Last Week in AI #248: ChatGPT's 1st birthday, Amazon's new AI chatbot Q, the energy cost of generative AI, Sam Altman makes statement after return, and more!
How the world has changed after 1 year of ChatGPT, Amazon releases Q chatbot for companies, generative AI requires heavy energy usage, OpenAI's board to include Microsoft as non-voting observer
ChatGPT, the AI-powered language model developed by OpenAI, is celebrating its one-year anniversary. Since its launch, ChatGPT has become the fastest-growing consumer application in history, allowing users to engage in interactive and conversational dialogue with the model. The tool has received both praise and criticism, with some hailing it as a disruptive technology and others calling it unreliable. OpenAI has released updates to the model, including the GPT-4, which offers improved problem-solving abilities and the ability to handle longer-form content. ChatGPT has also faced competition from other companies, such as Anthropic and Google DeepMind. OpenAI has expanded the capabilities of ChatGPT by introducing plugins and giving it access to the internet for up-to-date information. Despite some controversies and leadership changes at OpenAI, ChatGPT has made a significant impact in the past year and is poised to continue evolving in the future.
Amazon has entered the AI chatbot field with "Amazon Q," designed by its cloud division, AWS, to assist in the workplace by streamlining tasks such as summarizing documents and providing company policy information. Unlike consumer-oriented AI solutions, Q is built on the Bedrock platform, integrating multiple AI systems, crafted to align with established enterprise security protocols. Offered to businesses at a competitive pricing model, Q represents Amazon's effort to catch up with tech giants like Microsoft and Google in the AI sphere, which was highlighted as part of a series of AI-focused announcements at its AWS conference. However, soon after its launch, internal concerns arose about Q's reliability, with reports of the chatbot hallucinating and potentially exposing confidential data during its public preview—a situation categorized by employees as severe but downplayed by Amazon as a lack of identifying security issues and a part of the feedback process for improvement.
A recent study by Hugging Face and Carnegie Mellon University, discussed in an article by Melissa Heikkilä, has found that using AI to perform various tasks such as generating images or writing emails incurs a significant energy and carbon cost. Generating one image using a powerful AI model can consume as much energy as charging a smartphone, while text generation is less energy-intensive. The research, still awaiting peer review, emphasizes that while training large models is energy-consuming, the usage of these models contributes the most to their carbon footprint. The study calculated the emissions from different AI tasks across multiple models, highlighting that image generation is notably more energy and carbon-intensive than text generation. Generative AI models, which perform various tasks simultaneously, were found to use much more energy than specialized models designed for specific tasks. The findings suggest a need to reconsider the use of generative AI in favor of less carbon-intensive alternatives and have sparked discussions about the sustainability and carbon accountability of AI technologies.
In his first statement after returning as OpenAI’s CEO, Sam Altman expressed his excitement and confidence in the company's trajectory and its future success in achieving its mission despite recent uncertain times. He acknowledged the resilience of the team, thanked various individuals including Ilya, who although no longer on the board, might still contribute to OpenAI's work, as well as partners, users, and the previous interim board members for guiding the organization through this period. Altman highlighted the company's immediate priorities which include solidifying governance structures with oversight from the newly formed initial board consisting of Bret Taylor, Larry Summers, and Adam D’Angelo. Notably, Microsoft will have a non-voting observer role on OpenAI’s new board.
Stable Diffusion Turbo XL can generate AI images as fast as you can type - Stability AI has launched Stable Diffusion XL Turbo, an AI image-synthesis model that can rapidly generate imagery based on a written prompt, using a technique called Adversarial Diffusion Distillation (ADD) to produce image outputs in a single step, significantly reducing the time required for image generation.
Microsoft Paint’s DALL-E 3 integration is rolling out on Windows 11 - Microsoft has rolled out the highly-anticipated Paint's DALL-E integration on Windows 11, allowing users to create AI-generated images using the power of OpenAI's ChatGPT through a "Cocreator" button in Paint.
Amazon finally releases its own AI-powered image generator - Amazon has released its own AI-powered image generator called Titan Image Generator, which can create new images or customize existing ones, and includes built-in mitigations for toxicity and bias, as well as tamper-resistant invisible watermarks to combat AI-generated misinformation and abuse imagery.
Mastercard launches Shopping Muse, an AI to help consumers find the perfect gift - Mastercard has launched Shopping Muse, an AI-powered service that provides personalized gift recommendations based on individual consumer profiles and preferences, aiming to save time and effort for consumers and boost sales and revenue for retailers.
Bitmagic launches public test for AI-based tool to create games with text prompts - Bitmagic has launched a generative AI-based game development platform that allows gamers to create fully playable games by typing a description into a text prompt, with the platform producing the game based on the words.
