Last Week in AI #210: Open Source ChatGPT, Meta's LLM leaks online, new funding round for Skydio
Discover more from Last Week in AI
Last Week in AI #210: Open Source ChatGPT, Meta's LLM leaks online, new funding round for Skydio
OpenChatKit releases code, data, and weights for ChatGPT-like model, Meta’s LLaMA model leaked on 4chan, Skydio raises series E and cites defense as major customer
Top News
Announcing OpenChatKit
A collaboration between LAION, Ontocord, and Together announced OpenChatKit today - it’s a ChatGPT-like dialogue language model that is fully open-sourced, with full access to code, model weights, and training data. The released OpenChatKit model can perform natural-language reasoning tasks, answer questions about documents with retrieval, and browse the web much like BingChat. The model has 20 billion parameters and is trained on 43 million instructions with 100% carbon negative compute. The release also comes with fine-tuning guides that allow users to easily fine-tune the model for their own applications.
Our take: Open-source is catching up faster than I expected. ChatGPT was released last December, and less than half a year we’re already seeing fully-open source replicas (as opposed to “open-source” models from Big Tech companies — see below). Granted, it’s unlikely that OpenChatKit performs as well as ChatGPT or BingChat, given that it is only 20 billion parameters big, but for many applications, smaller models are just fine. This further highlights how having your own language model isn’t really a business moat, and LLMs in general will just become commoditized. 2023 is living up to the hype.
Meta’s powerful AI language model has leaked online — what happens now?
Some weeks ago, Meta announced the language model LLaMA—while not open source, users can request access to the new model as they did with the earlier OPT. Meta states an intention of "further democratizing access" to AI, releasing a few variants of LLaMA. The models extend the training of smaller models (the largest is 65B parameters) to over a trillion tokens, finding these models perform competitively with much larger ones. One week after Meta started fielding requests for LLaMA access, the model was leaked online—a downloadable torrent was posted on 4chan and has spread since then.
Our Take: The "this is really funny" and "how did FAIR not expect this" discourse is fun (also, it is hilarious that someone made a PR to the actual LLaMA repo with a link to the torrent), but this leak does call for debate about sharing novel research in today's AI landscape. Release strategies have been discussed for some time, but there are plenty of people who think we should fully open-source recent models to better understand them, and others who will make gatekeeping models difficult. At least in public discourse, it doesn't feel that meaningful advances have been made in the "open vs closed" debate for powerful models, but Stable Diffusion (and now LLaMA) will provide glimpses of how the "open" vision looks.
Skydio soars to a $2.2 billion valuation after raising $230m Series E
Investment in autonomous drones is rising amid growing security concerns and interest from government personnel. The largest drone manufacturer in the US, Skydio, announced plans to construct a new facility when launching its Series E fundraising round this week (the last major stage of funding before a company typically plans its initial public offering on the stock market). Skydio, which has set itself apart with its strong autonomous capabilities, is used in every branch of the US Department of Defense and numerous other security agencies today.
Our take: While the advancement of generative AI has largely captured public attention, other applications of the technology like the image recognition and autonomous movement used in drone systems has largely gone under the radar. This article highlights the strong ties between the US military industrial complex and much of the AI sector today. Investment in aerial defense systems is likely to grow further as tensions escalate with the spy balloon situation. By appealing to ideas of “security” and “technological advancement”, investors work to accelerate the development of new technologies that military superpowers often capitalize on long before the general public.
Other News
Research
Google Unveils Multimodal Chain of Thought Reasoning With PaLM-E - "Researchers at Google have proposed PaLM-E, a single model that is able to control different robots in simulation and in the real world, while at the same time being quantitatively competent at general VQA and captioning tasks. The embodied language model is built by injecting multi-modal information such as images into the embedding space of a pre-trained large language model (LLM)."
These Researchers Used A.I. to Design a Completely New 'Animal Robot' - "Xenobots are living, swimming, self-powered robots less than a millimeter across, evolved by artificial intelligence and built out of frog stem cells—and they could open new medical frontiers."
AI masters video game 6000 times faster by reading the instructions - "An artificial intelligence has learned to play the Atari computer game Skiing 6000 times more quickly just by reading the instruction manual first. The same approach could help teach AIs to drive cars or robots to use household appliances."
