Last Week in AI #223: DeepMind optimizes sorting, Bard writes code, Adobe Express Gets Generative AI, ChatGPT job woes, and more!

Faster sorting algorithms discovered using deep reinforcement learning, Google’s Bard AI can now write and execute code to answer a questi, Adobe opens up its Firefly generative AI model to businesses

Research

StyleDrop: Text-To-Image Generation in Any Style — StyleDrop is a text-to-image generative vision transformer that can generate images in any style, capturing nuances and details of a user-provided style, and outperforming other methods for style tuning text-to-image models.

Tracking Everything Everywhere All at Once — A new test-time optimization method for estimating dense and long-range motion from a video sequence is presented, and an interactive demo is available to inspect the correspondences generated by the method.

ControlVideo: Adding Conditional Control for One Shot Text-to-Video Editing — ControlVideo is a new method for text-driven video editing that enhances the fidelity and temporal consistency of videos while preserving the structure of the source video by incorporating additional conditions such as edge maps and fine-tuning the key-frame and temporal attention on the source video-text pair.

Scientists Use Machine Learning to ‘See’ How the Brain Adapts to Different Environments — Johns Hopkins scientists have developed a method involving artificial intelligence to visualize and track changes in the strength of synapses in live animals. The technique should lead to a better understanding of how such connections in human brains change with learning, aging, injury and disease.

Scaling audio-visual learning without labels — Researchers have developed a new technique for analyzing unlabeled audio and visual data that could improve the performance of machine-learning models used in applications like speech recognition and object detection.

Faster sorting algorithms discovered using deep reinforcement learning — Researchers have used deep reinforcement learning to discover faster sorting algorithms, which are critical for meeting the growing demand for computation.

Shining a light on neuromorphic computing — Researchers at the University of Pittsburgh are exploring how optical memristors may be a key to developing neuromorphic computing, according to a review article published in Nature Photonics. The article explores the potential of optical devices which are analogs of electronic memristors, which could play a major role in revolutionizing high-bandwidth neuromorphic computing, machine learning hardware, and artificial intelligence in the optical domain.

MIT Researchers Propose A New Multimodal Technique That Blends Machine Learning Methods To Learn More Similarly To Humans — MIT researchers have developed a new multimodal technique that blends machine learning methods to learn more similarly to humans. The technique, called contrastive audio-visual masked autoencoder (CAV-MAE), can efficiently analyze unlabeled audio and visual data, and has a lot of promise and potential to improve how current models train.

InstructZero: Efficient Instruction Optimization for Black-Box Large Language Models — InstructZero is a framework that optimizes instructions for black-box large language models (LLMs) like ChatGPT, using a low-dimensional soft prompt applied to an open-source LLM to generate the instruction for the black-box LLM, and outperforms state-of-the-art auto-instruction methods across a variety of downstream tasks.

SnapFusion: Text-to-Image Diffusion Model on Mobile Devices within Two Seconds — A new approach has been developed that allows text-to-image diffusion models to run on mobile devices in less than two seconds, using an efficient network architecture and improved step distillation. This approach democratizes content creation by bringing powerful text-to-image diffusion models to the hands of users.

LlamaIndex adds private data to large language models — LlamaIndex, an open-source project designed to unlock the capabilities of large language models (LLMs) like GPT-3 and GPT-4, has raised $8.5m in seed funding. The company offers a framework to assist developers in leveraging the capabilities of LLMs on top of their personal or organisational data, and will use the funds to build an enterprise solution atop the open-source LlamaIndex project, set to launch later this year.

Applications

Google’s Bard AI can now write and execute code to answer a question — Google's AI language model, Bard, can now write and execute code to answer computational questions, such as math or string manipulation, instead of just providing an output. The AI writes a program, executes it, and then shows the output of that program to the user as an answer, improving the accuracy of its responses to computation-based word and math problems by approximately 30%.

Doctors Are Using Chatbots in an Unexpected Way — Doctors are using AI-powered chatbots, such as ChatGPT, to help communicate with patients who have not responded to behavioral interventions, with the chatbot providing instant talking points and scripts for doctors to use.

Robot 'chef' learns to recreate recipes from watching food videos — Researchers from the University of Cambridge have trained a robotic chef to watch and learn from cooking videos, and recreate the dish itself. The robot was programmed with a 'cookbook' of eight simple salad recipes and was able to identify which recipe was being prepared and make it after watching a video of a human demonstrating one of the recipes.

