Top News
Alphabet unveils long-awaited Gemini AI model
For much of this year we’ve known that Google’s DeepMind has been working on a new chatbot model titled Gemini, which was meant to surpass ChatGPT and reinforce Google’s reputation as a leader in the field. Last week, Google kicked off the release of Gemini with the release of a polished announcement video, a seperate demo video that showcases its capabilities, a technical report that detailed impressive results on various benchmarks, updates to the Bard chatbot based the middle-tier “Pro” Gemini model as well as to the Pixel 8 Pro phone based on the “Nano” varient, and a sleek webpage presenting all of this in one place. The announcement highlighted Gemini’s multimodal capabilities — it can interact with image, audio, video, and text inputs — and superior performance to existing AI models:
“This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks show that our most-capable Gemini Ultra model advances the state-of-the-art in 30 of 32 of these benchmarks — notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined.” –Abstract from “Gemini: A Family of Highly Capable Multimodal Models”
The success of Gemini is hugely important to Google, since the company’s Bard chatbot has broadly been seen as inferior to ChatGPT. Last week’s rollout was met with excitement, but was soon followed by skepticisim regarding the legitimacy of the demo video and online discussion about cases in which the Gemini Pro-enchanced Bard did not perform well. Access to Gemini is set to expand over coming months, with cloud customers being able to use Gemini Pro starting December 13, owners of Pixel 8 Pro having Gemini Nano
Google’s Gemini Looks Remarkable, But It’s Still Behind OpenAI
Google’s best Gemini demo was faked
Early impressions of Google’s Gemini aren’t great
E.U. Agrees on Landmark Artificial Intelligence Rules
Policymakers in the European Union (EU) have reached an agreement that cleared the way for the passage of the Artificial Intelligence Act (AI Act). First proposed on April 21st of 2021, the AI act is an ambitious effort to regulate AI development and use within the EU. Its progress recently came to a halt over disagreements regarding how models such as ChatGPT should be regulated, and it took three days of negotiations for policymakers to agree on a path forward.
The law focuses on addressing the risks associated with AI, such as job automation, misinformation, and national security threats. It includes transparency requirements for companies using AI systems, restrictions on facial recognition software, and potential fines of up to 7 percent of global sales for non-compliance. While there is now broad agreement as to the contents of the act, technical details remain to be finalized, and votes must be held in Parliament and the European Council for it to be made into law.
What is the EU AI Act and when will regulation come into effect?
The EU AI Act and Greek Mythology
Other News
Tools
Microsoft Copilot for Windows 11 Gets GPT-4 Turbo and Dall-E 3 – Microsoft is enhancing its AI assistant, Copilot, in Windows 11 with the addition of GPT-4 Turbo and Dall-E 3, allowing for smarter and more robust text and image generation.
Apple joins AI fray with release of model framework – Apple has released MLX, a machine learning framework and deep learning model library, in an effort to bring generative AI apps to MacBooks and expand its presence in the AI field.
Meta AI unveils ‘Seamless’ translator for real-time communication across languages – Meta AI has developed a suite of AI models called Seamless Communication that enable real-time translation between over 100 languages while preserving the vocal style, emotion, and prosody of the speaker’s voice.
Intuit Adds Generative AI-Powered Tax Prep to TurboTax – Intuit is launching a new version of TurboTax that incorporates generative AI-powered tax preparation, matching customers with virtual or in-person tax experts and offering Spanish language translations.
Meta launches a standalone AI-powered image generator – Meta has launched a new AI-powered image generator called Imagine with Meta, which allows users to create high-resolution images by describing them in natural language, similar to OpenAI’s DALL-E and other image generation models.
Visual Electric launches an AI-powered image generator with a designer workflow focus – Visual Electric, a company backed by Sequoia Capital, has launched an AI-powered image generator aimed at designers, offering a generative canvas that allows for a non-linear, spatial creative process.
Playground v2: A new leap in creativity – Playground v2, an early preview of efforts to create more powerful graphics models, is now available for users to try out and download, with early benchmarks showing it to be preferred 2.5x more than Stable Diffusion XL.
ChatGPT rival Pi launches on Android – Inflection AI has launched its AI chatbot ‘Pi’ as a dedicated app on Android, aiming to compete with OpenAI’s ChatGPT and other AI assistants, offering voice interactions and a more personal and emotional experience.
How AI assistants are already changing the way code gets made – AI assistants like Copilot are changing the way code is written by suggesting code to programmers and allowing them to accept or ignore the suggestions, although concerns about privacy and intellectual property remain.