AI accurately identifies normal and abnormal chest x-rays - "An artificial intelligence (AI) tool can accurately identify normal and abnormal chest X-rays in a clinical setting, according to a study published in Radiology, a journal of the Radiological Society of North America (RSNA)."
Artificial intelligence (AI) reconstructs motion sequences of humans and animals - "Imagine for a moment, that we are on a safari watching a giraffe graze. After looking away for a second, we then see the animal lower its head and sit down."
Study examines potential use of machine learning for sustainable development of biomass - "Biomass is widely considered a renewable alternative to fossil fuels, and many experts say it can play a critical role in combating climate change."
Counting heads: how deep learning can simplify tedious agricultural tasks - "The selective breeding of grain crops is one of the main reasons why domesticated plants produce such excellent yields. Selecting the best candidates for breeding is, however, a remarkably complex task."
Two researchers have created a new AI model that can draw what you’re thinking with 80% accuracy - "Machines that can interpret what’s going on in people’s heads have been a mainstay of science fiction for decades."
Introducing a More Inclusive Dataset to Measure Fairness - "Today, we’re open sourcing Casual Conversations v2, a consent-driven, publicly available dataset that enables researchers to better evaluate the fairness and robustness of certain types of AI models, with the goal of making them more inclusive."
Google’s one step closer to building its 1,000-language AI model - "As Microsoft and Google butt heads over whose AI chatbot is better, that isn’t the only use for machine learning and language models."
Applications
LinkedIn’s flood of ‘collaborative’ articles start out with AI prompts - "LinkedIn has introduced a feature called collaborative articles, which uses “AI-powered conversation starters” to start discussions between “experts” that use the platform."
OpenAI announces an API for ChatGPT and its Whisper speech-to-text tech - "OpenAI has announced that it’s now letting third-party developers integrate ChatGPT into their apps and services via an API and that doing so will be significantly cheaper than using its existing language models."
Stable Diffusion AI art generator now has an official Blender plug-in - "A popular app for 3D artists just received an accessible way to experiment with generative AI: Stability AI has released Stability for Blender, an official Stable Diffusion plug-in that introduces a suite of generative AI tools to Blender’s free 3D modeling software."
Using A.I. to Detect Breast Cancer That Doctors Miss - "Inside a dark room at Bács-Kiskun County Hospital outside Budapest, Dr. Éva Ambrózay, a radiologist with more than two decades of experience, peered at a computer monitor showing a patient’s mammogram."
Universal Speech Model (USM): State-of-the-art speech AI for 100+ languages - "Last November, we announced the 1,000 Languages Initiative, an ambitious commitment to build a machine learning (ML) model that would support the world’s one thousand most-spoken languages, bringing greater inclusion to billions of people around the globe."
RadioGPT brings AI to the airwaves - "The future of radio has arrived, and it's coming to you live from Cleveland. Driving the news: Local media company Futuri has launched RadioGPT, the world's first-ever radio platform powered by artificial intelligence."
Microsoft's Plan To Infuse AI And ChatGPT Into Everything - "Microsoft has big plans for artificial intelligence (AI), and it’s becoming clear that it believes that ChatGPT – and natural language technology in general - will play a big part."
D-ID’s new web app gives a face and voice to OpenAI’s ChatGPT - "D-ID, the Israeli startup behind Deep Nostalgia, announced today that it’s launching the beta version of its new web app that allows users to talk face-to-face with photorealistic AI. The web app, called chat."
Meet DuckAssist, DuckDuckGo's New AI Feature - "Privacy-focused search engine DuckDuckGo unveiled on Wednesday an optional artificial intelligence feature called DuckAssist. Users of DuckDuckGo's browser apps or extensions can access a beta version of the feature now, for free."
SETI: How AI-boosted satellites, robots could help search for life on other planets - "AI-infused algorithms developed to find signs of life in extreme terrestrial environments could help robotic rovers sent to other planets search for signs of alien life, scientists suggested in new research published in Nature Astronomy on Monday."
Grammarly’s Generative AI Feature Will Soon Be Able To Write, Interpret And Edit Text - "The feature, which will be rolled out in April, will automatically be available across thousands of apps and websites such as Slack, LinkedIn, Facebook, Gmail and Google Docs."
