It's AI Time

 AI Edition - It's AI Time - 199 Pushing the Frontiers of Audio Generation  DeepMind has talked a little bit more about the audio generation models used to power NotebookLM. OpenAI's new hallucination benchmark  OpenAI has released the SimpleQA benchmark, which measures models' abilities around simple factual questions. Evaluating feature steering: A case study in mitigating social biases  The feature steering in AI models to interpretably modify outputs. It reveals a "steering sweet spot", where changes do not degrade capabilities. The study results show steering can alter social bias in targeted domains but also brings unexpected off-target effects. Further research is required to refine feature steering for safer, more reliable outcomes in AI models.

AI Edition - It's AI Time - 198 An Interview with Salesforce CEO Marc Benioff about AI Abundance Salesforce's CEO Marc Benioff recently discussed the company's new AI initiative, Agentforce, highlighting its potential to revolutionize enterprise applications and customer interactions. Benioff criticized Microsoft's Copilot, positioning Salesforce's approach as more integrated and effective due to its robust platform and data architecture. In the interview, he emphasized the importance of AI-driven "agentic" layers, which aim to enhance customer service and operational efficiency across industries. Apple releases first batch of Apple Intelligence features Apple launched new AI features, called Apple Intelligence, on recent devices for text processing and photo editing. The revamped iMac is powered by the M4 chip, which features a Neural Engine and boasts triple the AI performance of past chips. Future AI updates will enhance Siri and integrate with ChatGPT ...

AI Edition - It's AI Time - 197 Google Expands Voice Technology Support to 15 More African Languages Google has introduced voice recognition for 15 additional African languages across its platforms, including Voice Search, Gboard talk-to-type, and Translate dictation. This update empowers approximately 300 million more Africans to interact with digital content in their native languages. Cohere releases state-of-the-art multimodal AI search model Cohere announced that its Embed 3 AI model is now multimodal, enabling fast and accurate search across important enterprise image data sources like graphs, charts, product catalogs, and design files. With this upgrade, Embed 3 is the most generally capable multimodal embedding model on the market. Arcjet: protect your AI app from bots and abuse AI costs money. Enforce user limits and prevent abusive bots using Arcjet security as code. Thinking Like an AI LLMs function as sophisticated autocomplete systems, predicting the next token based on...

AI Edition - It's AI Time - 196  AI jewelry Arcade AI has built a generative product platform for creating unique and quality jewelry items from text prompts. The best part is, you can buy what you prompt. IBM debuts open source Granite 3.0 LLMs for enterprise AI IBM is expanding its enterprise AI offerings with the new Granite 3.0 LLMs, focusing on open-source solutions and performance. These models, available on multiple platforms, include safety features and are tailored for various enterprise use cases. IBM emphasizes the importance of real open-source licensing with Apache 2.0 to enable flexible enterprise adoption and innovation. Microsoft introduces ‘AI employees' that can handle client queries Microsoft is launching autonomous AI agents to enhance productivity, allowing businesses to build custom bots for tasks like customer service and supply chain management. ByteDance intern fired for planting malicious code in AI models ByteDance has confirmed an intern interfered w...

AI Edition - It's AI Time - 195 Government AI Acquisition Guidelines: The U.S. government is tightening AI acquisition standards to ensure responsible, innovative use, particularly when AI impacts privacy or safety. The Office of Management and Budget (OMB) recently issued detailed guidelines to regulate AI procurement, emphasizing data protection, risk management, and public safety. This aligns with a broader effort to secure and ethically advance AI's role in federal operations (The White House). Generative AI in Business: Businesses continue to invest heavily in generative AI (Gen AI) tools for tasks such as supply chain management and customer service, aiming to increase revenue and operational efficiency. McKinsey reports that, while many companies achieve cost savings, inaccuracy in generative models is a common challenge, leading organizations to implement strict governance practices to ensure reliability and security. Health and Climate Applications: AI applications in ...

AI Edition - It's AI Time - 194 MIT AI Conference in New York: The MIT AI Conference is bringing together AI leaders to discuss the transformative role of AI across sectors like business, healthcare, and government. This gathering of experts showcases real-world case studies, with a particular focus on addressing ethical challenges in AI and applications of generative AI technology in practical fields like entertainment and public administration. GenAI Hackathon Finale: In Hyderabad, India, the GenAI Hackathon 2024 is spotlighting innovative generative AI prototypes. With participants competing for cash prizes, the event emphasizes the creativity and potential of AI in developing new applications, such as personalized educational tools and customer service automation, aimed at improving user experience and accessibility. Stanford AI Index Report 2024 : Stanford’s AI Index report reveals trends like a high adoption rate of AI in Fortune 500 companies and growing diversity within AI...