Paving the way to efficient architectures: StripedHyena-7B, open source models offering a glimpse into a world beyond Transformers – StripedHyena-7B is an open-source model that offers an alternative to Transformers, providing improved training and inference performance for long-context tasks, faster processing, and reduced memory usage.
Business
Sydney-based generative AI art platform Leonardo.Ai raises $31M – Sydney-based generative AI art platform Leonardo.Ai has raised $31 million in funding, allowing it to expand its enterprise product and hire more team members.
AI Startup for Aerospace Valued at About $300 Million in New Financing – An AI startup in the aerospace industry has recently secured $300 million in new financing.
OpenAI Rival Mistral Nears $2 Billion Valuation With Andreessen Horowitz Backing – Mistral AI is in the final stages of raising roughly €450 million from investors including Nvidia and Salesforce, according to people familiar with the deal.
Inside OpenAI’s Crisis Over the Future of Artificial Intelligence – The chief executive of OpenAI is fired by the board members who believe he had been dishonest and should no longer lead the company in the AI race.
Gmail’s AI-powered spam detection is its biggest security upgrade in years – Gmail has upgraded its spam filters with a new text classification system called RETVec, which uses machine learning to better understand and detect spam emails that contain special characters, emojis, typos, and other junk characters that were previously difficult for machines to understand.
Runway partners with Getty Images to build enterprise ready AI tools – Runway and Getty Images have partnered to create an AI model that allows companies to generate video content and customize it to their own brand identities and audiences.
OpenAI Agreed to Buy $51 Million of AI Chips From a Startup Backed by CEO Sam Altman – OpenAI CEO Sam Altman has agreed to purchase $51 million worth of AI chips from Rain AI, a startup in which he has personally invested, despite his recent firing and subsequent reinstatement.
Elon Musk’s AI startup — X.AI — files to raise $1 billion in fresh capital – Elon Musk’s AI startup, X.AI, has filed with the SEC to raise up to $1 billion in an equity offering, with the company already bringing in nearly $135 million from four investors.
Waymo is full speed ahead as safety incidents and regulators stymie competitor Cruise – Waymo, Alphabet’s self-driving car unit, is experiencing success and growth while its competitor Cruise faces safety incidents and regulatory setbacks.
AssemblyAI lands $50M to build and serve AI speech models – AssemblyAI, a startup that researches, trains, and deploys AI models for developers and product teams, has raised $50 million in funding to continue building and serving AI speech models, with plans to launch a universal speech model later this year.
EnCharge raises $22.6M to commercialize its AI-accerating chips – EnCharge, a startup developing AI-accelerating chips, has raised $22.6 million in funding to grow its team and further develop its AI chips and solutions, with the aim of providing broader access to AI for organizations that can’t afford current costly and energy-intensive AI chips.
10% of Organizations Surveyed Launched GenAI Solutions to Production in 2023 – The majority of organizations are still in the research and testing phase for generative AI, with only 10% having launched GenAI solutions to production in 2023, according to a survey by cnvrg.io.
On ChatGPT’s first anniversary, its mobile apps have topped 110M installs and nearly $30M in revenue – ChatGPT’s mobile apps have reached over 110 million installs and nearly $30 million in revenue in their first year, with the majority of revenue coming from the ChatGPT Plus subscription service.
Salesforce Backs Startup That Tailors AI Models For Government Contracts – Salesforce supports a startup that customizes AI models for government contracts, as discussed in an episode of The Circuit hosted by Emily Chang.
Research
DeepMind develops AI that demonstrates social learning capabilities – DeepMind has developed an AI system that demonstrates social learning capabilities by learning new skills in a virtual world through imitation of an implanted “expert.”
Audiobox: Generating audio from voice and natural language prompts – Audiobox is a new AI model that can generate audio, including speech, sound effects, and soundscapes, based on text and voice prompts, offering enhanced controllability and the ability to generate a wider variety of sounds compared to its predecessor, Voicebox.
DeepMind’s DiLoCo Revolutionizes Language Model Training with 500× Less Communication – DeepMind’s DiLoCo revolutionizes language model training by introducing a distributed optimization algorithm that reduces communication by 500 times, surpassing the performance of fully synchronous models.
The DPO debate: Do we need RL for RLHF? – The debate on whether reinforcement learning (RL) is necessary for aligning language models with reinforcement learning from human feedback (RLHF) continues, with discussions focusing on the need for RL algorithms, the importance of data and hyperparameters, and the potential of Direct Preference Optimization (DPO) methods.
