four to eight GPUs without needing to compress the model. Meta had to offer an 8 bit quantized version of their model to make it run on existing hardware, but ML2's smaller size means it can process queries faster. This higher throughput is because smaller models don't put as much pressure on the memory system, allowing them to produce responses quicker. The crazy thing is that if you have a powerful enough system, you can even try running Mistral Large 2 at home. Mistral has also put a lot of effort into making ML2 more reliable. They've fine tuned it to be cautious and avoid hallucinations where the model generates incorrect information. ML2 is designed to recognize when it doesn't know something and handle complex instructions better, especially in long conversations. Another cool feature is that ML2 is optimized for concise responses. While long answers might score higher in some tests, they're not always practical for business use, where shorter, more efficient responses are often better. Now about the model's availability. ML2 is open and available on popular repositories like Hugging Face, but there's a catch. The license is more restrictive than some of Mistral's previous models. For non commercial and research use it's fine, but if you want to use it commercially, you'll need a separate license. This isn't too surprising given the resources needed to develop these large models. Alright now. Amazon has just rolled out a major redesign for Prime Video and it's packed with some cool AI upgrades to make your streaming experience even better. So what's new? Basically, they've revamped their interface to simplify the way we watch and browse content. The main goal here is to make it easier for us to find movies, TV shows, live sports and even manage our subscriptions. One of the standout features of this update is the new navigation bar. Now you can browse content by type, like home movies, TV shows, sports and live tv, all from one easy to use menu. They've even added a dedicated tab for active prime video channels. Add on subscriptions, which is super handy. Amazon has also integrated some generative AI upgrades into the mix. These AI enhancements are aimed at personalizing your experience even more. For example, you'll see revised synopses for movies and shows that are tailored to your preferences. This means you'll get more relevant recommendations and easier access to the content you love. Another cool addition is the Hero Rotator, just below the navigation bar. This feature promotes content that you can rent or subscribe to, making it even easier to find the latest releases or hidden gems. Kaam Kashmiri, the vice president of design for Prime Video, shared that the update is all about presenting an easy to navigate entertainment hub. The idea is to help you discover new titles and enjoy your favorites without any hassle. Plus you can sign up for or switch add on subscriptions using a single login. The global rollout of this new UI started on July 23 and it should be available to all customers in the coming weeks. So if you haven't seen the new design yet, keep an eye out. It's coming your way soon. Alright now an interesting trend happening in the financial markets. Shares in both the US and Asia have taken a pretty sharp dive, especially those in the tech sector and even more so in AI related stocks. Just recently in New York, The S&P 500 dropped by 2.3% and the NASDAQ, which is heavy on tech stocks, fell by a whopping 3.6%. These are their biggest one day falls since 2022. The Dow Jones Industrial Average didn't escape either, it was down by 1.2%. The big players that got hit hard included Nvidia, Alphabet, Microsoft, Apple and Tesla. Over in Asia, the decline was led by Japan's Nikkei Index, which which fell by more than 3% this year. AI stocks have been a major driver of market gains, but now it looks like investors are reassessing things. Nvidia, which has been riding high on the AI wave, saw its shares drop by 6.8% and it's lost about 15% of its value over the last two weeks. They've got some financial results coming up at the end of August, so that's one to watch. Tesla also took a hit, dropping more than 12% after its latest latest financial results didn't quite meet investor expectations. Alphabet, the parent company of Google and YouTube, saw its stock price drop by 5%. They reported financial results that beat expectations, but they also mentioned high spending through the rest of 2024, mostly on AI development in Asia. Some of the big fallers included chip makers like Renesis Electronics and Tokyo Electron in Japan and South Korea's SK Hynix. According to Junbei Liu, portfolio manager at Tribeca Investment Partners, investors are starting to worry about all this spending on AI without seeing the revenue benefits. She doesn't think this signals the end of belief in AI, but suggests that investors will now be more focused on returns in the AI sector rather than just buying into the hype. There's also some investor wariness due to surprises in the US Presidential election campaign and questions around when the US Central bank will cut interest rates. Rates alright, if you like this video, please give it a thumbs up and subscribe for more AI updates. Share your thoughts in the comments below. Thanks for watching and I'll catch you in the next one.