AI Edition - It's AI Time - 193 Grammy-Winning Musician and Google Unveil MusicFX DJ Jacob Collier, a Grammy-winning musician, has teamed up with Google DeepMind and Google Labs to create MusicFX DJ, an AI-powered music tool. The interface has been redesigned to encourage creativity and help users easily enter a "flow state" of artistic inspiration. MusicFX DJ is available now, offering intuitive controls for all skill levels. The AI Investment Boom The AI boom is driving significant US investment in data centers, computing infrastructure, and advanced hardware, with data center construction reaching a record-high $28.6 billion annually. This surge is fueled by the growing demand for powerful computing resources necessary for training and deploying advanced AI models. While revenue rebounds in the tech sector, job growth remains concentrated in semiconductor manufacturing and infrastructure, diverting focus from traditional programming roles. Apple internally believes tha...

AI Edition - It's AI Time - 192 Simplifying, Stabilizing, and Scaling Consistency Models OpenAI has released some work discussing improvements to consistency models (which generate in two steps as opposed to 1,000 for diffusion). They still rely on distillation from an existing diffusion model, but this work aims to improve and stabilize them as they scale. Sabotage evaluations for frontier models Anthropic's Alignment Science team developed evaluations to test AI models' potential for sabotage, including human decision sabotage, code sabotage, sandbagging, and undermining oversight. These evaluations aim to identify dangerous capabilities in future AI systems and inform necessary safety mitigations. Their findings suggest minimal risks with current models but highlight the need for more robust evaluations and safeguards as AI capabilities advance. AI Dreams: Microsoft @ 50, Chapter 1 Microsoft's research paper on AI robustness prompted the company to invest billions in...

AI Edition - It's AI Time - 191 Stable Diffusion 3.5 Stability AI has released a new series of models with improved performance and speed. They have Diffusers support with training out of the gate. The Second $100B AI Company New AI companies driven by advancements in AI applications, especially in consumer AI, will join OpenAI in surpassing a $100B market cap. Despite the current dominance of established tech giants in the AI infrastructure and model layers, the application layer remains ripe for innovation and growth, presenting opportunities for consumer AI to thrive. The potential for large-scale success in consumer AI, particularly in areas like video creation, online shopping, and gaming, mirrors previous tech revolutions in cloud and mobile sectors. ChatGPT has a Windows app now OpenAI is testing a ChatGPT app for Windows. Adobe's new image rotation tool is one of the most impressive AI concepts we've seen Adobe's Project Turntable uses AI to rotate 2D vector art...

AI Edition - It's AI Time - 190 State of AI Report 2024 The 2024 State of AI Report highlights the consolidation of foundation models into practical products, with OpenAI leading significant revenue generation. Key trends include convergence in frontier lab performance, increasing focus on planning and reasoning in LLM research, and the expansion of foundation models into multimodal domains. Despite regulatory challenges, AI enterprises have surged in value, though questions about long-term sustainability persist. xAI, Elon Musk's AI startup, launches an API xAI has released a Grok-based generative AI API. It currently offers a single model, "grok-beta," with function-calling capabilities. The API is part of xAI's strategy to compete with AI leaders by leveraging data from Musk's companies, including X. Despite some functionality issues, xAI is rapidly expanding Grok's integration into the X platform. Mini-Omni2: Towards Open-source GPT-4o with Vision, Spe...

AI Edition - It's AI Time - 189 Anthropic just made it harder for AI to go rogue with its updated safety policy Anthropic updated its Responsible Scaling Policy to include Capability Thresholds for AI models that pose significant risks, such as bioweapons and autonomous AI research. This policy aims to set industry standards by introducing AI Safety Levels that require stricter safeguards based on model capabilities. By publicly disclosing safety practices and maintaining a Responsible Scaling Officer, Anthropic hopes to lead in AI governance and inspire similar approaches industry-wide. Sam Altman's Worldcoin becomes World and shows new iris-scanning Orb to prove your humanity The World project, co-founded by Sam Altman, aims to verify human identity online using iris-scanning technology amidst privacy concerns and ongoing investigations in the EU. It plans to incorporate human verification into AI platforms and potentially redistribute AI-generated wealth through Worldcoins. ...