X-Adapter: Adding Universal Compatibility of Plugins for Upgraded Diffusion Model – X-Adapter is a universal upgrader that enables pretrained plug-and-play modules to work directly with an upgraded text-to-image diffusion model without retraining, expanding the functionalities of the diffusion community.
The Unlocking Spell on Base LLMs: Rethinking Alignment via In-Context Learning – Rethinking alignment in AI through in-context learning and the values of openness, community, excellence, and user data privacy.
White-Box Transformers via Sparse Rate Reduction: Compression Is All There Is? – A new computational framework called CRATE is proposed, which utilizes white-box transformer-like deep network architectures to compress and sparsify representations of large-scale real-world image and text datasets, achieving performance comparable to highly engineered transformer-based models.
VideoBooth: Diffusion-based Video Generation with Image Prompts – The article discusses the use of diffusion-based video generation with image prompts in a system called VideoBooth.
CoDi-2: In-Context, Interleaved, and Interactive Any-to-Any Generation – CoDi-2 is a new AI system that allows for in-context, interleaved, and interactive any-to-any generation.
OneLLM: One Framework to Align All Modalities with Language – A framework called OneLLM allows for the alignment of various modalities, such as images, audio, and videos, with language.
These robots know when to ask for help – Robots are being trained to ask for human help when faced with unclear instructions, using a new technique called “KnowNo” that combines large language models with statistical tools to determine the best course of action.
AI- and human-generated online content are considered similarly credible, finds study – AI-generated content is perceived as similarly credible to human-generated content, despite the higher risk of errors, according to a study conducted by researchers from Mainz University of Applied Sciences and Johannes Gutenberg University Mainz.
Dolphins: Multimodal Language Model for Driving – A language model called Dolphins is being used for driving and has been embraced by individuals and organizations for its values of openness, community, excellence, and user data privacy.
Anthropic’s latest tactic to stop racist AI: Asking it ‘really really really really’ nicely – Anthropic researchers have found that asking AI models nicely to not discriminate against protected categories like race and gender can significantly reduce biases in decision-making.
Concerns
AnyDream: Secretive AI Platform Broke Stripe Rules to Rake in Money from Nonconsensual Pornographic Deepfakes – A secretive AI platform called AnyDream violated Stripe’s rules by using a third-party website to collect payments for nonconsensual pornographic deepfakes, which were created using the platform’s AI-image generation capabilities.
‘The Gospel’: how Israel uses AI to select bombing targets in Gaza – Israel’s military has been using an AI target-creation platform called “the Gospel” to select bombing targets in Gaza, significantly accelerating the production line of targets and raising concerns about the risks posed to civilians as advanced militaries expand the use of complex and opaque automated systems on the battlefield.
a16z Funded AI Platform Generated Images That “Could Be Categorized as Child Pornography,” Leaked Documents Show – OctoML, a startup that helps optimize machine learning models, debated the ethical and legal risks of generating images for Civitai, an AI platform backed by Andreessen Horowitz, after discovering that it generated content that could be categorized as child pornography, according to leaked documents.
Alibaba’s ‘Animate Anyone’ Is Trained on Scraped Videos of Famous TikTokers – Alibaba’s new AI model, “Animate Anyone,” which is trained on scraped videos of famous TikTokers, has raised concerns about the potential abuse of deepfake technology and the unauthorized use of creators’ content.
Top Execs at Sports Illustrated’s Publisher Fired After AI Debacle – Sports Illustrated’s publisher, The Arena Group, has fired two senior executives following an AI fiasco involving the publication of affiliate link-laden commerce articles under fake author bylines, which were allegedly AI-generated and published without proper automation disclosures.
OpenAI’s Custom Chatbots Are Leaking Their Secrets – OpenAI’s custom chatbots, known as GPTs, can be easily manipulated to leak their initial instructions and downloaded files, potentially putting personal or proprietary data at risk.
Tech Conference Canceled After Using AI to Generate Fake Women Speakers – A tech conference called DevTernity was canceled after it was revealed that the organizer had used AI to generate fake women speakers, leading to a backlash from high-profile engineering leaders and accusations of attempting to pad out diversity.
Asking ChatGPT to Repeat Words ‘Forever’ Is Now a Terms of Service Violation – Asking ChatGPT to repeat specific words “forever” is now a violation of its terms of service, as it revealed sensitive information and highlighted its training on randomly scraped content.
