#AI - Page 7
486 Stories
YouTube CEO Says OpenAI Could’ve Violated Platform Rules For Sora Training
Arjun Sha
View quick summary
Speaking to Bloomberg, the YouTube CEO, Neal Mohan said that OpenAI would violate YouTube's Terms of Service if the company used YouTube videos to train its Sora model. Google also uses some corpus of YouTube videos to train its Gemini model, but signs licensing contracts with individual creators. Earlier, the CTO of OpenAI, Mira Murati, refused to clarify whether Sora was trained on YouTube videos.
Ajaay Srinivasan
DALL-E can generate solid visuals but has lagged behind Midjourney in editing, thus limiting its creative appeal. However, thanks to OpenAI's new update, the tool now offers a powerful new editing interface that allows users refine portions of an images with modified prompts. The new interface boasts a "Select" option that offers precise targeting of areas for modification, enabling you to add, remove, or adjust elements.
Just like Microsoft Copilot, you can now use ChatGPT without an account. You no longer need to sign in or create an account to use the free version of ChatGPT. The GPT-3.5 model is readily available to all free users including those without an account. Keep in mind, you won't get some features like custom instructions, chat history, chat sharing, and voice conversations without an account.
The new AI PC era is here. Basically, AI PCs have a special hardware component known as Neural Processing Unit (NPU). They are also designed with a better GPU than previous-generation PCs for the upcoming AI era of computing. While an AI PC can be worth it if there are features that can benefit your work, we are a long way away from the AI PC maturing. So, it's not worth buying an AI PC for now, unless you can take advantage of the features.
OpenAI's ChatGPT will now cite sources in its responses similar to Microsoft's Copilot, providing transparency and legitimacy in reponses. This feature is limited to paid versions: ChatGPT Plus, Team, and Enterprise unlike Copilot which offers it for free. The announcement for this was made by OpenAI through an X post on Friday.
Why Google Keep’s ‘Help Me Create’ Is the Best Feature for Procrastinators
Abubakar Mohammed
View quick summary
Google is rolling out a new generative AI feature, this time inside the Google Keep app. The feature called "Help me create a list" takes input and can generate a list. It makes it easier to add things to your list on the get-go and skip the research part. The feature appears as you tap the "+" icon in Google Keep to create a new Note. The "Help me create a list" button in gradient appears on the bottom-right corner.
After introducing Sora, a remarkable text-to-video AI model, OpenAI has previewed its Voice Engine model that can clone voices with a single 15-second audio sample. It can translate voices into different languages with high accuracy. Since there are risks associated with the technology, the company is not releasing the speech synthesis model now. The company encourages society to understand the capabilities of AI and adapt to a new reality.
Elon Musk's xAI firm has launched a new intermediate model called the Grok-1.5. It comes with improved reasoning capabilities and supports a large context length of 128K tokens. On the NIAH test, the model has shown great retrieval capability. xAI says early testers and existing Grok users will be able to access the model on X (formerly Twitter) in the coming days.
Snapchat offers an AI Caption feature that uses AI to analyze the image in your Snap and generate a caption based on its content. The feature is currently only available in a limited number of regions, including the US and UK and can only be used by Snapchat+ members. You can create AI Captions on Snapchat by taking a Snap first and then navigating to T icon > AI captions.
The ChatGPT maker, OpenAI, is reportedly planning to release the GPT-5 model during the summer of this year. A new report says that OpenAI has already demoed the GPT-5 model to a few enterprise customers and one of the CEOs called it "materially better" than GPT-4. OpenAI is also working on AI agents that can perform complex actions.
Microsoft Appoints Google Deepmind Co-founder Mustafa Suleyman to Lead Its AI Efforts
Arjun Sha
View quick summary
In another reshuffle, Microsoft has hired Mustafa Suleyman, the co-founder of Google DeepMind, and Inflection to lead AI innovations at Microsoft. Suleyman will oversee the development of Copilot, Edge, and Bing and how AI can be leveraged to make meaningful product improvements.
