Last Week in AI #215: Fully open ChatGPT alternative, Google's AI product plans, AI remains nonpartisan, and more!

Release of OpenAssistant, a completely open ChatGPT-like LLM, Google advances AI plans to defend its search businesses, no culture wars on AI (yet)

Top News

OpenAssistant RELEASED! The world's best open-source Chat AI!

Large language models (LLMs), like ChatGPT, have become more user-friendly and widely adopted due to alignment techniques that sync their behavior with human preferences. Two such techniques are supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), which make these powerful AI tools more accessible across different areas. One challenge is that current techniques like RLHF require expensive, high-quality data that often remain private. In an effort to democratize large-scale alignment research, OpenAssistant Conversations was released—a dataset containing 161,443 messages in 35 languages from over 13,500 volunteers. A ChatGPT-like LLM model, OpenAssistant, was built using this dataset and performed comparably to GPT-3.5-turbo (ChatGPT). All code and data have been released with unrestricted licenses.

Our take: This is big news for the open-source community and the AI community in general. There are many reasons why one might not want to build on top of closed models hosted by OpenAI or Microsoft (e.g. concerns over leaking business secrets, less flexibility on fine-tuning), and other powerful open-source models often have restrictive licenses that prevent commercial use. OpenAssistant is completely open (open data, model, code, weights, paper, training reports, etc) and will no doubt accelerate both the development and deployment of LLM in the near future. As we’ve seen with Stable Diffusion, open source has a way of quickly realizing creative use cases of AI technology as well as optimizing the underlying tech for specific contexts (e.g. generating images on a single laptop instead of GPU clusters). I’m sure we’ll see similar things happen with LLMs, and the day that everyone has a pocket ChatGPT that’s completely their own might come sooner than we anticipated.

Google Devising Radical Search Changes to Beat Back A.I. Rivals

Google is developing an all-new AI-powered search engine and updating its current engine with AI features in response to mounting competition from AI-enhanced rivals, such as Microsoft's Bing. The tech giant was alarmed by Samsung's consideration to replace Google with Bing as the default search engine on its devices, potentially jeopardizing $3 billion in annual revenue. Called Magi, the new search engine aims to provide users with a more personalized experience and anticipate their needs. It's still in the early development stages, with the release timetable unclear. Until recently, Google's widely-used search engine faced little challenge but now its $162 billion business is at risk from AI competitors.

Our take: Google’s Bard was Google’s first public product response to OpenAI’s ChatGPT and Microsoft’s Bing Chat. Bard received mixed reviews and remained limited in test releases. Now it seems like Google is planning to release an actual LLM-powered search product in Magi, although it’s still taking a measured pace, and time will tell how effective Magi is at helping Google maintain market share in search. I’m also interested to see how Microsoft responds and what new features they’ll release with Bing this year.

Where’s the AI Culture War?

There are more than enough debates about AI right now: what it is, what we're trying to do with it, where it's going, whether current methods are the right ones, and whether it will take our jobs and/or kill us all, among others. But for most people, the techno-utopian/techno-pessimist debate is the dominant one. Notably, this debate does not break along the same lines as other topics that have turned into "culture war" issues—in other words, AI has managed to remain nonpartisan so far. It's possible that this will change when controversial figures stake out positions, or our nation's rival factions may find common cause in issues such as their wariness of Big Tech. It's unclear whether and how positions on AI will evolve politically, but if any nation can make the topic political, it's America.

Our Take: I think it's right that there are many ways for an AI "culture war" to start, but I sincerely hope that's not what happens. There are many issues we will confront in the development and application of AI systems that will require a lot of political will and coordination to solve. Education and literacy around AI are vital at this moment, and I feel we can't afford to let political ire leave us divided on what to do when it comes to these nascent technologies.

Other News

Research

China scientists carry out ‘rule-breaking’ AI experiment in space - "Researchers from Wuhan University say they gave the technology full control of a satellite and set it free for 24 hours."

Another Large Language Model! Meet IGEL: An Instruction-Tuned German LLM Family - "IGEL is the Instruction-tuned German large Language Model for Text."

Meet AUDIT: An Instruction-Guided Audio Editing Model Based on Latent Diffusion Models - "Diffusion models are rapidly advancing and making lives easier. From Natural Language Processing and Natural Language Understanding to Computer Vision, diffusion models have shown promising results in almost every domain."