Voicemod will now let you create and share your own AI voices - Voicemod has launched new features that allow users to create and share their own bespoke AI voices, offering a wide range of options for customization and fine-tuning.
Perplexity AI Introduces New Online LLMs for Real-Time Information Access - Perplexity AI has introduced two new online models that tap into real-time information from the internet, allowing them to deliver accurate responses to time-sensitive queries, surpassing the capabilities of leading proprietary models.
OpenAI delays launch of custom GPT store to early 2024 - OpenAI has delayed the launch of its custom GPT store until early 2024, citing the need for improvements based on customer feedback.
Using Midjourney "Blend" for Fun and Profit. (But Mostly for Fun.) - The Midjourney /blend feature allows users to blend two or more images together for fun and creative purposes, such as visualizing someone in a movie role, virtual try-ons, style transfer, creating hybrid creations, and making memes.
How Jensen Huang’s Nvidia Is Powering the A.I. Revolution - The revelation that ChatGPT, an AI chatbot, was trained on an Nvidia supercomputer led to a significant increase in Nvidia's stock value, making it one of the most valuable corporations in the world, and highlighting the company's role as a major player in the AI industry.
Broadcom adds silicon AI features to speed new Trident networking chip - Broadcom has added artificial intelligence features to its new Trident networking chip, aiming to improve efficiency and help move information around data centers more effectively.
Amazon and Salesforce Expand Partnership to Add New AI Capabilities - Salesforce and Amazon Web Services (AWS) have announced a significant AI-focused expansion of their partnership, making it easier for customers to manage their data across Salesforce and AWS and infuse the latest generative AI technologies into their applications and workflows.
The Inside Story of Microsoft’s Partnership with OpenAI - Microsoft's partnership with OpenAI has been seen as a huge victory for the company and a strong endorsement of its approach to developing AI, with OpenAI executives choosing Microsoft as the best and safest place to continue their work.
Cradle’s AI-powered protein programming platform levels up with $24M in new funding - Biotech and AI startup Cradle has secured $24 million in funding for its generative approach to protein design, which has attracted major drug development companies like Johnson & Johnson and Novozymes, and has the potential to significantly reduce the time and number of experiments required for protein development.
Pika Labs, which is building AI tools to generate and edit videos, raises $55M - Pika Labs, a startup building AI tools for video editing and generation, has raised $55 million in funding and launched its Pika 1.0 suite of videography tools, which includes a generative AI model capable of editing videos in various styles.
Microsoft's $3.2 bln UK investment to drive AI growth - Microsoft plans to invest $3.2 billion in the UK over the next three years, with a focus on artificial intelligence (AI) infrastructure and development, in order to drive future growth in the country.
Indian AI Video Startup Rephrase.ai Announces Acquisition by Adobe - Adobe is reportedly acquiring Indian AI video startup Rephrase.ai, expanding its generative AI capabilities and positioning itself in the growing text-to-video AI generation space.
Meet the first Spanish AI model earning up to €10,000 per month - A Spanish agency has created Aitana, the first Spanish model created by artificial intelligence, who earns up to €10,000 per month and has gained over 121,000 followers on Instagram, receiving private messages from celebrities who are unaware she is not a real person.
Together lands $102.5M investment to grow its cloud for training generative AI - Together, a startup creating open source generative AI and AI model development infrastructure, has closed a $102.5 million Series A funding round to expand its cloud platform and provide developers with the ability to build on open and custom AI models.
Interview: Sam Altman on being fired and rehired by OpenAI - Sam Altman discusses his initial reaction to being fired and then rehired by OpenAI, expressing his love for the company and its mission of safe and beneficial AGI, while also acknowledging the need for an independent review and not interfering with the process.
Report: Stability AI Positioning Itself for Acquisition - Stability AI, a British AI startup, is considering selling the company due to pressure from investors over its financial position, engaging in preliminary discussions with potential buyers.
GM to cut spending on Cruise after accident - FT - General Motors is cutting spending on its self-driving unit, Cruise, following a pedestrian accident last month.
AI under the microscope: the algorithms powering the search for cells - Researchers are using deep learning algorithms to develop computational tools that can segment cellular and subcellular features in biological imaging experiments, with the aim of reducing the time it takes to analyze data sets.
📄 NeurIPS 2023 Primer - The NeurIPS 2023 conference will feature 3586 papers, with a focus on natural language processing (NLP) and topics such as large language models, learning from human feedback, LLM alignment, LLM pre-training, LLM fine-tuning, emergent abilities and in-context learning, and reasoning.
AI tech helps in early detection of pancreatic cancer - A deep learning approach using AI technology developed by Alibaba Group's Damo Academy has been successful in detecting and classifying pancreatic lesions with high accuracy via non-contrast computed tomography, potentially serving as a new tool for large-scale pancreatic cancer screening.