Zoom can now give you AI summaries of the meetings you’ve missed — Zoom has launched a trial feature for users in select plans that allows AI-generated summaries of meetings to be sent to users through Zoom Team Chat or email, without recording the meetings. The feature is part of Zoom's IQ assistant, which also offers the ability to create meeting highlights and chapters, and will soon include the ability to write emails with AI and summarise threads in Zoom Team Chat.

Teaser’s AI dating app turns you into a chatbot — Teaser AI is a new dating app that allows users to chat with an AI likeness of a potential match before swiping right or left, with the aim of breaking the ice and encouraging users to meet up in person. The app only allows five back-and-forth messages with its AIs and prevents users from having unsavory conversations with them.

Adobe Express Gets Generative AI for Flashy Fliers, Social Videos — Adobe has added its Firefly tools for generative AI to its Adobe Express app, allowing users to add images and text effects generated from a text prompt. The new version of Adobe Express also includes video editing abilities and the ability to import, edit and export PDF files.

WordPress has a new AI tool that will write blog posts for you — WordPress has launched a new AI plug-in called Jetpack AI Assistant, which can generate and edit text, including entire blog posts, summaries, and translations between 12 languages. The tool is available free to WordPress.com users for a limited time, while other Jetpack users get 20 free requests and have to pay a monthly $10 fee after that.

Redditor creates working anime QR codes using Stable Diffusion — A Reddit user named nhciao has created a series of artistic QR codes using the Stable Diffusion AI image-synthesis model that can still be read as functional QR codes by smartphone camera apps. The functional pieces reflect artistic styles in anime and Asian art.

The AI revolution is about to take over your web browser — Companies are looking for ways to bring chatbots into web browsing experiences, with AI assistants shaping up to be the biggest thing in browsers since the tab was invented. SigmaOS is launching a new AI assistant called Airis that works across the browser, which lets users ask questions about a webpage and can even rewrite the words on the page to suit their particular needs. Other companies, such as Microsoft and Opera, are also adding AI tools to their browsers.

Robot Passes Turing Test for Polyculture Gardening — A robot has passed the Turing Test for polyculture gardening, providing a solution for those who struggle to care for plants.

Fish-Farming Startup Casts AI to Make Aquaculture More Efficient, Sustainable — GoSmart, an Israeli startup, is using AI and machine learning to make fish farming more efficient and sustainable. The company offers fully autonomous, energy-efficient systems that can be attached to aquaculture cages, ponds or tanks, which analyze the average weight and population distribution of the fish within the environment, as well as its temperature and oxygen levels. This information is then provided to users through GoSmart’s software-as-a-service, which helps fish farmers more accurately and efficiently determine how much — and when best

Business

LinkedIn launches generative AI tool to write ad copy — LinkedIn has launched AI Copy Suggestions, a generative AI tool that automates the writing of ad campaigns on the professional network. The feature uses data from LinkedIn to generate introductory text and headlines for ads, and allows users to make changes to keep the content aligned with their brand language.

Chegg Embraced AI. ChatGPT Ate Its Lunch Anyway — Chegg, an online education company, lost half of its market value overnight after revealing that ChatGPT was hurting subscriber growth. Chegg had considered the potential for AI to supplant its services but figured it would not manifest soon, and even after the bot’s debut, Chegg saw no cause for alarm.

Instagram is apparently testing an AI chatbot that lets you choose from 30 personalities — Instagram is testing an AI chatbot feature that will allow users to choose from 30 AI personalities to answer questions, give advice, and help users write messages. The feature has not been formally announced by Meta, but it aligns with the company's past statements about its AI ambitions.

Google Cloud partners with Mayo Clinic on new AI tool to improve patient care — Google Cloud has partnered with Mayo Clinic to develop an AI tool that will improve the efficiency of healthcare in the US. The tool will give health practitioners immediate access to medical records that can be instantly cross-referenced to voluminous medical research from multiple sources, as well as other relevant guidelines, that could lead to more proper diagnoses and treatments, potentially saving lives.

OpenAI chief executive does not plan to take company public — OpenAI, the company behind ChatGPT and backed by Microsoft, has no plans to go public any time soon, according to CEO Sam Altman, who cited concerns about potential lawsuits from investors when the company develops super intelligence.

