WEBVTT 00:00.449 --> 00:07.400 It's never a dull moment in the world of A I. There have been so many cool updates that happened this past week including an 00:07.400 --> 00:11.840 A I that can detect diabetes open A is DEV day announcements 00:11.840 --> 00:17.149 Elon Musk announcing his new A I called Grok as well as Amazon announcing their new A I 00:17.149 --> 00:22.139 as well as nvidia's supercomputer breaking record highs in A I model training 00:22.440 --> 00:29.379 Here are the top items in A I news and research. You don't want to miss out on this first article shows how A I 00:29.379 --> 00:37.959 can be used to identify faces of geese and other animals. Researchers have developed an A I tool for facial recognition in geese 00:37.959 --> 00:47.009 which is about 97% accurate in identifying geese by their beaks. Facial recognition with A I is also being used for other animals such as 00:47.009 --> 00:56.169 bears and lemurs. This is important for ecology and conservation because it allows researchers to determine population size and social interactions among individuals 00:56.880 --> 01:04.660 In this next article, researchers have developed an A I that can detect type two diabetes just by yzing a person's voice 01:05.029 --> 01:13.319 Nearly half of adults with diabetes are undiagnosed especially in low and middle income countries using only voice recordings 01:13.319 --> 01:21.480 They created an A I model that can detect diabetes with 89% accuracy in women and 86% accuracy in men 01:21.480 --> 01:26.680 This is possible because diabetes may affect the vocal cords and respiratory system 01:26.680 --> 01:35.120 affecting voice quality. This is just a very preliminary study and more research is needed to test the A I in larger and more diverse 01:35.120 --> 01:41.730 populations. Factors to consider include race, ethnicity and socio-economic background. Last week 01:41.730 --> 01:50.459 open A I held their dev day with exciting new announcements. They introduced GP T four turbo which is an improved version of GP T 01:50.680 --> 01:57.080 The context length increased to 100 and 28 K allowing for more information to be processed 01:57.809 --> 02:03.580 It can now follow instructions better and provide consistent outputs, knowledge cut off 02:03.580 --> 02:12.460 Also extended to April 2023 for more up to date information. These changes are now part of the paid plan of chat GP T 02:12.839 --> 02:18.979 It's also already faster than previous G BT versions but more speed improvements are coming 02:19.229 --> 02:25.220 Users can now build their own chat GP T and publish it themselves without much technical knowledge 02:25.220 --> 02:30.820 GP TS can be customized for various purposes such as lesson planners or graphic designs 02:30.990 --> 02:35.880 An entire GP T store will be published later this year in other news 02:35.880 --> 02:41.339 Hugh Maine. A start up in San Francisco has developed a wearable device called the A I PIN 02:41.710 --> 02:46.779 This is a small device that can be worn on the chest and has features like taking photos 02:46.779 --> 02:55.910 sending texts and projecting a visual interface onto the user's palm. It's designed to replace smartphones and reduce dependency on 02:55.910 --> 03:04.080 them. The A I PIN will be available for purchase in the US starting November 16th with a price of only 699 plus a monthly 03:04.080 --> 03:11.020 fee for unlimited calling, texting and data. The device is meant to be worn all day and is lightweight and comfortable 03:11.110 --> 03:17.070 It uses taps, hand gestures and voice commands for control and has a voice assistant called A I mic 03:17.559 --> 03:24.690 The device runs on Google's Android and will support navigation and shopping capabilities in the future 03:24.850 --> 03:31.389 It comes with a camera, a laser projector and other sensors and has a distinctive design made from aluminum 03:31.559 --> 03:39.639 The battery life is not disclosed but it comes with two battery packs and a portable charging case humane plans to add more features and 03:39.639 --> 03:47.000 integrations to the pin in the future. This is one of the first in a new trend of wearable devices that uses A I and 03:47.000 --> 03:54.740 is aimed at a wide range of users in other news. An A I model called Grok was launched by Elon Musk's Xa I 03:54.740 --> 04:00.500 Company Grok is designed to answer questions with humor and has real time knowledge of the world 04:00.559 --> 04:04.899 It's still in early beta stage but will improve with user feedback 04:05.289 --> 04:10.110 Xa I wants to create A I tools that benefit humanity and empower users 04:10.110 --> 04:19.320 So the goal for GROK is to assist in research innovation. And the pursuit of understanding GROK has shown strong performance in reasoning and coding 04:19.320 --> 04:24.549 tasks and surpasses other models in its compute class on various benchmarks 04:24.869 --> 04:28.950 Grok also performed well on a real life math exam, but again 04:28.950 --> 04:33.549 this is still in early beta. So more updates are going to come soon. Jumping over to Google 04:33.549 --> 04:41.809 Google has developed a new weather model called MET net three forecasting weather has historically been