Former Amazon executive David Limp demonstrated a new generative AI-powered version of Alexa in September 2023, but the new Alexa was not ready for a prime time rollout. The Alexa large language model (LLM) that powers the new Alexa is far from state-of-the-art and lacks data and access to specialized computer chips. Amazon has deprioritized the new Alexa in favor of building generative AI for its cloud computing unit, AWS. The organizational structure and siloed business units within Amazon have hampered the development of the new Alexa. The new Alexa LLM has not met the go/no-go criteria for a widespread rollout due to concerns about safety and lack of data. Amazon's internal bureaucracy and infrastructure issues have slowed down the development of the new Alexa. Many former employees are skeptical that the new Alexa will ever be ready or that it will be able to compete with nimbler competitors like OpenAI. Despite challenges, Amazon remains committed to delivering a generative AI Alexa.
Former Senior Machine Learning Scientist at Alexa AI, Mihail Eric, details the challenges faced in integrating advanced AI into Amazon's Alexa. He criticizes the fragmented organizational structure, poor data access, and lack of collaboration. A Fortune report highlights further struggles, including Alexa's design limitations and the lack of a clear monetization strategy. The Alexa AI team also faces obstacles in accessing Anthropic's Claude AI model. These challenges hinder the transformation of Alexa into a more sophisticated and profitable AI assistant. [75ad874e] [1d57c36b]
Former senior machine learning scientist Mihail Eric shared his perspectives on why Amazon's Alexa fell behind in the AI competition. He cited mismanagement and mislabeling of data, difficulty accessing and utilizing internal data, fragmented organizational structure, misalignment between product goals and scientific research, and conflicts over success metrics as primary reasons for Alexa's lag. Eric highlighted the cumbersome process of rectifying mislabeled data and the lack of collaboration due to decentralized teams. He also mentioned the tension between customer-oriented results and experimental projects, leading to constant conflict and project abandonment. [19838cb4]
Amazon Web Services (AWS) has announced new tools to make generative AI applications easier to create and improve their accuracy. The additions include contextual grounding checks, which evaluate AI-generated answers in real time by cross-referencing source material, and a Guardrails API that evaluates user prompt inputs and AI model responses. The API also helps identify and filter sensitive or inappropriate content. In tests, contextual grounding checks found and filtered up to 75% of hallucinations in AI model responses and blocked up to 85% more content when used with the Guardrails API. Other capabilities announced at the AWS Summit NY include the AWS App Studio for creating AI apps from text prompts and the expansion of Amazon Q Apps for building custom AI apps. [93c1522c]
Apple may try to monetize its Apple Intelligence (A.I.) at some point in the future as the company looks for new revenue streams. The long-term plan is to make money off the capabilities by charging for top-tier A.I. features, similar to how it charges for some iCloud features. Apple could potentially launch a paid service called 'Apple Intelligence+' with extra features that users pay monthly fees for. Additionally, Apple will receive a cut of the subscription revenue from every AI partner it brings onboard. This move would make Apple less reliant on hardware upgrades and more profitable from AI. Apple has been successful in increasing its services revenue, and its integration of AI has demonstrated real-world benefits. The company aims to open up a new revenue stream as its A.I. features continue to evolve and offer consumers an improved experience. [ce8cebff]
Bank of America has analyzed the potential financial impact of a paid tier for Amazon's Alexa devices. Assuming 10% of active users are willing to pay for an AI-supercharged Alexa, a $5/month tier could represent a $600 million incremental revenue opportunity. This would translate to 1.3 percentage points of growth to the 2025 subscription sales estimate at $49 billion. The Alexa fee initiative is seen as critical for Amazon's goal of turning the Alexa division into a profitable one. Amazon plans to stop paying developers to create apps for Alexa at the end of June. Bank of America has a Buy rating on Amazon with a price objective of $210. Shares of Amazon are up 0.13% in premarket trading and 24% year-to-date. [0f61d2f1] [6de295bc]
Apple is considering launching a paid service called 'Apple Intelligence+' to monetize its AI capabilities. The service would offer extra features that users would pay a monthly fee for, similar to how Apple charges for some iCloud features. Apple also plans to receive a portion of the subscription revenue from AI partners. This move would help Apple diversify its revenue streams and rely less on hardware upgrades. The company has seen success in increasing its services revenue and aims to continue this trend with its evolving AI features. [ce8cebff]