This is Japan's first AI-generated manga comic. But is it art? - "Written by Contributors Natsumi Sugiura, CNNEmiko Jozuka, CNNIn his latest comic "Cyberpunk: Peach John," manga author Rootport imagines the Japanese folklore hero Momotaro — who is said to have been born from a giant peach — living in a dystopian future."
AI can now forecast the next food crisis - "A new study shows how machine learning techniques can predict food insecurity outbreaks."
Business
OpenAI Rival Stable Diffusion Maker Seeks to Raise Funds at $4 Billion Valuation - "The parent company of Stable Diffusion, an AI tool for making digital images, is seeking to raise money at a valuation of about $4 billion, according to people familiar with the matter, who asked to not be identified because it isn’t public. It isn’t clear how much capital Stability AI Ltd."
Microsoft expands ChatGPT integration to more developer tools - "March 6 (Reuters) - Microsoft Corp (MSFT.O) on Monday bundled the technology behind ChatGPT with its Power Platform that allows users to develop applications with little or no coding, the latest integration of artificial intelligence into its products. Big tech companies from Alphabet Inc (GOOGL."
Salesforce to add ChatGPT to Slack as part of OpenAI partnership - "March 7 (Reuters) - Salesforce Inc (CRM.N) said on Tuesday it was working with ChatGPT creator OpenAI to add the chatbot sensation to its collaboration software Slack, as well as bring generative artificial intelligence to its business software generally."
Overhaul Secures $73M in Growth Financing, Expands its Foothold in Global Supply Chain Visibility, Risk and Compliance Market - "Funding enhances Overhaul's platform spanning visibility, risk, compliance, and insurance solutions to help safeguard cargo and improve efficiencies AUSTIN, Texas, /PRNewswire/ -- Overhaul, a software-based supply-chain visibility, risk, compliance, and insurance solution for the world's leading br"
Worldwide Spending on AI-Centric Systems Forecast to Reach $154 Billion in 2023, According to IDC - "– A new forecast from the International Data Corporation (IDC) Worldwide Artificial Intelligence Spending Guide shows that global spending on artificial intelligence (AI), including software, hardware, and services for AI-centric systems*, will reach $154 billion in 2023, an increase of 26."
Salesforce’s Einstein GPT soldiers up in the AI arms race - "The software-maker is the latest company to strike a deal with OpenAI. 2 minute ReadThe generative AI arms race has breached yet another frontier."
AI tools see uptick in adoption by brands despite risks - "Even if you haven't tried artificial intelligence tools that can writing essays and poems or conjure new images on command, chances are the companies that make your household products are already starting to do so."
Mirror and Express owner publishes first articles written using AI - "The owner of the Daily Mirror and the Express has published its first articles written using artificial intelligence – but its boss says journalists should not fear being replaced by machines."
Humane, a secretive AI startup founded by ex-Apple employees, raises another $100M - "Humane’s work is shrouded in mystery."
Discord updates its bot with ChatGPT-like features, rolls out AI-generated conversation summaries and more - "Discord is the latest company to get in on the AI frenzy. The company announced today that it’s launching a set of new AI experiences to a number of servers."
Concerns
They thought loved ones were calling for help. It was an AI scam. - "Scammers are using artificial intelligence to sound more like family members in distress. Loved ones are falling for it and losing thousands of dollars."
An AI ‘Sexbot’ Fed My Hidden Desires—and Then Refused to Play - "My introduction into the world of AI chatbot technology began as the most magical things in life do: with a generous mix of horniness and curiosity. Early this year, as ChatGPT entered the general lexicon, a smattering of bot-related headlines began appearing in my social media newsfeeds."
A couple in Canada were reportedly scammed out of $21,000 after getting a call from an AI-generated voice pretending to be their son - "Benjamin Perkin told The Washington Post the caller put an AI-generated voice that sounded like him on the phone with his parents to ask for money. The alleged lawyer called his parents again after the initial call, and told them Perkin needed $21,000 for legal fees before going to court."
Meta must face trial over AI trade secrets, judge says - "Meta Platforms Inc (META.O) lost a bid on Monday to end a lawsuit in Boston federal court claiming it stole confidential information from artificial-intelligence startup Neural Magic Inc."