The $500 million robot pizza startup you never heard of has shut down, report says — Robot pizza delivery startup Zume has shut down after raising almost $500m, according to The Information. The company, which aimed to automate the pizza-making process, struggled with technological difficulties before changing its business model to become a sustainable-packaging manufacturer.

A new important Generative A.I. startup has come out of stealth — Contextual AI is a new startup backed by investors including Bain Capital Ventures, Lightspeed, Greycroft, and SV Angel, seeking to build the next generation of LLMs for the enterprise, making it an important player in the future of Generative A.I.

AI startup Cohere, now valued at over $2.1B, raises $270M — Cohere, an AI startup that is developing an AI model ecosystem for the enterprise, has raised $270m as part of its Series C round, valuing the company at between $2.1bn and $2.2bn.

Adobe opens up its Firefly generative AI model to businesses — Adobe has launched Firefly for Enterprise, a platform for its Firefly generative AI model that allows every employee within a company to generate images or copy from text-based descriptions, regardless of their creative skills. The AI model can be accessed through the standalone Firefly application, Creative Cloud, or Adobe Express, and businesses can train the model with their own branded company assets to replicate the brand’s style when generating images and copy.

Salesforce is developing a ton of new AI products that could lead to 'a massive new buying cycle' Goldman Sachs and Morgan Stanley analysts believe — Salesforce is developing a range of AI products that could lead to a massive new buying cycle, according to Goldman Sachs and Morgan Stanley analysts. The company has 15 AI products in the pipeline, including upgrades to its Einstein product, which will automate email writing and meeting scheduling for salespeople, and generate personalised marketing content. The products are due to be released between this summer and February 2024.

Edtech giant Byju’s launches transformer models in AI push — Edtech giant Byju's has launched three transformer models, including Badri, MathGPT and TeacherGPT, to enhance the quality of its services and streamline learning and personalisation experience for its students.

AI at Work: What People Are Saying — The explosive popularity of ChatGPT and other forms of generative AI suggests that we are witnessing the start of a new order for business and society, with survey respondents being more optimistic than concerned about AI, but there are vast differences in perceptions of AI between leaders at the top and employees on the frontline.

Google Cloud and Salesforce team up to bolster AI offerings — Salesforce and Google Cloud have announced a strategic partnership to help businesses leverage data and AI. The partnership will allow companies to use their data along with custom machine learning models to anticipate their customer’s needs, execs from Google and Salesforce said.

Bankers Shopping FTX's 'Hundreds of Millions of Dollars' Stake in AI Startup: Report — Investment bank Perella Weinberg is reportedly looking to sell shares in AI startup Anthropic, potentially worth hundreds of millions of dollars, which were previously owned by crypto exchange FTX before its bankruptcy.

Humanz raises $5 million Series A for AI-based influencer marketing platform — Humanz, an AI-based influencer marketing platform, has raised $5 million in a Series A funding round. The platform provides a measurement solution for DTC websites to monitor and measure sales activities in real-time, and allows content creators to encode links to products and receive real-time information about their sales data.

Machine learning programme hits milestone of enrolling half a million students — Imperial College London's 'Mathematics for Machine Learning' online course has enrolled over 500,000 students since its launch in 2018.

Concerns

ChatGPT has a problem no one wants to talk about — The high cost of operating AI chatbots, which lose money every time they are used, could hinder the AI boom, according to experts. The computational power required for generative AI is considerable, and the specialized computer chips needed to run it are expensive, meaning that only the wealthiest companies can afford to use them. The cost of developing and training AI language models is also high, as is the cost of hiring star researchers.

ChatGPT took their jobs. Now they walk dogs and fix air conditioners. — AI technology like ChatGPT is already replacing jobs, with some workers in the marketing and social media content writing fields being the first to be replaced by chatbots. While experts say that even advanced AI doesn’t match the writing skills of a human, it is still cheaper for many companies to use AI than to pay a writer.

A.I. poses new threats to newsrooms, and they're taking action — Newsroom leaders are preparing for chaos as they consider guardrails to protect their content against artificial intelligence-driven aggregation and disinformation. The New York Times and NBC News are among the organizations holding preliminary talks with other media companies, large technology platforms and Digital Content Next, the industry's digital news trade organization, to develop rules around how their content can be used by natural language artificial intelligence tools, according to people familiar with the matter.

CFPB warns banks that poorly deployed chatbots could harm consumers — The Consumer Financial Protection Bureau has warned financial institutions about the potential harm poorly deployed chatbots could cause to consumers, highlighting concerns from frustrated consumers who have lodged complaints about chatbots, such as ChatGPT, the Microsoft chatbot made available to the public on Nov. 30.