AI Edition - It's AI Time - 188 Google's AI Search Enhancements: Google has unveiled significant updates to its Search capabilities, integrating AI to allow users to search through video content and recognize moving objects. This includes an enhanced Google Lens, which now supports more interactive queries and offers AI-generated overviews to assist with complex questions. X (formerly Twitter) Privacy Changes: Elon Musk’s platform is updating its privacy policy, enabling third parties to utilize user posts for training AI models. This move has raised concerns regarding user data privacy and the ethical implications of such practices. New Developments in AI Models: Mistral, a notable AI company, has released new models optimized for laptops and smartphones, showcasing a trend towards making AI technology more accessible on personal devices. ChatGPT on Windows: OpenAI's ChatGPT has made its way to Windows, providing users with easier access to AI tools directly from their de...

AI Edition - It's AI Time - 187 Small Language Models (SLMs): While large language models (LLMs) like GPT remain powerful, smaller models are gaining traction due to their ability to run on devices with fewer computing resources, like smartphones, while still maintaining high performance. This is making AI more accessible and affordable, enabling broader adoption. Multimodal AI: AI systems capable of processing multiple data types (text, images, audio, etc.) are becoming more advanced. These systems enhance tools like design apps and voice generation platforms, creating richer user experiences by integrating different sensory inputs. AI in Science: AI is playing a critical role in speeding up scientific discoveries. In fields like agriculture, climate change mitigation, and medicine, AI-driven tools are accelerating research and offering innovative solutions, such as helping farmers with weather prediction and optimizing crop management. Sustainability in AI: Companies are incre...

AI Edition - It's AI Time - 186 Customize Audio Overviews with Google's NotebookLM NotebookLM now allows users to tailor their Audio Overview experience, giving more control over the AI hosts' focus and expertise. Businesses can apply for the new NotebookLM Business pilot program, which offers enhanced tools for professional use. MIT Researchers Combine Next-Token Prediction and Video Diffusion for Robotics A new method combines next-token prediction and video diffusion, enabling neural networks to clean corrupted data while predicting future steps. This approach enhances a model's ability to generate high-quality video, create flexible plans for robots, and assist AI agents in navigating digital environments. Nvidia just dropped a new AI model that crushes OpenAI's GPT-4 Nvidia's new AI model, Llama-3.1-Nemotron-70B-Instruct, surpasses performance benchmarks set by OpenAI and Anthropic. This model marks Nvidia's strategic expansion from hardware to sophisti...

 AI Edition - It's AI Time - 185 Meta AI's New Tracking Model CoTracker is a new tracking model that addresses the performance gap between synthetic and real video data by using semi-supervised training. Former Palantir CISO Dane Stuckey joins OpenAI to lead security OpenAI hired former Palantir CISO Dane Stuckey to strengthen its security operations. The company seeks to enhance its AI security infrastructure and has been working with the Pentagon on cybersecurity-related projects. OpenAI aims to build secure AI systems while exploring government partnerships including with the U.S. Department of Defense. Can AI really compete with human data scientists? OpenAI's new benchmark puts it to the test OpenAI's MLE-bench tests AI capabilities in machine learning engineering through 75 Kaggle-style competitions. The benchmark highlights AI progress but also its limitations in creative problem-solving compared to humans. Open-sourcing MLE-bench could standardize AI evaluation ...

AI Edition - It's AI Time - 184 Meta's open AI hardware vision Meta showcased open AI hardware designs, including the Catalina rack and the expanded Grand Teton platform, at the OCP Global Summit. Training the Llama 3.1 405B model required 16,000 NVIDIA H100 GPUs, highlighting Meta's scaling infrastructure. Open AI hardware systems are required to continue advancing AI capabilities. The New York Times warns AI search engine Perplexity to stop using its content The New York Times has issued a cease and desist to AI startup Perplexity, accusing it of using its content without permission for AI search operations. Perplexity claims it doesn't scrape content for training but indexes web pages to surface factual content. The company is engaging with publishers and aims to resolve the issue by working with the Times and other media outlets. Decagon raises $65m Series B Decagon has raised $65 million in Series B funding to enhance its AI customer support agents, which are alrea...

