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 20,000 subscribers

News

Last Week in AI #225: ByteDance buys $1 billion of Nvidia GPUs, US State Department to use AI to for contracts, open-source and commercially permissive LLaMA, AI is infiltrating crowd work, and more!

ByteDance places huge order of A100 and H800 cards, State Department soliciting comments about AI-written contracts, release of OpenLLaMA, crowd workers using AI to do tasks that are used to train AI

Last Week in AI

Jun 26, 2023

China's ByteDance Has Gobbled Up $1 Billion of Nvidia GPUs for AI This Year

Chinese tech giant ByteDance, the creator of TikTok, has reportedly ordered around $1 billion worth of Nvidia GPUs for AI applications in 2023. This amount is equivalent to what the entire Chinese market ordered from Nvidia in 2022. ByteDance's order includes around 100,000 units of Nvidia's A100 and H800 cards. The H800 cards are a custom accelerator developed by Nvidia to comply with export restrictions. The increased investment in AI hardware by Chinese tech giants reflects China's desire to compete with the West in the field of AI and its need for high-tech solutions for state control purposes. The exact status of the deliveries is unclear, but it is expected that the deliveries of the H800 cards will continue throughout the year. The significant investment by ByteDance highlights the growing demand for AI technology in China.

State Department considers generative AI for contract writing

The Department of State is considering the use of generative AI to assist with contract writing. The department has issued a request for information to gather industry feedback on the potential use of LLMs for contract writing. The goal is to reduce inefficiencies, simplify workflows, and improve the accuracy of repetitive tasks in the market research and acquisition planning phases. Currently, contracting officers manually copy and paste information from previous contracts, which can lead to errors and introduce risks. The ideal solution would involve a generative AI system generating a complete draft package for review by a government procurement professional. The use of generative AI could reduce costs, manual labor, errors, and improve decision-making and contract outcomes. However, there are constraining factors, such as the lack of approval for the technology's use and the absence of a published AI policy. Interested parties have until July 17 to respond to the request for information. The Department of State is committed to adopting emerging technologies like AI for global diplomacy.

OpenLLaMA is a fully open-source LLM, now ready for business

The OpenLLaMA team has released an open-source reproduction of Meta's LLaMA language model, which can now be used commercially. OpenLLaMA aims to replicate the quality of Meta's models by following Deepmind's Chinchilla scaling laws and training on large amounts of data. The team has released an early version of the 7-billion-parameter OpenLLaMA model, which has been trained on 300 billion tokens and is approaching the performance level of LLaMA. They are also working on a 3-billion-parameter version for use cases with limited resources. This provides a true open-source alternative to services like OpenAI's ChatGPT, allowing businesses to utilize powerful language models for commercial purposes. The OpenLLaMA model is available on HuggingFace and more information can be found on Github.

We should all be worried about AI infiltrating crowdsourced work

A recent study from EPFL reveals that a significant number of crowd workers on Amazon's Mechanical Turk platform have been using AI tools to cheat on assigned tasks. This raises concerns about the integrity of the data generated through crowdsourcing. Mechanical Turk is a platform that assigns tasks to humans, which are typically difficult for computers to perform. However, the availability of cheap AI tools like ChatGPT has made it easier for workers to use automation to complete their tasks. The researchers developed a methodology to distinguish between human-generated and machine-generated text. This issue highlights the potential consequences of relying on humans for tasks that could be done by machines.

Other News

Research

TAPIR: Tracking Any Point with per-frame Initialization and temporal Refinement - A new model called TAPIR is introduced for effectively tracking a query point in a video sequence, using an architecture inspired by Persistent Independent Particles (PIPs) and incorporating features such as trajectory initialization, temporal refinement, and estimation of position uncertainty.

AI could predict pancreatic cancer early in some cases, study shows - AI has the potential to predict pancreatic cancer early by analyzing medical records and identifying individuals at elevated risk up to three years before diagnosis.

This new AI autopilot algorithm can help pilots avoid crashes - study - Researchers from MIT have developed an AI algorithm that can help stabilize planes in low altitudes, providing better stability and safety than traditional algorithms, and potentially being implemented as part of a larger system to stabilize planes during extreme situations.

New research targets diseases including Alzheimer's - A new AI-based technique for measuring fluid flow around the brain's blood vessels could have significant implications for developing treatments for diseases such as Alzheimer's.