Researchers populated a tiny virtual town with AI (and it was very wholesome) - "What would happen if you filled a virtual town with AIs and set them loose? As it turns out, they brush their teeth and are very nice to one another! But this unexciting outcome is good news for the researchers who did it, since they wanted to produce “believable simulacra of human behavior” and"

Exploring Creativity in Large Language Models: From GPT-2 to GPT-4 - "In recent weeks, people have used large language models (LLMs) to generate a variety of creative content, such as books, flash fiction, rap battles, and music chords. But is it possible to measure the level of creative process more broadly in these models?"

Google Tells AI Agents to Behave Like 'Believable Humans' to Create 'Artificial Society' - "Researchers at Google and Stanford used ChatGPT to generate human-like characters who live and interact in a contained, video game-like world called Smallville."

Researchers used machine learning to improve the first photo of a black hole - "Researchers have used machine learning to tighten up a previously released image of a black hole."

Microsoft AI Open-Sources DeepSpeed Chat: An End-To-End RLHF Pipeline To Train ChatGPT-like Models - "There is no exaggeration in saying that ChatGPT-like concepts have had a revolutionary effect on the digital world. For this reason, the AI open-source community is working on some projects (such as ChatLLaMa, Alpaca, etc.) that aim to make ChatGPT-style models more widely available."

Tennis Robot Could Pave Way for Advancement in Fast-Movement Robotics - "Matthew Gombolay sees a future for human-scale robots in sports and athletic training. He imagines robots that can test the skills of professional athletes as well as novices looking to learn a new sport. His latest invention may serve as the touchstone to getting there."

Applications

Once More, with Feeling: Exploring Relatable Robotics at Disney - "This is a guest post. The views expressed here are solely those of the author and do not represent positions of IEEE Spectrum or the IEEE."

Poe’s AI chatbot app now lets you make your bots using prompts - "An app called Poe will now let users make their own chatbot using prompts combined with an existing bot, like ChatGPT, as the base."

Developer creates “regenerative” AI program that fixes bugs on the fly - "Debugging a faulty program can be frustrating, so why not let AI do it for you? That's what a developer that goes by "BioBootloader" did by creating Wolverine, a program that can give Python programs "regenerative healing abilities," reports Hackaday. (Yep, just like the Marvel superhero.)"

Stability AI’s new model is slightly better at generating hands - "Stability AI, the startup funding a range of generative AI experiments, has released a new version of Stable Diffusion, the text-to-image AI system that was among the first to rival OpenAI’s DALL-E 2."

Meta has open-sourced an AI project that turns your doodles into animations - "Meta has open-sourced an artificial intelligence project that lets anyone bring their doodles to life. The company hopes that by offering Animated Drawings as an open-source project other developers will be able to create new, richer experiences."

I played Dungeons and Dragons with ChatGPT. It was great at some things—but there’s an important reason I’m not threatened - "As a kid in the 1980s, I begged my parents for Heroes’ Quest, a fantasy role-playing board game that looked so fun in TV ads. When I eventually got it, I found I had no friends actually interested in playing with me, so I spent a lot of time playing with it alone."

AI is flooding the workplace, and workers love it - "Who’s afraid of ChatGPT? Not these workers. A lot of what Conor Grennan does as a dean of students at NYU’s Stern School of Business could be done at least in part by bots. Brainstorming and planning are prime examples of tasks that can be easily handled by generative AI tools like ChatGPT."

Using AI in electronic medical records to save the lives of children - "Artificial intelligence, including machine learning, is everywhere these days. From news headlines to talk show monologues, it seems like everyone is talking about artificial intelligence (AI) and how it is rapidly changing the world around us."

Adobe launches AI-powered text-based video editing - "Adobe announced new versions of its family of video applications, including the fastest and most reliable version of Premiere Pro, which adds AI-powered text-based video editing to make life simpler for professionals and creators."

Kuwaiti news outlet unveils country’s first AI-generated anchor - "An AI-generated news anchor has made her debut on a Kuwaiti media outlet, which plans to use the virtual presenter to read news bulletins. “I’m Fedha, the first presenter in Kuwait who works with artificial intelligence at Kuwait News."