 AI Edition - It's AI Time - 183 Adobe Announced Video Models Adobe announced the expansion of its Firefly family of creative generative AI models to video during the Adobe MAX conference. OpenAI appoints international expansion boss OpenAI has appointed Oliver Jay to lead international expansion, focusing on AI strategy and operations. The company announced a new APAC office in Singapore and is developing local language datasets. The o1 model, which utilizes "chain of thought" techniques, is aimed at enhancing AI accuracy. Microsoft may have an audio-to-image generator in the works Microsoft has filed a patent for an AI system that converts live audio into images using LLMs. The system aims to enhance communication by generating real-time images from audio streams. If developed, it could be integrated into Microsoft Teams via Copilot. Al Will Take Over Human Systems From Within Yuval Noah Harari, an Israeli historian and the author of “Sapiens,” “Homo Deus,” and “Nexus,”...

AI Edition - It's AI Time - 182 Silicon Valley is debating if AI weapons should be allowed to decide to kill Debate over fully autonomous weapons continues as Anduril's Palmer Luckey expresses openness to them, contrasting with Shield AI's previous stance against AI making lethal decisions. The U.S. government remains non-committal, with no binding ban on such weapons, amidst fears that adversaries like China or Russia might develop them first. Defense tech companies, including Anduril and Palantir, are lobbying heavily to influence policy on the integration of AI in military systems. This AI-Powered Invention Machine Automates Eureka Moments Iprova's AI-driven software analyzes diverse technical literature to suggest patentable inventions by connecting previously unlinked ideas, aiding companies like Procter & Gamble and Panasonic. The tool uses semantic search and generative AI to identify novel inventions and relies on human insight for practical application, foc...

AI Edition - It's AI Time - 181  Google's Nobel Laureates and AI Recognition: Several 2024 Nobel Prizes were awarded to researchers affiliated with Google, particularly from the DeepMind team. Demis Hassabis and John Jumper received the Nobel Prize in Chemistry for their AI-driven research on micro-proteins, while Geoffrey Hinton won the Physics Nobel for his contributions to AI systems, specifically related to advancements in neural networks. These awards signal the Nobel Committee's growing recognition of AI's critical role in scientific breakthroughs, though there has been some debate about the categorization of these achievements. Microsoft's €4.3 Billion AI Investment in Italy: Microsoft is making a significant investment to boost its AI infrastructure in Italy. This initiative will not only create one of Europe’s largest AI and cloud hubs but also train over a million people in digital skills by 2025. The effort focuses on enhancing AI’s role in key industries...

AI Edition - It's AI Time - 180 Generative AI Adoption: Companies across various industries are increasingly adopting generative AI. A recent McKinsey survey revealed that industries like human resources, supply chain management, and marketing are experiencing significant cost reductions and revenue gains thanks to AI integration. However, businesses are also grappling with risks related to data privacy, IP infringement, and model inaccuracies, which have become top concerns. TikTok and AI: TikTok has been in the spotlight due to its plan to reduce workforce numbers and shift toward more AI-driven content generation, underscoring the ongoing trend of automation replacing human labor in digital content creation. AI in the Nobel Prize Arena: Artificial intelligence played a crucial role in this year's Nobel Prize categories, with scientists and the tech industry celebrating the contribution of AI in scientific advancements. This marks a significant recognition of AI’s impact on ...

AI Edition - It's AI Time - 179 Uber's EV ride hailing business is maturing Uber's new EV-only Uber Green service is expanding to 40 cities worldwide, encouraging more environmentally friendly rides. The company is introducing a GPT-4 powered chatbot to assist drivers with EV-related queries, launching in 2025. A mentorship program for EV drivers will also provide support and incentives for transitioning drivers. Amazon's new AI guides can help shoppers find what they need Amazon has launched AI-powered "Shopping Guides" to streamline product searches by offering tailored recommendations. These guides are integrated into Amazon's mobile platforms across 100+ product types, enhancing user decision-making with targeted insights based on customer reviews and popularity. The AI feature organizes products visually, simplifying the filtering process for users. Measuring AI's Engineering Skills MLE-bench is a benchmark designed to test AI agents' abilitie...

AI Edition - It's AI Time - 178 OpenAI reducing dependency on Microsoft data centers OpenAI is reducing its dependence on Microsoft data centers by securing its own compute capacity. Microsoft is also decreasing its reliance on OpenAI as it competes in the AI product market. This company is merging patented next-gen AI, compression & VR technology with the highest grossing character IP in history  and is offering investors a piece for a limited time Elf Labs, known for winning historic trademark cases for iconic characters like Snow White, Cinderella, Sleeping Beauty, and the Little Mermaid, is raising an investment round to bring them to life using AI, AR, and VR. With two projects funded and their goal surpassed, limited space remains. Invest before the round closes on Oct. 30 and earn up to 40% bonus shares and exclusive investor perks! TikTok parent company ByteDance has a tool that's scraping the web 25 times faster than OpenAI ByteDance is aggressively using its web c...

