AI Shocker: Mind-Blowing Experiments, Disturbing Recalls, and Unbelievable Extinction Events Unveiled This Week!

Keeping up with an industry that moves as fast as AI it is a difficult task. So until an AI can do it for you, here’s a helpful roundup of the past week’s stories in the world of machine learning, along with notable research and experiments we didn’t cover on their own.

YouTube has started experimenting with AI-generated summaries for videos on the watch and search pages, though only for a limited number of English-language videos and viewers.

Certainly, abstracts could be useful for discovery and accessibility. Not every video creator can be bothered to write a description. But I do worry about the possibility of errors and biases embedded by the AI.

Even today’s best AI models tend to “hallucinate.” OpenAI freely admits that its latest text generation and summarization model, GPT-4, makes major errors in reasoning and fabricates “facts”. Patrick Hymel, an entrepreneur in the health technology industry, I wrote about the ways in which GPT-4 creates references, facts and figures without any identifiable link to actual sources. and fast company tested the ability of ChatGPT to summarize articles, finding it… pretty bad.

One can imagine AI-generated video summaries skyrocketing, given the added challenge of analyzing video content. It’s difficult to assess the quality of YouTube’s AI-generated roundups. But it is well established that AI is not that good to sum up text content.

YouTube subtly acknowledges that AI-generated descriptions are not a substitute for real ones. On the support page, he writes, “While we hope these roundups are helpful and give you a quick overview of what a video is about, they’re not a replacement for video descriptions (which are written by the creators!).”

We hope the platform doesn’t roll out the feature too quickly. But considering Google’s half-baked AI product launches of late (see its attempt at rival ChatGPT, Bard), I’m not too confident.

Here are some other standout AI stories from the past few days:

Dario Amodei is coming to Disrupt: We’ll interview the Anthropic co-founder about what it’s like to have so much money. And AI stuff too.

Google Search gets new AI features: Google is adding contextual images and videos to its AI-powered system generative search experiment the AI-powered generative search feature announced at the I/O conference in May. With the updates, SGE now shows images or videos related to the search query. The company is also reportedly spinning your project Wizard to a generative AI similar to Bard.

Microsoft kills Cortana: Echoing the events of the Halo game series from which the name was drawn, Cortana has been destroyed. Fortunately, this was not a rogue general AI, but an also executed digital assistant whose time had come.

Meta embraces AI generative music: Goal announced this week audio crafta framework for generating what it describes as “high quality” and “realistic” audio and music from brief text descriptions or prompts.

Google pulls out AI Test Kitchen: Google has pulled its AI Test Kitchen app from the Play Store and App Store to focus solely on the web platform. The company launched the AI ​​Test Kitchen experience last year to allow users to interact with projects powered by different AI models, such as LaMDA 2.

Robots learn from small amounts of data: On the subject of Google, DeepMind, the tech giant’s AI-focused research lab, has developed a system that it claims allows robots to effectively transfer concepts learned in relatively small data sets to different scenarios.

Kickstarter enacts new rules around generative AI: Kickstarter announced this week that projects on its platform that use AI tools to generate content will be required to disclose how the project owner plans to use the AI ​​content in their work. Additionally, Kickstarter requires new projects involving the development of artificial intelligence technology to detail information about the training data sources the project owner intends to use.

China cracks down on generative AI: Several generative AI apps were removed from Apple’s App Store in China this week, thanks to new rules that will require AI apps operating in China to obtain an administrative license.

Stable Diffusion launches new model: Stability AI released Stable Diffusion XL 1.0, a text-to-image model that the company describes as its “most advanced” version to date. Stability claims that the model’s images are “more vibrant” and “accurate” colors and have better contrast, shadows, and lighting compared to its predecessor’s artwork.

The future of AI is video: Or at least a large part of the generative AI business is, as Haje says.

AI.com has switched from OpenAI to X.ai: It’s unclear if it’s been sold, rented, or is part of some kind of ongoing scheme, but the coveted two-letter domain (which is likely worth $5-10 million) is now pointing to the X.ai research team of Elon Musk instead of the ChatGPT interface. .

Other machine learning

AI is finding its way into myriad scientific domains, as I have the opportunity to document here on a regular basis, but you can be forgiven for not being able to list more than a few specific applications in advance. This review of the literature in Nature it is as comprehensive a description of the areas and methods in which AI is having an effect as you are likely to find anywhere, as well as the advances that have made them possible. Unfortunately, it’s paid, but you can probably find a way to get a copy.

A deeper dive into the potential of AI to enhance the global fight against infectious disease can be found. here in scienceand some takeaways in the UPenn summary. One interesting part is that models created to predict drug interactions could also help “unravel the complex interactions between infecting organisms and the host’s immune system.” The pathology of the disease can be ridiculously complicated, so epidemiologists and doctors are likely to accept any help they can get.

Another interesting example, with the caveat that not all algorithms should be called AI, is this multi-institutional work algorithmic identification of “potentially dangerous” asteroids. Sky surveys generate a ton of data and sorting through it for faint signals like those from asteroids is hard work that is very amenable to automation. The 600ft SF289 2022 was found during a test of the algorithm in ATLAS data. “This is just a small taste of what you can expect from the Rubin Observatory in less than two years, when HelioLinc3D discovers an object like this every night,” said Mario Jurić of the University of Washington. I can not wait!

A kind of halo around the world of AI research is the research being done. in AI: how it works and why. These studies are generally quite difficult for non-experts to analyze, and this one from the ETHZ researchers is not an exception. But the main author, Johannes von Oswald, also did an interview explaining some of the concepts in simple language. It’s worth reading if you’re curious about the “learning” process that happens inside models like ChatGPT.

It is also important to improve the learning process, and as these Duke researchers find, the answer is not always “more data”. In fact, more data can hamper a machine learning model, said Duke professor Daniel Reker: “It’s like if you trained an algorithm to distinguish between images of cats and dogs, but gave it a billion photos of dogs to learn and only a hundred photos of cats. The algorithm will be so good at identifying dogs that everything will start looking like a dog and it will forget about everything else in the world.” His approach used an “active learning” technique that identified such weaknesses in the set of data and proved to be more effective by using only 1/10 of the data.

A University College London study found that people could only distinguish real speech from synthetic 73 percent of the time, both in English and Mandarin. We’ll all probably get better at this, but in the short term the technology will probably outpace our ability to detect it. Stay ice cold out there.