This list will offer some of the best options in the realm of AI powered tools that can reproduce human like voice from given text. These include options like the popular Speechify voice generator and other tools like Play.HT that generate almost natural voice-overs. You can try out any of the ones that we have mentioned in this list and see which one works best for you.
Partner Content
If you are a developer building AI products, LimeWire AI API offers great flexibility and pricing. It hosts multiple Diffusion models from OpenAI, Stability AI, Google, etc. to deliver generative AI features. You can use the API to generate images, upscale images, inpaint, and oupaint images as well. In our testing, the API performed pretty well. Not to mention, the API implementation is straightforward and supports many popular languages.
The Google DeepMind team has developed a generalized AI agent called SIMA that can perform actions in a wide range of virtual games. Google has trained the AI agent on nine video games including No Man's Sky, Teardown, Valheim, and more. It can carry out tasks even on unseen video games which makes it a unique AI agent.
OpenAI might be working to release an intermediate GPT-4.5 Turbo model before launching the next-gen GPT-5 model. It was leaked through the OpenAI Blog page which was indexed by Bing and DuckDuckGo. The upcoming model promises better "speed, accuracy, and scalability" than the GPT-4 Turbo model. However, strangely, the captured description says the model will have a knowledge cutoff of up to June 2024.
Microsoft is slowly adding new features to Copilot on Windows 11. Copilot has finally received the file upload capability on Windows 11. You can upload a range of documents including PDF, DOC, XLS, PPT, TXT, and more. Copilot seamlessly ingests the documents and answers from the provided document accurately. From analyzing financial sheets to summarizing PDFs and understanding private code documentation, the feature can be immensely helpful to all kinds of users.
After filing a lawsuit against OpenAI for becoming a closed-source company, Elon Musk has now announced that xAI's Grok model will be open-sourced this week. Musk founded xAI in March last year to rival OpenAI and released its first preview model in November. However, Grok has not been received well and it's highly prone to hallucination. Open-sourcing the model may help the company improve Grok on many fronts.
Microsoft launched the Seeing AI app with AI features during its Ability Summit 2024. The app is powered by AI and allows low-vision users to make sense of the world by hearing descriptions of scenes, images, products, currency, and more. You can install the app on Android and iOS for free. It's currently available in 33 languages, and Microsoft is planning to bring support for 36 languages by the end of 2024.
Google's latest update for Gemini allows users to modify and regenerate specific portions of a response according to their preferences. The feature can be used to replace portions of text with a different prompt or you can simply ask Gemini to regenerate selected text or make it shorter or longer. The new feature is available for everyone but is limited to the Gemini web app.
Microsoft is Holding a Surface, Windows & Copilot AI Event Later This Month
Arjun Sha
View quick summary
Microsoft has announced a new event for March 21st where the company is likely to unveil Windows 11 improvements, new Copilot features, and refreshed Surface devices. The company may release refreshed versions of Surface Laptop 6 and Surface Pro 10. Microsoft is also preparing to showcase an advanced version of Copilot with a new feature called "AI Explorer".
If you have a low-end computer, you can download and run Google's open-source Gemma model on Windows, macOS, and Linux. The model is just 1.5GB in size and takes up around 1.4GB RAM. For creative tasks in English, the model does a good job while running offline. You can download LM Studio and load the model to start using it right away.
Anthropic has immediately released APIs for its Claude 3 models including Opus and Sonnet. The company says API access for the smallest Haiku model is coming pretty soon. While the API pricing for the Claude 3 models is high, users and developers are keen to test the Opus model, which according to Anthropic, beats GPT-4 and Gemini 1.0 Ultra. We have also added some code examples so go through our detailed tutorial.
Anthropic Announces Claude 3 AI Models; Beats GPT-4 and Gemini 1.0 Ultra
Arjun Sha
View quick summary
Anthropic has released a new family of Claude 3 models -- Opus, Sonnet, and Haiku. The largest and most capable Claude 3 Opus model beats GPT-4 and Gemini 1.0 Ultra in all major benchmarks. According to Anthropic, all three models support a context window of 200K tokens and deliver 99% accuracy with great recall. Plus, they come with vision capability as well.
