Discover more from Last Week in AI

Weekly text and audio summaries of the most interesting AI news, as well as editorials commenting on recent events

Over 23,000 subscribers

Last Week in AI #244: Biden Issues Executive Order to Create A.I. Safeguards, global AI risk management at UK summit, OpenAI's ChatGPT goes 'Voltron-style', Musk's Grok bot spices up AI, and more!

At UK's AI Summit developers and govts agree on testing to help manage risks, ChatGPT is combining its different abilities into a single chat, Elon Musk debuts 'Grok' AI bot to rival ChatGPT

Last Week in AI

Nov 6, 2023

Biden Issues Executive Order to Create A.I. Safeguards

President Biden has signed an executive order on artificial intelligence (AI) that requires companies to report to the federal government about the risks their systems pose in aiding countries or terrorists in creating weapons of mass destruction. The order also aims to address the dangers of "deep fakes" that can manipulate audio and video to spread fake news, commit fraud, or smear reputations. Biden expressed concern about the potential for fraudsters to manipulate a person's voice and create a viral statement that never happened, leading to political or national security crises. The order demonstrates the US's commitment to leading in AI regulation, as Europe is already moving forward with its own rules. Vice President Kamala Harris will represent the US at an international conference on AI regulation in Britain.

At UK's AI Summit developers and govts agree on testing to help manage risks

Leading AI developers and governments have agreed to collaborate on testing new AI models before their release to manage the risks associated with the technology. The agreement was reached at the UK's artificial intelligence summit, where political leaders from the US, EU, and China signed a declaration to share a common approach to identifying and mitigating risks. The summit also saw the establishment of an international panel on risk and the delivery of a "State of the Science" report by AI expert Yoshua Bengio. The aim is to rigorously assess AI models before and after deployment to ensure safety. The summit brought together politicians, academics, and tech executives to discuss the future of AI and establish global oversight.

ChatGPT is combining its different abilities into a single ‘Voltron-style’ chat

OpenAI has released an update to its AI chatbot, ChatGPT, that combines multiple capabilities into a single chat session. Previously, users had to toggle each capability independently, but now they can analyze and answer questions about documents, browse the web, perform data analysis, and generate images using OpenAI's DALL-E model all within the same session. This update improves the efficiency of the service and has been well-received by users. However, some have raised security concerns about the combination of browsing and code interpretation capabilities, as it could potentially be exploited for malicious purposes. OpenAI has not yet formally announced the update.

Elon Musk debuts 'Grok' AI bot to rival ChatGPT, others

Elon Musk has unveiled Grok, an AI bot developed by his new company xAI. Grok is designed to answer "spicy questions" and has a rebellious and witty personality. It has access to data from X, giving it an advantage over other AI models. However, xAI acknowledges that Grok can still generate false or contradictory information. The prototype is currently in beta testing and will eventually be available as a feature of X Premium+. In initial tests, Grok outperformed other models in its compute class but was surpassed by bots with larger data sets. The goal of xAI is to build AI that advances our collective understanding of the universe and is accessible to people of all backgrounds and political views.

Other News

Tools

Chinese tech giant Alibaba launches upgraded AI model to challenge Microsoft, Amazon - Alibaba has launched Tongyi Qianwen 2.0, an upgraded AI model that demonstrates remarkable capabilities in understanding complex instructions, copywriting, reasoning, memorizing, and preventing hallucinations, as the Chinese tech giant aims to compete with Amazon and Microsoft.

MetNet-3: A state-of-the-art neural weather model available in Google products - Google has developed a new weather model called MetNet-3, which uses direct observations of the atmosphere to provide high-resolution weather forecasts up to 24 hours in advance, outperforming traditional physics-based models, and has integrated these forecasts into various Google products and technologies.

Teachers in India help Microsoft Research design AI tool for creating great classroom content - Teachers in India are collaborating with Microsoft Research to design an AI tool called Shiksha copilot, which uses generative AI to help teachers quickly develop personalized learning experiences, design assignments, create hands-on activities, and more, saving them hours of time spent on daily planning.

Runway’s Gen-2 update is blowing people’s minds with incredible AI video - Runway's Gen-2 update for its AI video model has impressed users with its improved fidelity and consistency, leading to claims that it is a game-changer in generative AI.

ChatGPT Plus members can upload and analyze files in the latest beta - ChatGPT Plus members now have the ability to upload and analyze files, as well as access multimodal support, in the latest beta update.

New Maps updates: Immersive View for routes and other AI features - Google Maps is introducing Immersive View for routes, an AI-powered feature that allows users to preview their journey in a multidimensional view with detailed turn-by-turn directions, simulated traffic, and weather conditions.