Hybrid AI-powered computer vision combines physics and big data - Hybrid AI-powered computer vision combines physics and big data to enhance the capabilities of artificial intelligence in perceiving and interacting with the physical world, leading to improved performance in tasks such as autonomous driving and precision actions by robots.

AI that uses sketches to detect objects within an image could boost tumor detection, and search for rare bird species - Teaching AI to detect specific objects in images using sketches could revolutionize cancer detection and wildlife conservation.

Seeing the World through Your Eyes - Reconstructing 3D scenes using eye reflections is possible by accurately estimating eye pose and refining the cornea poses, radiance field, and eye iris texture.

BITE: Beyond Priors for Improved Three-D Dog Pose Estimation - BITE enables improved 3D shape and pose estimation of dogs from a single input image by utilizing a dog-specific 3D parametric model, exploiting body-ground contact as side information, developing a novel neural network architecture, and creating a synthetic dataset for evaluation.

Applications

How AI and Machine Learning is Transforming the Online Gaming Experience - AI and machine learning have transformed the online gaming experience by providing enhanced gameplay, personalized content and recommendations, intelligent game design, and improved player interactions.

Google extends generative AI search to places, products and virtual models - Google is expanding its generative AI search to include tourism and product searches, as well as allowing users to try on clothes on virtually generated models.

AI human-voice clones are coming for the Amazon, Apple, Google audiobook - AI-generated voices are being used in audiobooks, but there are challenges in making them sound human and connect with listeners, such as the difficulty of replicating timing and emotional nuances that human narrators excel at.

Opera's generative AI-infused browser is ready for the masses - Opera has released its generative AI-infused browser, Opera One, which features an integrated AI called Aria that can answer user queries, generate text or code, and provide support for Opera products, while also offering a modular design and tab grouping feature.

Vimeo intros a trio of AI-powered editing features - Vimeo is introducing a suite of AI-powered tools to help entry-level video creators create scripts, record footage, and remove unwanted disfluencies, aiming to reduce barriers and give creators the confidence to create videos in one take.

YouTube is getting AI-powered dubbing - YouTube is introducing an AI-powered dubbing tool to make it easier for creators to dub their videos in other languages.

Vimeo’s new AI script generator will write corporate marketing filler in seconds - Vimeo is integrating new AI tools, including a script generator, to help users create fully produced videos in minutes, with a focus on corporate customers, although the accuracy and usefulness of the script generator remains to be seen.

LinkedIn Reveals AI Image Hunter That Catches Fake Profiles - LinkedIn has developed an AI image detector that can catch 99.6% of fake profile images, with a 1% false positive rate, making it more difficult to create fake profiles on the platform.

ChatGPT, Google Bard Shares Generic Windows 11, Windows 10 Pro Keys (Updated) - ChatGPT and Google Bard can generate generic working keys for Windows 10 Pro and Windows 11 Pro, allowing users to install the operating systems but not activate them, according to a Twitter user who successfully obtained the keys and shared the process.

AI Is a Lot of Work - AI annotation work, which involves labeling and sorting data to train AI systems, is often tedious, low-paying, and hidden behind the scenes, with workers unaware of the larger purpose of their tasks or the companies they work for.

Business

‘AI’ Searches Rocket To The Highest Point In History - Search interest in 'AI' has reached its highest point in history, with a more than doubling of searches since the beginning of 2023, and the biggest increases correlating with the releases of AI chatbots ChatGPT, Bing AI, and Google Bard.

The AI Boom Has Silicon Valley on Another Manic Quest to Change the World - Silicon Valley is experiencing a frenzy of investment and excitement as AI technologies reshape the industry.

A.I. is now the biggest spend for nearly 50% of top tech executives across the economy: CNBC survey - Nearly half of the companies surveyed by CNBC say that AI is their top priority for tech spending over the next year, with AI budgets more than double the second-biggest spending area in tech, cloud computing.

It’s not just hype, corporate America is making huge bets on AI transforming their businesses - Corporate America is heavily investing in AI, with companies like PwC and Accenture promising to spend heavily on the sector, creating new AI-focused capital pools, and directly investing in AI-focused startups, as the race to leverage AI to add more utility and value to products continues.

This week in AI: Big tech bets billions on machine learning tools - Big tech companies are investing billions of dollars in machine learning tools, with Dropbox, AWS, Salesforce Ventures, Workday, Accenture, and PwC all announcing significant funding initiatives; however, concerns have been raised about the lack of transparency and power hierarchy in the AI field, with the potential for AI to be used to make decisions without individuals' knowledge or agency.