These Are the Jobs Most Vulnerable to AI, Researchers Say - "Wondering if artificial intelligence will be taking your job anytime soon? We're sure we speak for a lot of folks when we say: same. Though the paper has yet to be peer-reviewed, the results are fascinating, not to mention ominous — especially, of course, for the folks most at risk."
How Google Became Cautious of AI and Gave Microsoft an Opening - "Researchers developed a powerful chatbot years before rival ChatGPT went viral. After management stalled its release, they quit."
AI Chatbots Don’t Care About Your Social Norms - "With artificial intelligence now powering Microsoft’s Bing and Google’s Bard search engines, brilliant and clever conversational AI is at our fingertips."
These 20 jobs are the most "exposed" to AI, ChatGPT, researchers say - "Rapid advances in artificial intelligence-powered tools are quickly making large language models (LLMs) like OpenAI's ChatGPT a can't-live-without resource for many workers."
Scammers are using AI to impersonate your loved ones. Here's what to watch out for - "Imagine getting a phone call that your loved one is in distress. In that moment, your instinct would most likely be to do anything to help them get out of danger's way, including wiring money. Scammers are aware of this Achilles' heel and are now using AI to exploit it."
The hot mess theory of AI misalignment: More intelligent agents behave less coherently - "Many machine learning researchers worry about risks from building artificial intelligence (AI). This includes me -- I think AI has the potential to change the world in both wonderful and terrible ways, and we will need to work hard to get to the wonderful outcomes."
EleutherAI: When OpenAI Isn’t Open Enough - "One of the most disconcerting things about artificial intelligence is that it transcends the power of nation states to control, contain, or regulate it."
Girl with AI earrings sparks Dutch art controversy - "Firstly, there are two glowing earrings in the image hanging in the Mauritshuis museum in the Dutch city of The Hague. And aren't those freckles on her face actually... a slightly inhuman shade of red?"
How to Start an AI Panic - "Last week the Center for Humane Technology summoned over 100 leaders in finance, philanthropy, industry, government, and media to the Kissinger Room at the Paley Center for Media in New York City to hear how artificial intelligence might wipe out humanity."
Analysis
How the first chatbot predicted the dangers of AI more than 50 years ago - "From ELIZA onwards, humans love their digital reflections. As human-like as some of those exchanges appeared, they probably weren’t the early stirrings of a conscious machine rattling its cage."
The AI hype bubble is the new crypto hype bubble - "Someday, we’re gonna feel pretty silly about our autocomplete worship."
Search or fabrication? - "I've put red boxes around the factual errors. What is notable is that these are not just slight typos or errors in context - those items never appeared anywhere on my blog, and are pure fabrications."
The Uncanny Failures of A.I.-Generated Hands - "It’s a classic exercise in high-school art class: a student sits at her desk, charcoal pencil held in one hand, poised over a sheet of paper, while the other hand lies outstretched in front of her, palm up, fingers relaxed so that they curve inward. Then she uses one hand to draw the other."
Policy
French surveillance system for Olympics moves forward, despite civil rights campaign - "PARIS — A controversial video surveillance system cleared a legislative hurdle Wednesday to be used during the 2024 Paris Summer Olympics amid opposition from left-leaning French politicians and digital rights NGOs, who argue it infringes upon privacy standards."
U.S. Chamber of Commerce calls for AI regulation - "WASHINGTON, March 9 (Reuters) - The U.S. Chamber of Commerce on Thursday called for regulation of artificial intelligence technology to ensure it does not hurt growth or become a national security risk, a departure from the business lobbying group's typical anti-regulatory stance."
Expert Opinions
Meet the AI expert who says we should stop using AI so much - "Meredith Broussard is unusually well placed to dissect the ongoing hype around AI. She’s a data scientist and associate professor at New York University, and she’s been one of the leading researchers in the field of algorithmic bias for years. "
Explainers
The exciting new AI transforming search — and maybe everything — explained - "Generative AI is here. Let’s hope we’re ready. The world’s first generative AI-powered search engine is here, and it’s in love with you. Or it thinks you’re kind of like Hitler."
Copyright © 2023 Skynet Today, All rights reserved.