LinkedIn's new AI chatbot wants to help you get a job - LinkedIn has introduced an AI-powered chatbot called the "job seeker coach" to help users with job applications and has also unveiled other generative AI tools for Premium members.

Brave responds to Bing and ChatGPT with a new ‘anonymous and secure’ AI chatbot - Brave, the privacy-focused browser, has launched Leo, an AI chatbot that prioritizes user privacy and offers features similar to other chatbots like Bing Chat and Google Bard, but without recording conversations or requiring login information.

Business

Defense Tech Startup Shield AI Raises $200M At $2.7B Valuation - Defense tech startup Shield AI has raised $200 million in a Series F funding round, bringing its valuation to $2.7 billion, as it continues to develop AI pilots for autonomous aircraft.

Palantir Surges After AI Demand Powers Record Quarterly Profit - Palantir Technologies Inc. reported its highest earnings since its founding 20 years ago, driven by strong demand for its new artificial intelligence offerings, leading to a surge in its stock price.

Aurora, With New Autonomous-Truck Terminal, On Track for Commercial Launch - Aurora Innovation is on track for the commercial launch of its autonomous-truck program by the end of 2024, with the opening of a new terminal in Houston that will support driverless trucks on a lane between Dallas and Houston.

Britain to invest 300 million pounds in AI supercomputing - Britain plans to invest 300 million pounds in two supercomputers to support research into making advanced AI models safe, as well as drive breakthroughs in drug discovery and clean energy.

AMD gives soft fourth-quarter guidance, but expects to sell $2 billion of AI chips next year - AMD reported third-quarter earnings that beat expectations, but issued a weaker-than-expected forecast, while also announcing plans to sell $2 billion worth of AI chips next year.

Hiring Data Shows Why San Francisco Is AI Capital of the World - San Francisco is the AI capital of the world, with around 22% of AI job postings coming from companies headquartered in the city, and when factoring in the broader Bay Area, the proportion of AI job postings balloons to around 59%, more than every other city in the country combined.

Quora’s Poe introduces an AI chatbot creator economy - Quora's AI chatbot platform Poe is introducing a creator monetization program, allowing bot creators to generate income through subscriptions and per-message fees.

Research

A glimpse of the next generation of AlphaFold - The latest generation of AlphaFold, developed by Google DeepMind and Isomorphic Labs, has significantly improved accuracy and expanded coverage to include ligands and other biological molecules, revolutionizing the understanding of proteins and their interactions.

AI can catalogue a forest’s inhabitants simply by listening - Researchers have developed an artificial intelligence system that can identify bird species in the rainforest by analyzing their calls, potentially making it easier and faster to measure biodiversity in a given area.

Say what: AI can diagnose type 2 diabetes in 10 seconds from your voice - An AI model trained by Canadian researchers can diagnose type 2 diabetes in patients after listening to their voice for just six to 10 seconds, potentially revolutionizing the way the medical community screens for the condition.

HyperFields: Towards Zero-Shot Generation of NeRFs from Text - A new technique called HyperFields shows promising progress in generating detailed 3D models directly from text prompts, using a dynamic hypernetwork to predict weights for a 3D generation network, resulting in faster convergence time and the ability to adapt to new text prompts, although it has limitations in fine details.

RT-Sketch: Goal-Conditioned Imitation Learning from Hand-Drawn Sketches - RT-Sketch is a goal-conditioned imitation learning model that uses a sketch-conditioned IL policy trained on a dataset of trajectories paired with sketches representing the goal state, generated using an image-to-sketch GAN translation network and augmented with various colorspace and affine transforms.

RT-Trajectory: Robotic Task Generalization via Hindsight Trajectory Sketches - Language-conditioned policies like RT-1 struggle to generalize to new scenarios that require extrapolation of language specifications, as demonstrated in the article through the use of robotic task generalization via hindsight trajectory sketches.

Self-Critique, Self-RAG, NEFTune, Safe RLHF and More - Recent AI research has focused on analyzing the capabilities of large language models (LLMs), improving the alignment process, and making practical tools for LLMs more effective, including self-critique, self-RAG, NEFTune, safe RLHF, and more.

Learning From Mistakes Makes LLM Better Reasoner - The Librarian Bot found similar papers and recommends them using the Semantic Scholar API.

Codebook Features: Sparse and Discrete Interpretability for Neural Networks - Codebook features introduce sparse and discrete hidden states to neural networks, enabling improved interpretability and control, as demonstrated through experiments with sequence modeling and language models.

Personas as a Way to Model Truthfulness in Language Models - Using personas can be an effective way to model truthfulness in language models, as it helps to ensure openness, community, excellence, and user data privacy.