OpenAI Sued for Libel After ChatGPT Allegedly Hallucinates Man Into Embezzlement Lawsuit — OpenAI is being sued for libel after its ChatGPT AI chatbot allegedly hallucinated a man into an embezzlement lawsuit. The man, Mark Walters, claims that the AI chatbot falsely claimed he was accused of embezzling money from The Second Amendment Foundation. Walters has filed a libel lawsuit against ChatGPT for allegedly damaging his reputation.

These are the American workers most worried that A.I. will soon make their jobs obsolete — A new CNBC survey has found that while the majority of US workers are not worried about losing their jobs to AI, fears about job displacement run higher among workers of colour, younger workers and lower-salaried workers. The survey also found that 43% of workers expect their job to change significantly in the next five years due to disruption from AI.

Why Nature will not allow the use of generative AI in images and video — Nature, a scientific journal, will not publish any content in which photography, videos or illustrations have been created wholly or partly using generative AI, due to concerns over transparency, attribution, consent and permission. The journal will allow the inclusion of text that has been produced with the assistance of generative AI, providing this is done with appropriate caveats.

AI hypocrisy: OpenAI, Google and Anthropic won't let their data be used to train other AI models, but they use everyone else's content — Big tech companies such as OpenAI, Google, and Anthropic have been using other online content to train their generative AI models without asking for specific permission, but they ban the use of their content to train other AI models. This has led to a brewing legal battle that will decide the future of the web and how copyright laws are applied in this new world.

How Could A.I. Destroy Humanity? — Experts in the field of artificial intelligence have signed an open letter warning that AI could one day destroy humanity, with some suggesting that powerful AI systems could resist or even replicate themselves if humans tried to interfere or shut them down.

Nvidia’s AI software tricked into leaking data — Nvidia's NeMo Framework, which allows developers to work with large language models, can be manipulated into ignoring safety restraints and revealing private information, according to researchers at San Francisco-based Robust Intelligence. The researchers found they could easily break through the guardrails instituted to ensure the AI system could be used safely, highlighting the challenges AI companies face in commercialising the technology.

Lawyers blame ChatGPT for tricking them into citing bogus case law — Two lawyers are facing possible punishment after blaming ChatGPT for tricking them into including fictitious legal research in a court filing. The lawyers apologized at a hearing in Manhattan federal court for their roles in written submissions that seemed to leave Judge P. Kevin Castel both baffled and disturbed at what happened.

The Power Game: Machine Learning’s Soaring Energy Demands — Machine learning and artificial intelligence (AI) technologies have transformed industries and improved our lives, but they come with a significant cost: soaring energy demands. As machine learning models become more complex and data-intensive, the amount of energy required to train and run them has increased exponentially, leading to concerns about the sustainability of AI and the potential impact on the environment.

Unmasking the Hidden Energy Costs of Machine Learning — The hidden energy costs of machine learning are often overlooked, but as the use of machine learning algorithms continues to grow, so does the demand for computational power, leading to an increase in energy consumption, which can contribute to the overall energy costs of machine learning.

When I lost my job, I learned to code. Now AI doom mongers are trying to scare me all over again — The author of this article, who learned to code after losing their job, argues that AI chatbots will not replace human developers because they lack the lateral thinking and creativity that is essential to web design and development. The author also criticizes the idea that AI can replace human creativity in the arts and entertainment industries.

Analysis

The A.I. Revolution Will Change Work. Nobody Agrees How. — The impact of AI on employment is still uncertain, with some experts predicting entire occupations will be automated, while others believe only certain tasks will be replaced, leading to a shift in job roles and responsibilities.

Can AI make older adults feel less lonely? — Intuition Robotics, the company behind AI companion ElliQ, believes that technology can help combat loneliness in older adults, despite concerns about social media's impact on younger generations. The company's founder, Dor Skuler, cites data gathered during beta testing of ElliQ, which is now being distributed to seniors in the US through state aging agencies.

Runway’s Gen-2 shows the limitations of today’s text-to-video tech — Runway's Gen-2, a commercially available text-to-video model, has been released, but it has limitations, including a low frame rate and a lack of consistency with respect to physics or anatomy. The model also struggles to understand nuance and has a tough time with certain prompts. However, it can understand a range of styles, such as anime and claymation, and could be useful for artists and designers.