AI Models in India to Require Govt Approval; What are the Implications?
Arjun Sha
View quick summary
The Indian IT Ministry (MeitY) has issued a new advisory for tech companies offering AI models and services in India. The advisory asks tech firms to get approval before deploying "untested" AI models in India. The government later clarified that the advisory applies to large companies like Google, OpenAI, Microsoft, etc., and not startups. The advisory also requests companies to embed permanent metadata in generated data to easily identify the first originator.
You can sign up for the Gemini 1.5 Pro waitlist via Google AI Studio and get early access to the flagship model with a context window of 1 million tokens. The model is currently in preview, so Google is offering access for free to test and evaluate the model. That said, there is no API available for Gemini 1.5 Pro yet, just like Gemini 1.0 Ultra.
In a surprising turn of events, Elon Musk has sued OpenAI and its CEO, Sam Altman in the San Francisco Superior Court on Thursday. Musk claims that OpenAI has become a closed-source company and it's only developing new technologies to maximize profits. Further, Musk says that Microsoft is leveraging its power and influencing OpenAI's operations. Musk is seeking an injunction against OpenAI and Microsoft from taking advantage of the AGI technology and cashing in for profits.
I Got Access to Gemini 1.5 Pro, and It’s Better Than GPT-4 and Gemini 1.0 Ultra
Arjun Sha
View quick summary
We got our hands on the Gemini 1.5 Pro model via Google AI Studio, and after probing the model on a multitude of tests, we can say that Google has finally delivered an immensely powerful AI model. It's easily on par with GPT-4 model by OpenAI and surpasses Google's largest Gemini 1.0 Ultra model. It's excellent at advanced reasoning, can process videos, handles large corpus of data in a single window, and you can do so much more. Read our detailed comparison between Gemini 1.5 Pro, Gemini 1.0 Ultra, and GPT-4.
What the Gemini Image Generation Fiasco Tells Us About Google’s Approach to AI
Arjun Sha
View quick summary
After Gemini generated some inaccurate and offensive images, Google has been accused of anti-white bias by critics from many quarters. In response, Google has temporarily turned off image generation of people in Gemini. Moreover, many accuse Google of aggressively tuning the model to represent diversity which seems to have backfired. So what explains this debacle and Google's overall approach to AI? Read on to find out.
You Can Now Set Copilot As the Default Assistant on Android; Here’s How
Anshuman Jain
View quick summary
Microsoft's Co-Pilot AI can now be set as the default assistant on Android devices. The latest beta version allows users to replace Google Assistant with Co-Pilot, which can be triggered from any screen. However, it currently lacks voice activation and screenshot capabilities.
Microsoft is looking to expand its AI investment beyond OpenAI. The software giant has partnered with a French AI startup called Mistral AI and invested close to $2 billion to accelerate their new LLMs on Azure's infrastructure. Mistral AI's new model, Mistral Large, is already available on Azure AI Studio for customers and developers. The Mistral Large model supports advanced reasoning, understands several international languages, and comes with native function calling support.
Qualcomm has finally released its AI Hub and AI Stack to run AI models locally on Snapdragon platforms, be it your smartphone or PC. The chipmaker has optimized over 75 AI models to run locally on your device. You can generate texts and images, enhance low-light images, perform image segmentation, and do much more locally. Qualcomm's AI Engine delivers 3x-4x better performance than Intel's latest Core Ultra processors.
Meet Groq, a Lightning Fast AI Accelerator that Beats ChatGPT and Gemini
Arjun Sha
View quick summary
Groq is a company formed by ex-Google TPU engineers who have developed an LPU (Language Processing Unit) that can generate outputs at a blistering speed. It can generate over 500 tokens per second while using a 7B model and close to 250 tokens per second while using a 70B model. ChatGPT and Gemini generate responses at a speed of 50 to 60 tokens per second. The Groq LPUs are said to be highly performant with much less latency and minimum energy consumption. With the introduction of LPUs, expect instant interaction with AI models soon.