Lost in DALL-E 3 Translation - DALL-E 3, the latest AI image generation model from OpenAI, incorporates automatic prompt transformations that modify the original prompt into a more descriptive version, but also translates all prompts into English, resulting in biased and stereotypical image generation across different languages.

Concerns

Scarlett Johansson takes legal action against use of image for AI - Scarlett Johansson has taken legal action against an AI app that used her name and likeness in an AI-generated advertisement without her permission, joining other actors and authors who have spoken out against the unauthorized use of their images in AI.

Managing AI Risks in an Era of Rapid Progress - The rapid progress of AI systems poses large-scale risks, including social injustice, criminal activities, and loss of human control, and urgent priorities for AI research and governance are needed to mitigate these risks.

Cruise Grew Fast and Angered Regulators. Now It’s Dealing With the Fallout. - Cruise, a self-driving car company, is facing safety concerns and skepticism about its management and viability after a driverless taxi hit and severely injured a woman in San Francisco.

The Finals Criticised by Actors and Devs for Using AI-Generated Voice Overs - The game "The Finals" has faced criticism for using AI-generated voice overs instead of casting actors, with voice actors expressing concerns about the impact on their livelihoods and the quality of the AI-generated voices.

Image of Palestinian carrying children out of rubble shows signs of AI - An image of a man carrying children through rubble, which has been widely shared on social media in connection to the bombing of Gaza, shows signs of artificial intelligence manipulation, according to experts.

Justine Bateman's Fight Against Generative AI In Hollywood - Justine Bateman is taking a stand against generative AI in Hollywood, as discussed in this episode of the Fixing the Future podcast.

Analysis

Do You Want Artificial General Intelligence? - The article discusses the uncertainty surrounding artificial general intelligence (AGI) and questions whether the world actually wants AGI at all.

The Secret Sauce Behind DALL-E 3 : How Is It So Great At Following Instructions? - DALL-E 3, a text-to-image model, stands out for its exceptional prompt adherence due to the use of better image captions in its training dataset, according to a research paper by OpenAI.

Decoding the White House AI Executive Order’s Achievements - The new Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (EO) demonstrates the Biden Administration's commitment to fostering a vibrant AI ecosystem, harnessing and governing AI, and attracting AI talent through immigration, while also addressing concerns such as foundation models and content provenance.

Policy

UK, US, EU and China sign declaration of AI’s ‘catastrophic’ danger - The UK, US, EU, Australia, and China have all signed a declaration acknowledging the potentially catastrophic risks posed by artificial intelligence and agreeing to work together on AI safety research.

San Francisco Launches Ground War to Rein in Robotaxis - San Francisco lawmakers are introducing new legislation to tighten controls over fleet charging sites for autonomous vehicles, following the California Department of Motor Vehicles' decision to block Cruise from operating robotaxis in the city.

Tesla Wins Suit That Blamed Its Software for Deadly Crash - Tesla wins lawsuit blaming its software for a deadly crash, potentially impacting pending cases and consumer perceptions of the company's vehicles.

UN creates AI advisory body to ‘maximise’ benefits for humankind - UN Secretary-General António Guterres has unveiled an AI advisory body to harness the technology's power for good and mitigate its risks through international collaboration and governance, with preliminary recommendations to be made by the end of the year.

AI Regulation Is Needed to Support US Workers in Future, Business Officials Say - AI regulation is necessary to protect and support American workers in the future, according to business officials discussing the balance between new regulations and risk.

Midjourney, Stability AI and DeviantArt win a victory in copyright case by artists — but the fight continues - A copyright infringement lawsuit against AI art generators Stability AI, Midjourney, and DeviantArt has been largely dismissed by a US District Court judge, who ruled that the artists' complaint was defective in various respects, including the failure to file copyrights on their art and the lack of substantial similarity between the AI-generated images and the original copyrighted works.

US to launch its own AI safety institute - The United States will launch an AI safety institute to evaluate risks of "frontier" AI models and establish a partnership with the UK Safety Institute, according to Secretary of Commerce Gina Raimondo.

SAG-AFTRA Awaits Studios’ Response To Latest AI Proposals On Day Of No Formal Talks - SAG-AFTRA is awaiting a response from studios on their AI counter proposal and comprehensive counter proposal, as negotiations continue amidst the actors' strike.

Regulating AI by Executive Order is the Real AI Risk - The President’s Executive Order on Artificial Intelligence is a premature and pessimistic political solution to unknown technical problems and a clear case of regulatory capture at a time when the world would be best served by optimism and innovation.

Against open sourcing Automatized Knowledge Interpolators - The article discusses the potential dangers of open sourcing AI technology, arguing that the automatic interpolation of knowledge by AI systems poses significant risks that should not be freely distributed.