AI Edition - It's AI Time - 177 Navigating Gen AI's middle miles As we approach the “middle mile” moment in the adoption of generative AI technologies, only 34% of software and platform companies have embedded AI into all business processes. Learn the 3 key recommendations that will help address the challenges to come. A Nobel Prize for neural network pioneers Hinton and Hopfield The Royal Swedish Academy of Sciences awarded the Nobel Prize in physics to deep learning “godfather” Geoffrey Hinton and machine learning pioneer John Hopfield for their work on the artificial neural networks that underpin today’s AI revolution. Meta debuts movie generation AI model The social media company unveiled Movie Gen, a powerful generative AI model that can create high-quality short videos from text prompts. Text prompts can also be used to edit the videos and the model can automatically create AI-generated sound effects or music appropriate to the scene an advance over other text-to-video ...

AI Edition - It's AI Time - 176 Kobold Materials raises $491m Kobold Materials, a mining startup that uses AI to identify and extract metals, has raised a substantial new round of funding. OpenAI gets $4 billion revolving credit line, giving it more than $10 billion in liquidity OpenAI has secured over $10 billion in liquidity, reaching a $157 billion valuation after its latest funding round. The company raised $6.6 billion from major investors, including Microsoft and Nvidia, but faces high operational costs, primarily due to the need for more GPUs to support LLM training. OpenAI is exploring restructuring options to optimize financial growth and sustainability in the AI ecosystem. Google brings ads to AI Overviews as it expands AI's role in search Google will show ads in its AI Overviews to counter alternatives like ChatGPT while enhancing user engagement. Despite generating controversies over misinformation, AI Overviews have boosted search engagement, especially in younger ...

AI Edition - It's AI Time - 175 Meta Movie Gen Meta has trained a state-of-the-art movie model. It has 30B parameters and took 6,144 H100s to train. The model was trained on 1B images and 100m highly curated videos. Interestingly, it is a Temporal Autoencoder and Flow matching Llama. Meta released a 92 page paper, one of the most detailed research reports on the topic. Embers of Autoregression for o1 Large language models have limitations due to next token prediction. However, OpenAI's o1 model was trained with a new objective over reasoning traces. Despite this change, o1 still shows signs of limitations due to next token prediction. Liquid Foundation Models: Our First Series of Generative AI Models Liquid AI has announced its first generation of Liquid Foundation Models (LFMs), which deliver state-of-the-art performance with reduced memory usage. The LFMs, optimized for various hardware, include a 1B, 3B, and 40B model and are already available on platforms like LIQUID PLAYGR...

AI Edition - It's AI Time - 174 Google Chrome Book Google introduced new AI-powered features for its Chromebook Plus lineup, including tools like "Help me read" for summarizing and clarifying text, and "Live Translate," which enables real-time translations across over 100 languages during video calls, making it easier to overcome language barriers. Apple Apple also launched new AI capabilities for its devices, integrating deeper natural language understanding into Siri. Users can now communicate more fluidly with Siri, which will soon be able to perform actions based on contextual understanding across apps, such as finding content from texts or emails and making personalized recommendations. Apple continues to emphasize privacy with these advancements, ensuring AI processing happens on-device for increased user control​. Meta AI Updates Meta announced new features for their AI assistant, which is now being used by over 400 million people monthly. The assistant c...

AI Edition - It's AI Time - 173 Google Google introduced significant upgrades to its search capabilities, integrating AI more deeply with visual and voice-based queries. Google Lens now allows users to search not just static images but videos, understanding moving objects in real-time and offering AI-generated summaries and explanations. This feature is designed to improve interaction and engagement, especially for younger users​. European AI Startup Program Meta, in collaboration with Hugging Face and Scaleway, announced the five winners of its European AI Startup Program. Among the winners is Batisia, a French startup developing AI tools for real estate professionals, and Kodex AI, a German company focusing on automating compliance tasks for financial institutions​. These initiatives demonstrate the expanding AI landscape, with startups across sectors developing niche solutions to transform industries. OpenAI OpenAI launched its Realtime API in public beta. This API allows develo...