Business

Elon Musk is moving forward with a new generative-AI project at Twitter after purchasing thousands of GPUs - "Elon Musk is said to be moving forward with an artificial-intelligence project within Twitter, despite recently signing an open letter calling for an industrywide halt to any AI training for several months."

Elon Musk Creates New Artificial Intelligence Company X.AI - "Elon Musk has created a new artificial intelligence company called X.AI Corp. that is incorporated in Nevada, according to a state filing."

Google Introduces TPU v4: A Machine-Learning Super-Computer With Hardware Support For Embeddings That May Be Optically Reconfigurable - "TPU v4 outperforms TPU v3 by 2.1x, and The TPU v4 pod is 4x larger at 4096 chips and thus ~10x faster overall. Machine learning (ML) models are still developing in challenging ways, both in terms of size and technique."

OpenAI Surprises Open-Source Community, Unveils Consistency Models - "In a surprising turn of events, OpenAI has decided to open-source its technology. The company had released the paper about Consistency Models last month and now have decided to open-source the technology. The fact that OpenAI has decided to open-source their technology is definitely surprising."

Sidewalk delivery robot company Neubility secures $2.42M investment - "Korean sidewalk delivery robotics company Neubility announced that it has secured a $2.42 million (3 billion won) investment from Samsung Venture Investment. This brings a cumulative total investment raised by the company to date to $24.2M (30 billion won)."

Alphabet's A.I. frenzy hits startup scene as CapitalG unit leads $100 million investment in AlphaSense - "Google's hefty investment in artificial intelligence and the latest boom in generative AI doesn't end with its homegrown products. Parent company Alphabet is also putting money to work in the startup world."

Publishers create task forces to oversee AI programs - "Editorial executives previously told Digiday in February they were encouraging employees in their newsrooms to familiarize themselves with generative AI and chatbots like ChatGPT to see if the technology could help them perform their jobs better."

Samsung wants to release EX1, a human assistant robot, this year - "Samsung Electronics has had an eye on the consumer-grade robotics niche for a while now, and during CES 2023, the company said it views robots as “a new growth engine."

Stability AI is on shaky ground as it burns through cash and looks at a management overhaul - "Stability AI stands apart from its rivals, which include Google, and the OpenAI-Microsoft partnership, in two fundamental ways. First, the company has no deep-pocketed partner that can absorb the expensive process of training AI models."

Amazon AWS expands generative AI efforts with Bedrock and CodeWhisperer updates - "Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More"

The man who unleashed AI on an unsuspecting Silicon Valley - "OpenAI CEO Sam Altman took the artificial intelligence company from nonprofit to kingmaker. Not everyone is happy with him."

Inside the AI talent wars: Tech companies are ransacking university AI programs at Stanford, MIT, and Cornell in search of rare talent - "On a Friday afternoon, Cristóbal Eyzaguirre pulled out a Hot Wheels toy car to show Insider over Zoom. For the most part, it looks like any children's toy, except for one thing: the "Tesla Model X" emblazoned across its packaging."

Concerns

Avocados and Invasive AI — The Societal Impact and The Illusion of AI Safety - "AI — the Mongoose of Society, how do we keep things safe and in check with invasive AI species? Pick at least one of the following examples, or read all of these ecological analogies at your pace before moving to the Shallow Dive into AI Safety of OpenAI:"

A.I. could lead to a ‘nuclear-level catastrophe’ according to a third of researchers, a new Stanford report finds - "Many experts in A.I. and computer science say the technology is likely a watershed moment for human society. But 36% don’t mean it as a positive, warning that decisions made by A.I."

Four to 11-year-olds deem it wrong to attack either semi-intelligent robot - "Most kids know it's wrong to yell or hit someone, even if they don't always keep their hands to themselves. But what about if that someone's name is Alexa? Four- to eleven-year-olds judged Alexa to have more human-like thoughts and emotions than Roomba."

‘I didn’t give permission’: Do AI’s backers care about data law breaches? - "Cutting-edge artificial intelligence systems can help you escape a parking fine, write an academic essay, or fool you into believing Pope Francis is a fashionista."

Don't Play Poker With ChatGPT - "The past few years have seen an explosion of progress in large language model artificial intelligence systems that can do things like write poetry, conduct humanlike conversations and pass medical school exams."

Finally, a realistic roadmap for getting AI companies in check - "It’s time for AI regulators to move fast and break things. New AI systems are coming at us so fast and furious that it might seem like there’s nothing we can do to stop them long enough to make sure they’re safe."