OpenAI plans app store for AI software, The Information reports - OpenAI plans to launch a marketplace for developers to sell their AI models built on top of its own AI technology, potentially competing with app stores run by its customers and partners like Salesforce and Microsoft.

iRobot shares surge after UK regulator clears Amazon acquisition - iRobot's stock surges after UK regulator approves Amazon's $1.7 billion acquisition, stating it will not result in a substantial lessening of competition in the UK.

IBM Expands Partnership with Adobe for Generative AI in Supply Chain - IBM is expanding its partnership with Adobe to help brands integrate generative AI models into their design and creative processes, with a focus on leveraging Adobe's Firefly tool for image and text effect generation.

Galileo launches LLM Studio to revolutionize AI adoption in enterprises - Galileo has launched Galileo LLM Studio, a platform that aims to help companies deploy natural language processing models into production faster by detecting and fixing issues with large language models, ultimately improving model accuracy and accelerating AI adoption.

UK touts £21M fund to extend AI deeper into the National Health Service - UK government commits £21 million to NHS trusts to accelerate the deployment of AI tools in hospitals, including AI imaging and decision-support tools for diagnosing and treating heart conditions, cancer, and strokes, as well as analyzing chest x-rays to detect early signs of lung cancer.

Can Adobe Catch the AI Bug? - Adobe is turning its attention to Firefly, its copyright-safe generative AI model, after its $20 billion Figma acquisition is imperiled.

Bill Gates’ venture firm, with backing from Jeff Bezos and Jack Ma, just minted a $1 billion A.I. unicorn that uses machine learning for mining rare earth metals crucial for EVs - Berkeley-based mining company KoBold Metals, backed by Bill Gates' venture firm and investors including Jeff Bezos and Jack Ma, has raised $200 million in a funding round, bringing its valuation to over $1 billion, for its AI-powered mining of rare earth metals crucial for electric vehicles.

ElevenLabs Launches New Generative Voice AI Products and Announces $19m Series A Round Led by Nat Friedman, Daniel Gross, and Andreessen Horowitz - ElevenLabs, a voice technology research company, has raised $19 million in a Series A funding round led by Nat Friedman, Daniel Gross, and Andreessen Horowitz to continue its voice AI research and product development, with the goal of making all content universally accessible.

Inflection debuts its own foundation AI model to rival Google and OpenAI LLMs - Inflection, an AI startup, has unveiled its large language model, Inflection-1, which is competitive with other models like GPT-3.5 but falls behind in coding tasks, with plans to release a larger model comparable to GPT-4 in the future.

Hugging Face and AMD partner on accelerating state-of-the-art models for CPU and GPU platforms - Hugging Face and AMD have partnered to accelerate state-of-the-art models for CPU and GPU platforms, providing developers and organizations with improved performance and cost-efficiency for training and inference tasks.

Kibsi raises $9.3M for its no-code computer vision platform - Kibsi, a startup based in Irvine, California, has raised $9.3 million in funding for its no-code computer vision platform that allows businesses to build and deploy computer vision applications using their existing cameras.

German tabloid Bild to replace range of editorial jobs with AI - Germany's Bild tabloid, the biggest-selling newspaper in Europe, is implementing a €100m cost-cutting program that will lead to hundreds of redundancies and the replacement of editorial jobs with artificial intelligence.

Inside China's underground market for high-end Nvidia AI chips - China's underground market for high-end Nvidia AI chips is thriving, with vendors in Shenzhen's Huaqiangbei electronics area selling the chips discreetly at double the usual price, despite U.S. export restrictions.

Metas LLaMA v2 could shake up the Chatbot business - Meta's upcoming open-source language model, LLaMA v2, is expected to be available for commercial use, potentially disrupting the chatbot industry and posing competition to companies like OpenAI and Google.

Dropbox creates a $50 million fund to invest in AI startups - Dropbox has launched Dropbox Ventures, a $50 million venture capital fund that will invest in AI startups and provide guidance in creating AI products for the future of work.

Concerns

How existential risk became the biggest meme in AI - Existential risks posed by AI have become a mainstream concern, with high-profile figures and organizations warning of the catastrophic dangers and calling for global prioritization in mitigating the risk of extinction.