Google DeepMind AI reveals potential for thousands of new materials - Google DeepMind has used artificial intelligence (AI) to predict the structure of over 2 million new materials, potentially improving real-world technologies such as batteries, solar panels, and computer chips.
Stanford Researchers Introduce BLASTNet: The First Large Machine Learning Dataset for Fundamental Fluid Dynamics - Stanford researchers have introduced BLASTNet-2, a revolutionary dataset for machine learning in fluid dynamics, which promises to revolutionize the understanding and application of fundamental fluid dynamics in various fields.
New method uses crowdsourced feedback to help train robots - Researchers from MIT, Harvard University, and the University of Washington have developed a new reinforcement learning approach that leverages crowdsourced feedback to guide AI agents as they learn new tasks, eliminating the need for expertly designed reward functions and allowing for faster learning.
Starling-7B: Increasing LLM Helpfulness & Harmlessness with RLAIF - Starling-7B is an open large language model trained using Reinforcement Learning from AI Feedback (RLAIF), which outperforms other models on MT Bench and AlpacaEval, and is released along with the Nectar dataset and Starling-RM-7B-alpha reward model.
Scalable Extraction of Training Data from (Production) Language Models - The article discusses the scalable extraction of training data from language models, emphasizing the importance of openness, community, excellence, and user data privacy.
Researchers quantify the carbon footprint of generating AI images - Generating images using artificial intelligence has a carbon footprint equivalent to charging a smartphone, while generating text requires much less energy, according to a study by researchers at Hugging Face and Carnegie Mellon University.
MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI - A new benchmark called MMMU is introduced for expert AGI, allowing for multi-discipline multimodal understanding and reasoning.
UK school pupils ‘using AI to create indecent imagery of other children’ - Children in UK schools are using artificial intelligence (AI) to create realistic and illegal indecent images of other children, raising concerns among experts on child abuse and technology.
ChatGPT Can Reveal Personal Information From Real People, Google Researchers Show - Google researchers have discovered that OpenAI's AI chatbot, ChatGPT, can be manipulated to reveal personal information from real people, including names, email addresses, and phone numbers, by using specific keywords to force the chatbot to divulge its training data.
Sports Illustrated Published Articles by Fake, AI-Generated Writers - Sports Illustrated published articles by fake, AI-generated writers, with AI-generated headshots and biographies, and some of the articles themselves were also churned out using AI, according to anonymous sources involved in the creation of the content; after the magazine's publisher was contacted with questions, all the AI-generated authors disappeared from the site without explanation.
Microsoft president says no chance of super-intelligent AI soon - Microsoft president dismisses the possibility of super-intelligent AI in the next 12 months, but emphasizes the need for safety measures in AI systems.
Ego, Fear and Money: How the A.I. Fuse Was Lit - The debate over whether artificial intelligence will elevate the world or destroy it has pitted some of the world's richest men against each other, as they fight for a piece of the business and the power to shape it.
Putin to boost AI work in Russia to fight a Western monopoly he says is ‘unacceptable and dangerous’ - Putin announces plan to endorse a national strategy for the development of AI in Russia, emphasizing the need to prevent a Western monopoly and ensure the use of Russian solutions in creating reliable and transparent AI systems that are safe for humans.
US, Britain, other countries ink agreement to make AI 'secure by design' - The United States, Britain, and other countries have signed an agreement to ensure that artificial intelligence (AI) systems are "secure by design," with guidelines including monitoring for abuse, protecting data, and vetting software suppliers.
Inside U.S. Efforts to Untangle an A.I. Giant’s Ties to China - American officials are concerned that G42, an artificial intelligence firm controlled by the United Arab Emirates, is hiding its extensive work with China and could potentially be a conduit for the transfer of advanced American technology and genetic data to the Chinese government.
Meta updates political advertising rules to cover AI-generated images and videos - Meta has updated its policies on political ads to require advertisers to disclose when they use AI to alter images and videos in certain political ads.
Deepfake-Chancellor announces ban of Nazi-party - An art collective in Germany has created a fake video of Chancellor Olaf Scholz calling for a ban on the right-wing extremist party AFD, using AI-generated voices and real evidence of extremist statements by AFD politicians, as part of an art installation and to raise questions about the use of AI in propaganda.
Generative AI a stumbling block in EU legislation talks -sources - EU lawmakers are struggling to agree on how to regulate generative AI systems like ChatGPT, posing a threat to landmark legislation aimed at controlling artificial intelligence.
Brazilian city enacts an ordinance that was secretly written by ChatGPT - City lawmakers in Brazil have enacted what appears to be the nation’s first legislation written entirely by artificial intelligence, sparking objections and raising questions about the role of AI in public policy.
Copyright © 2023 Skynet Today, All rights reserved.