quite challenging and not very accurate 04:42.119 --> 04:51.100 Met. Net three provides accurate 24 hour forecasts for precipitation, temperature, wind speed direction dew point as well as other weather variables 04:51.700 --> 04:57.630 It outperforms traditional weather prediction models and has a high resolution of 1 to 4 kilometers 04:57.750 --> 05:05.140 If you want to get into the details. They used a technique called densification to merge weather data into a single pass through a neural 05:05.140 --> 05:10.709 network which is then trained to spit out a probability distribution for each weather variable 05:10.709 --> 05:19.109 providing very rich information. The system uses near continuous inference to create full forecasts based on incoming data streams 05:19.350 --> 05:24.359 The forecasts are available on Google's products providing hyper local weather insights 05:25.420 --> 05:32.339 Moving on to NVIDIA, their EOS supercomputer broke its own A I training record training 05:32.339 --> 05:36.119 A large language model can take a long time using regular hardware 05:36.119 --> 05:45.450 but that's no problem for NVIDIA which unveiled their supercomputer called EOS that can train a 175 billion parameter language model 05:45.450 --> 05:54.760 in just under four minutes. EOS is powered by over 10,000 H 100 tensor cord GP US and has 860 terabytes 05:54.760 --> 06:02.200 of high bandwidth memory. And for non technical folks out there, this all just means it's really bad ass going into the specifics 06:02.200 --> 06:11.720 NVIDIA set six records in benchmark tests including training a stable diffusion model in 2.5 minutes and training a bet language model in 7.2 06:11.720 --> 06:20.809 seconds. This supercomputer can be rented for only $37,000 a month by companies looking to expand their A I capabilities 06:21.010 --> 06:28.500 NVIDIA plans to use this supercomputer for various tasks such as model development GP U design and autonomous driving 06:28.890 --> 06:35.059 In other news. Git lab introduced Duo Chat which is a set of A I features to help developers be more productive 06:35.359 --> 06:40.459 Duo is similar to chat GP T. It's basically a chatbot that includes issue 06:40.459 --> 06:49.500 summarization code, suggestions, vulnerability explanations and more tools for developers. Duo can access all code even if it's not open in the 06:49.500 --> 06:55.019 ID E setting it apart from others in the field and giving it more context to help developers with coding 06:55.600 --> 07:03.579 In other news, Pepsico has developed an A I to eliminate crunching sounds while gamers wearing headsets eat Doritos 07:03.579 --> 07:10.269 which is also owned by Pepsico. They trained this noise cancellation A I with over 5000 crunch sounds 07:10.570 --> 07:19.309 The goal is to allow gamers to enjoy their Doritos without disturbing their fellow players in other news runway's gen two model has just been 07:19.309 --> 07:28.329 updated. Gen two now produces more realistic and consistent video results. Users can extend A I generated videos and control camera 07:28.329 --> 07:35.470 movements with the director mode feature. This gives creators much more control when generating their A I videos 07:36.260 --> 07:42.420 In other news, Samsung has developed an A I called Gauss that can generate images and summarize documents 07:42.959 --> 07:47.500 Gauss consists of three parts Gauss language Gauss code and Gauss image 07:47.769 --> 07:52.380 The language portion can handle tasks such as writing emails and translation 07:52.660 --> 08:01.700 Gauss image can generate and modify images for creative use. The model is named after mathematician Carl Gauss and reflects Samsung's vision to 08:01.700 --> 08:10.269 improve consumers' lives with A I. It's possible that Gauss could be included in the upcoming galaxy S 24 smartphone in other news 08:10.269 --> 08:18.089 Google's A I powered search experience also known as SGE is expanding globally to over 100 and 20 countries 08:18.380 --> 08:25.269 Sge is basically like a A I chatbot in Google's search results think of it as like a competitor to bing chat 08:25.519 --> 08:33.820 Sge has been updated to include new features such as summarization of articles and definitions of unfamiliar words 08:33.820 --> 08:42.830 coding improvements, image generation and writing drafts. Google is also testing a way for users to ask follow up questions directly on the 08:42.830 --> 08:51.260 search results, page more updates to come on that. And finally, Amazon is working on a new A I model called Olympus 08:51.539 --> 08:55.739 They claim it will be twice the size of open A is GP T four 08:56.000 --> 09:05.450 Olympus has two trillion parameters compared to GP T four's one trillion Amazon plans to incorporate Olympus into its online store and Alexa smart 09:05.450 --> 09:10.400 speakers. And that's it for the top A I news and research this week 09:10.400 --> 09:16.140 If you enjoyed this content, be sure to like, share and subscribe and let me know in the comments what other types of videos you 09:16.140 --> 09:20.210 would like to see. Also, we built a site where you can search all the A I tools out there 09:20.210 --> 09:22.719 check it out at A I dash search dot IO