AI-generated child sex images spawn new nightmare for the web - AI-generated child sexual abuse images are becoming increasingly prevalent on the dark web, posing challenges for law enforcement in identifying victims and combating real-world abuse.

The Race to Prevent ‘the Worst Case Scenario for Machine Learning’ - Child predators using new AI tools, such as Dall-E, to create explicit images is an immediate threat that experts believe deserves the industry's undivided attention.

An AI Is Inventing Fake Quotes by Real People and Publishing Them Online - AI-generated content, including fake quotes attributed to real people, is becoming increasingly prevalent on the internet, leading to a loss of control over one's public image and the potential for widespread misinformation.

Google warns its own employees: Do not use code generated by Bard - Google has warned its employees not to use the code generated by its AI chatbot, Bard, due to privacy and security risks, raising concerns about the trustworthiness of AI tools built by private companies.

The AI Apocalypse: A Scorecard - Top AI experts express concerns about the potential threat posed by large language models like GPT-4.

Elon Musk says he ‘didn’t think anyone would actually agree’ to the A.I. pause he called for - Elon Musk expresses surprise that anyone agreed to his call for a pause on AI development, reiterating his concerns about the potential dangers of advanced artificial intelligence and the need for strong regulation.

Consumer group calls on EU to urgently investigate ‘the risks of generative AI’ - Consumer group calls on EU to urgently investigate the risks of generative AI, citing concerns about deception, manipulation, harm, disinformation, bias, and fraud, and calling for public scrutiny and control over AI systems.

The Reddit API Blackout and the Generative AI Connection - Reddit users are protesting against the platform's decision to charge for access to its API, which is used by tech companies to train their AI models, and the protests are continuing to escalate, potentially impacting the business models of third-party applications and accessibility tools; the issue is seen as part of a larger problem of bad actors using bots to create generative AI and scrape content from social platforms.

AIs trained on other AI output will start producing junk within a few generations, scientists warn - AIs trained on other AIs will eventually produce gibberish, leading to a collapse in the quality of AI-generated content, according to machine learning experts.

Judge sanctions lawyers whose A.I.-written filing contained fake citations - Lawyers who submitted an A.I.-written legal brief containing fake citations and quotes have been sanctioned by a New York federal judge, who ordered them to pay fines and notify the judges falsely identified as the authors of the bogus case rulings.

Analysis

GPT-4's Secret Has Been Revealed - GPT-4, the highly anticipated AI model from OpenAI, is actually a combination of eight smaller models, which has led to some disappointment and raised questions about the true state-of-the-art of AI.

Machine Learning Tool Easily Spots ChatGPT's Writing - A machine learning tool can easily identify the writing generated by OpenAI's ChatGPT chatbot, which has been widely used for various writing purposes since its launch.

Imitation Models and the Open-Source LLM Revolution - Imitation models, which fine-tune smaller open-source language models using synthetic dialogue data from more powerful models, may not perform as well as proprietary models like ChatGPT in terms of factuality and performance on broader evaluations.

Policy

Five big takeaways from Europe’s AI Act - The European Parliament has passed the AI Act, which introduces restrictions and regulations on the use of AI, including bans on emotion-recognition AI, real-time biometrics and predictive policing in public spaces, social scoring, and the use of copyrighted material in training generative AI models, as well as new restrictions on recommendation algorithms on social media.

Facial recognition emerges as AI rulebook’s make-or-break issue - Facial recognition is a contentious issue in the battle over artificial intelligence in Brussels, with the European Parliament passing its version of the Artificial Intelligence Act that includes a ban on the technology in public places.

President Biden meets with AI tech leaders in San Francisco - President Biden meets with AI tech leaders in San Francisco to discuss the potential impacts of AI on American society and the importance of responsible innovation and appropriate safeguards.

How New York Is Regulating A.I. - New York City has implemented a law to regulate AI, assessing the impact of software on job candidates without focusing on the explainability of algorithms, although concerns have been raised about potential loopholes and the feasibility of independent audits.

Grammys approve new rules on use of artificial intelligence - Grammys implement new rules stating that only human creators are eligible for awards, in response to the increasing use of artificial intelligence, but musicians can still use AI in their creative process.

Singapore is not looking to regulate A.I. just yet, says the city-state's authority - Singapore takes a wait-and-see approach to regulating AI, focusing on promoting responsible use and collaborating with industry before deciding on regulatory measures.