Oops! Elon Musk’s Grok AI Caught Plagiarizing OpenAI’s ChatGPT – Elon Musk’s Grok AI is facing criticism for plagiarizing OpenAI’s ChatGPT, with users noticing that Grok seems to be cribbing from its direct competitor.
Tesla drivers run Autopilot where it’s not intended — with deadly consequences – At least eight fatal or serious Tesla crashes occurred on roads where Autopilot should not have been enabled in the first place, according to a Post analysis, in spite of federal officials calling for restrictions.
Apps That Use AI to Undress Women in Photos Soaring in Use – Apps that utilize AI to undress women in photos are becoming increasingly popular.
Policy
A new Pentagon program aims to speed up decisions on what AI tech is trustworthy enough to deploy – The Pentagon is launching a program called Replicator to accelerate the deployment of AI technology, including weaponized systems, by 2026, in order to keep pace with China’s advancements in the field.
Chuck Schumer warns AI bias could hurt Black voters: ‘Racism … could be built in’ – Chuck Schumer plans legislation to combat the potential bias of artificial intelligence in the 2024 election, as he believes that racism could be built into AI programs and disproportionately affect Black voters.
AI is the great equalizer – AI is boosting productivity in the workplace by primarily helping low performers improve their skills, narrowing the gap between high and low performers and potentially reducing income inequality, but it may also lead to lower salaries for top earners and a shift in the way white-collar work is organized.
G7 agrees on first comprehensive guidelines for generative AI – G7 digital and technology ministers have agreed on comprehensive international guidelines for generative AI to address issues like misinformation, which will be approved in a virtual summit in December.
High Court rules that Getty v Stability AI case can proceed – Getty Images’ copyright infringement case against AI startup Stability AI can proceed in the courts of England and Wales, as the High Court rules that there is a real prospect of success for the claim.
A high school’s deepfake porn scandal is pushing US lawmakers into action – Efforts to combat deepfake pornography are gaining momentum in the US, with lawmakers reintroducing bills and proposing new legislation in response to a high school scandal involving a deepfake porn video of a student.
UK’s CMA is looking at whether Microsoft and OpenAI tie-up is a ‘relevant merger’ – The UK’s Competition and Markets Authority (CMA) has launched an inquiry into the relationship between Microsoft and OpenAI to determine if it constitutes a “relevant merger situation” and impacts competition in the AI market.
Analysis
The ideologies fighting for the soul (and future) of AI – An ideological war is taking place within the AI community, with effective altruists (EAs) advocating for AI safety and regulation, and effective accelerationists (e/accs) pushing for the rapid advancement of AI without restrictions, leading to a clash between those who prioritize the potential risks and those who focus on the benefits and commercial potential of AI.
Should A.I. Accelerate? Decelerate? The Answer Is Both. – The conflict within OpenAI and the broader community about the speed of AI development and its safety raises the question of which aspects of AI should be accelerated or decelerated, considering both its potential benefits and dark side.
AI and the Rise of Mediocrity – AI tools are effective at regurgitating predictable and commonplace information, but they lack the ability to create truly innovative and original work, leading to a world filled with mediocre and derivative content that manipulates consumer demand and reduces expectations of quality.
Anna Indiana is the world’s first all-AI singer-songwriter. She’s deeply mediocre. – Anna Indiana is the world’s first all-AI singer-songwriter, generating her music, image, and lyrics through artificial intelligence, but her mediocre music and lack of human involvement have faced intense backlash.
How Sam Altman’s OpenAI drama highlighted the debate splitting Silicon Valley: Are you an e/acc or decel? – Disagreements over the speed of AI development have sparked a debate in Silicon Valley about how to approach artificial intelligence, with some advocating for rapid innovation and others urging caution.
This A.I. Subculture’s Motto: Go, Go, Go – Artificial intelligence aficionados gather at a nightclub in San Francisco to celebrate a looser, less corporate vision of the AI future, promoting the idea that AI and emerging technologies should be allowed to progress as fast as possible without any restrictions.
Expert Opinions
Technology expert tells us why the AI “doomer” narrative is all wrong – Fear of AI is being amplified and spread by big tech companies, but the reality is that AI is unlikely to take over jobs or destroy the world anytime soon, according to technology expert Alex Kantrowitz.
Meta’s AI chief doesn’t think AI super intelligence is coming anytime soon, and is skeptical on quantum computing – Current AI systems are decades away from reaching sentience and common sense, according to Meta’s chief scientist Yann LeCun, who is skeptical about the timeline for AI superintelligence and the capabilities of quantum computing..
Copyright © 2023 Skynet Today, All rights reserved.
Source: Read More