After launching Gemini 1.0 Ultra and Gemini 1.5 Pro in the last few weeks, Google is back with new AI models that are actually open-source. Dubbed "Gemma", these are small open-source models which can run even on consumer laptops. They come in two sizes, one with 2B parameters and another with 7B parameters. The models are largely trained in the English language and are suited for text generation, summarization, reasoning, and Q&A. Google has given a commercial license to Gemma models subject to certain prohibited use policies.
OpenAI has unveiled its remarkable text-to-video model, Sora, that can generate AI videos for up to 1 minute. The generated videos are highly detailed and have a resolution of up to 1080p. Unlike existing solutions, Sora-generated videos don't have distortions in the scene, which makes everyone believe that it's trained on gaming engine simulations like Unreal Engine 5. OpenAI says it's a diffusion model built on the transformer architecture. The Sora model is going through safety checks now and currently, it's not available to regular users.
Google Introduces Gemini 1.5 Pro with a Massive 1 Million Context Window
Arjun Sha
View quick summary
After releasing Gemini Advanced with the Ultra 1.0 model last week, Google has announced its next-generation model called Gemini 1.5 Pro. The new model is built on the Mixture-of-Experts (MoE) architecture and supports a massive context window of up to 1 million tokens. It performs nearly the same as the much larger Gemini 1.0 Ultra model. The AI model is currently in limited preview and developers can test it on AI Studio after joining the waiting list.
Nvidia Launches Chat with RTX, an AI Chatbot You Can Run Locally on Windows
Satyam Kumar
View quick summary
Nvidia has released a new tool called Chat with RTX. This lets Windows PCs with Nvidia RTX 30 series or RTX 40 series GPU run an AI chatbot locally on their PC. Custom data can be given to the AI including documents, PDFs, and even YouTube videos. It includes the Mistral and Llama-2 open-source LLMs in the download. 16GB of RAM is also required and the GPU must have at least 8GB of VRAM capacity in order to work.
Google has launched a subscription plan called Gemini Advanced that is powered by the Gemini Ultra 1.0 model. Unlike the free Gemini (formerly Bard), Gemini Advanced is much better at highly complex tasks and performs close to OpenAI's GPT-4 model. It costs $20 per month, but you get two months of free trial. Gemini Advanced also bundles 2TB of storage, and Google One benefits. And soon, Gemini Advanced users will be able to access Gemini AI in apps like Gmail, Docs, and more.
The e-commerce giant, Amazon, has added an AI shopping assistant called Rufus to its shopping app. It's an AI chatbot trained on Amazon's vast product catalogs, customer reviews, community Q&A, and information sourced from the web. It can answer all your questions based on your shopping needs, find personalized products, offer recommendations for various occasions, and much more. The Rufus AI assistant is currently being rolled out to a small subset of users in the US only.
Google has finally added image generation into Bard and you can create images for free. Simply start your prompt with "create an image of" or "generate an image of" and Bard will generate two images at once. Google is using its in-house Imagen 2 AI model with ImageFX tool for added guardrails to power Bard image generation. Right now, it only supports prompts in English and some countries such as the UK, Switzerland, and EEA regions can't access the feature right away.
AI Breaks the Bank: 80% of Jobs Find Replacing Humans Costly, Reveals MIT Study
Vikhyat Rishi
View quick summary
A recent MIT study eases the job automation fears, revealing that only 23% of certain jobs can be cost-effectively replaced by AI. Economic hurdles and high upfront cost of AI installation and maintenance contribute to a prolonged timeline. The study further says that at least until 2046, human labor will be a valued asset.
Kanika Gogia
Sagnik Das Gupta
A group of researchers from Mohamed Bin Zayed University of Artificial Intelligence in Abu Dhabi developed a new AI tool to imitate human handwriting. Apparently, it just requires a couple of paragraphs to analyze the handwriting and copy it. Previously, generative adversarial method was used to mimic someone's handwriting. Although it worked great, it could not work on the subtle elements of a human handwriting. That's why they followed a vision transformer-based solution to formulate a handwritten text image generation approach. With this, the AI can be trained to accurately capture the essence of each human handwriting.