AI in schools – how metro districts are handing the rise in artificial intelligence chatbots - "The race for the best artificial intelligence is well underway. But is it too much, too fast? Just this month, tech leaders, including Tesla’s Elon Musk, signed a letter asking AI labs to slow down production until risks could be studied."

Deepfake porn could be a growing problem amid AI race - "Artificial intelligence imaging can be used to create art, try on clothes in virtual fitting rooms or help design advertising campaigns."

Analysis

AI Can Write a Song, but It Can’t Beat the Market - "Where is Wall Street’s AI revolution? Almost every industry, from architecture to entertainment, is testing generative artificial intelligence, hoping to profit from a technology that can produce writing, images and art much like humans."

A.I. Is Coming for Lawyers, Again - "More than a decade ago, lawyers were singled out as an endangered occupational species, their livelihoods at risk from advances in artificial intelligence. But the doomsayers got ahead of themselves."

Who Owns a Song Created by A.I.? - "Artificial intelligence tools that generate text, images and music are moving art into new territory — and that’s raising tricky questions for the business of creativity. For early adopters like Insider, the publication that this week announced an experiment with A.I."

Policy

China proposes measures to manage generative AI services - "HONG KONG, April 11 (Reuters) - China's cyberspace regulator unveiled draft measures on Tuesday for managing generative artificial intelligence services, saying it wants firms to submit security assessments to authorities before they launch their offerings to the public."

Eric Adams hails arrival of robot police dog called ‘creepy’ over 2 years ago: ‘Digidog is out of the pound’ - "The new devices, which also include a GPS tracker for stolen cars and a cone-shaped security robot, will be rolled out in a manner that is “transparent, consistent and always done in close collaboration with the people we serve,” said police Commissioner Keechant Sewell, who joined Mayor Eric Ad"

Baidu sues Apple, app developers over fake Ernie bot apps - "Chinese search engine giant Baidu (9888.HK), has filed lawsuits against "relevant" app developers and Apple Inc (AAPL.O) over fake copies of its Ernie bot app available on Apple's app store."

Expert Opinions

Why I am Not An AI Doomer - "I almost hate to add to the AI discourse these days, because I’m not confident my two cents of opinion is productive or helpful to anyone, but I’m finally giving into the peer pressure and laying out my views in one place."

Will LLMs learn more Agency in Doing? - "As large language models become multi-modal, they also begin to take on more functions of chain-of-prompting, agency and auto-GPT like capabilities. As LLMs go from billions to trillions of parametres many of us expect GPT-5 like LLMs to display AGI-like abilities."

The Surprising Thing A.I. Engineers Will Tell You if You Let Them - "Among the many unique experiences of reporting on A.I. is this: In a young industry flooded with hype and money, person after person tells me that they are desperate to be regulated, even if it slows them down. In fact, especially if it slows them down."

Explainers

The LLama Effect: How an Accidental Leak Sparked a Series of Impressive Open Source Alternatives to ChatGPT - "Edge 281: Our series about federated learning(FL) continues with an overview of cross-device FL, Google’s research about FL and differential privacy and the FedLab framework for FL simulation. Edge 282: We deep dive into LangChain, the uber popular framework for LLM-based development."

Fun

Without a Body, ChatGPT Will Never Understand What It's Saying - "When we asked GPT-3, an extremely powerful and popular artificial intelligence language system, whether you’d be more likely to use a paper map or a stone to fan life into coals for a barbecue, it preferred the stone."

AI art freaks me out. So I tried to make some. - "If Balenciaga Harry Potter is the future of entertainment, could Wes Anderson Lord of the Rings be its next chapter? (Spoiler: No!) In the past month, not one but two pieces of AI-generated content featuring the fashion brand Balenciaga went viral."

‘Slumdog Billionaires’: AI artist reimagines how world’s wealthiest would look if they were poor - "With the wide availability of artificial intelligence-backed art tools, people are able to recreate hyperrealistic photos against all kinds of backdrops."

What Is AI Doing To Art? - "In 1835, William Henry Fox Talbot finally succeeded in producing a crude photograph of his country estate. He triumphantly declared that his was the first house ever known to have drawn its own picture."

---

Copyright © 2023 Skynet Today, All rights reserved.