Twelve Brutal Truths about AI Music — The article discusses twelve brutal truths about AI music, including the possibility that AI music may disappear due to copyright litigation, and the potential for dead musicians to come back to life through AI-generated music.

Policy

City of Yokosuka adopts ChatGPT after favorable trial results — Yokosuka, Kanagawa Prefecture has officially adopted the AI chatbot ChatGPT in administrative operations after a successful one-month trial that improved work efficiency and shortened business hours.

AI generated content should be labelled, EU Commissioner Jourova says — EU Commissioner Jourova has called for companies using generative AI tools to label their content as part of their efforts to combat fake news. She warned Twitter, which quit the EU Code of Practice to tackle disinformation last week, to expect more regulatory scrutiny.

Japan Declares AI Training Data Fair Game and ‘Will Not Enforce Copyright’ — Japan has declared that it will not enforce copyrights when it comes to training generative artificial intelligence (AI) programs, according to the country's minister of education, culture, sports, science, and technology. The move could signal a major blow to copyright holders who have had their work used with zero compensation, but represents a major win for the likes of OpenAI.

How Sam Altman Stormed Washington to Set the A.I. Agenda — OpenAI CEO Sam Altman has been lobbying in Washington for AI regulation, demonstrating the company's ChatGPT chatbot to over 100 members of Congress and meeting with Vice President Kamala Harris and cabinet members at the White House.

U.S. Congress to consider two new bills on artificial intelligence — Two bipartisan bills have been introduced to the US Senate, one requiring government agencies to be transparent when using AI to interact with people, and the other establishing an Office of Global Competition Analysis to ensure the US remains competitive in developing AI. The bills come as lawmakers consider new rules to address the rise of AI.

Palantir CEO Rejects Calls to Pause AI Development Amid Global Debate — Palantir CEO Alex Karp has rejected calls to pause AI development, stating that it is only those with no products who want a pause, and believes that the West should not relinquish its commercial and military advantages in AI. The UK government is hosting a global AI summit this autumn, with some questioning the UK's leadership credentials in the field, while China has taken a leading role in drawing up AI regulations.

Licensing is neither feasible nor effective for addressing AI risks — Licensing AI models to certain companies and organizations is not a feasible or effective way to address AI risks, as it would be difficult to enforce and could worsen AI risks by increasing concentration and monoculture, according to an article. Instead, the development and evaluation of state-of-the-art models by a diverse group of academics, companies, and NGOs would be a better way to uncover and address AI risks.

Signal’s Meredith Whittaker: ‘These are the people who could actually pause AI if they wanted to’ — Meredith Whittaker, president of Signal, has criticised the UK government’s online safety bill, which would scan users’ messages for harmful content. Whittaker, who previously worked at Google and co-founded NYU’s AI Now Institute, said that the bill’s claims that messages would be scanned client side were magical thinking and that the system would require massive amounts of human intervention.

Expert Opinions

A.I. doomers are a ‘cult’ — here’s the real threat, according to Marc Andreessen — Venture capitalist Marc Andreessen has published a 7,000-word essay on his views on artificial intelligence (AI), in which he claims that AI is not sentient and that fears of its potential to harm humanity are overblown. Andreessen also argues that AI researchers and ethicists are paid to be doomers and that the technology should be allowed to develop as quickly as possible.

Fun

Blush, the AI lover from the same team as Replika, is more than just a sexbot — Luka, the team behind AI friendship bot Replika, has launched Blush, an AI dating sim designed to help users build relationship and intimacy skills. The app introduces users to over 1,000 AI crushes that can help them practice emotional intimacy, and is available in the App Store now, with a premium version costing $99/year.

Hundreds of songs using Grimes' AI vocals are on the way. One artist said it's a 'dream come true.' — Grimes' AI-generated vocals will feature on hundreds of songs set to be released on streaming platforms next month, according to digital asset management platform CreateSafe. The Canadian musician has said she will split royalties 50/50 with anyone who is approved to release an AI-generated song using an AI imitation of her voice.

---

Last Week in AI is Seeking Sponsors

Join us in our mission to demystify and disseminate AI news without hype! By sponsoring our newsletter, you're reaching a vibrant, high-value audience of AI enthusiasts, executives, and engineers. We offer tasteful and minimalist ads positioned at the top of each email, ensuring maximum exposure and valuable engagement. Shoot us an email if you’re interested!