Apple is considering a subscription-based model for advanced features of its AI assistant, Apple Intelligence. The core functionalities of Apple Intelligence will remain free, but the potential premium tier, called 'Apple Intelligence+,' would offer additional features and services. The free version of Apple Intelligence is set to debut in fall 2024, bringing enhanced capabilities to Siri and potentially new features like photo organization and summarization. Apple Intelligence+ could offer more sophisticated AI tools, such as advanced language models and personalized coaching services. It would follow Apple's existing model for services like iCloud+, where users pay a monthly fee for additional storage and features. The motivation behind introducing a paid tier is to provide a new revenue stream and offset the costs of developing and maintaining advanced AI technologies. The success of Apple Intelligence+ would depend on its value proposition and differentiation from the free version. Apple has not officially announced Apple Intelligence+ or provided details about its features, pricing, and release date. [123d3c17]
AI models in development today can cost up to $1 billion to train, with expectations that the cost will increase to $10 or even $100 billion in the next three years. Anthropic CEO Dario Amodei predicts that by 2025-2027, models could be better than most humans at most things. The development of AI from generative artificial intelligence (like ChatGPT) to artificial general intelligence (AGI) will be a gradual process. Hardware requirements for training AI models are expected to grow ten times more powerful each year. The estimated power consumption of all data center GPUs sold in 2023 could power 1.3 million homes. AI firms need to consider power supply and related infrastructure, including modular nuclear power for data centers. Elon Musk plans to purchase 300,000 B200 AI chips. OpenAI and Microsoft are reportedly planning a $100 billion AI data center. The future impact of AI on society remains a question. [c89421a3]
Nigerian startup NeuRaL AI has launched REACTOR, a composable data and AI compute infrastructure that simplifies the use of Generative AI (GenAI) for businesses. REACTOR eliminates the need for technical setups or complex infrastructure, self-hosting, managing machine learning pipelines, caching, or query engines. It provides access to a dedicated 'Metal Vault' within its virtual compute cluster data center and offers a user-friendly interface. The cost of using REACTOR starts at N99k per month. Developers and software companies can leverage REACTOR to host their custom models trained on their own data for a monthly fee of 99k. NeuRaL AI is also offering a free 1-hour GenAI adoption workshop for companies based in Lagos and virtual sessions for companies outside Lagos. [9c5f781e]
IBM is set to launch a GenAI in Kochi on July 10, 2024. Microsoft is testing AI radio on YouTube Music. The article also mentions various IT topics covered by IT Voice, including PC & Laptops, Visual Reality (AR, VR, MR), and e-Publishing. The article is published by IT Voice, an online media platform that focuses on providing news and updates related to the IT industry.
Advanced Micro Devices (AMD) has announced the acquisition of Silo AI, Europe's largest private AI lab, in a $665 million all-cash transaction. The deal is expected to close in the second half of 2024, pending regulatory approvals. The acquisition aims to accelerate AMD's development and deployment of AI models and software solutions, enhancing its open-source AI software capabilities. Silo AI, based in Helsinki, Finland, specializes in end-to-end AI-driven solutions and serves customers such as Allianz, Philips, Rolls-Royce, and Unilever. The company has also developed open-source multilingual large language models (LLMs) on AMD platforms. This acquisition is part of AMD's strategy to expand its AI ecosystem and deliver end-to-end AI solutions based on open standards. Silo AI's CEO and co-founder, Peter Sarlin, will continue to lead the team as part of AMD's Artificial Intelligence Group. [829ed975] [decec159]
ELLIS Unit Warsaw, a new unit of the pan-European Laboratory for Learning and Intelligent Systems (ELLIS), has opened in Warsaw with the goal of improving the energy efficiency of AI and machine learning systems. The unit aims to set standards for energy efficiency and develop taxonomy for AI models. It is based at IDEAS NCBR, an AI centre set up by the National Centre for Research and Development in Poland. The director of ELLIS Unit Warsaw, Tomasz Trzciński, wants the unit to become a standard-setter for energy efficiency and model taxonomy. The unit plans to increase awareness of the resources consumed by large AI models and deliver improvements to stimulate adoption of more efficient models. It is also collaborating with the European Lighthouse of AI for Sustainability and implementing the EXplainable ALgorithmic Tools (EXALT) project. ELLIS Unit Warsaw is the 41st unit in the ELLIS AI excellence network and aims to bridge academia and industry. It is classified as an SME and focuses on applied research. The unit is looking for funding and access to data from companies for applied research and aims to attract researchers from outside Poland in the future. [984e9ea7]