The week in AI: The pause request heard ’round the world

Keeping up with an industry that is changing as rapidly as AI is a tall order. In the meantime, here’s a handy wrap-up of last week’s articles in the world of machine learning, along with notable research and experiments we didn’t cover alone.

In one of the most surprising stories of the past week, the Italian Data Protection Authority (DPA) blocked OpenAI’s viral AI-powered chatbot ChatGPT, citing concerns that the tool violates the EU’s General Data Protection Regulation. Did. The DPA has reportedly launched an investigation into whether OpenAI has processed people’s data unlawfully and that it has no system in place to prevent minors from accessing its technology.

It is unknown what the outcome will be. Please allow 20 days for OpenAI to respond to your order. However, the DPA move could have significant implications for companies deploying machine learning models not only in Italy but anywhere in the EU.

As Natasha In her post on the news, she said many of OpenAI’s models were trained with data scraped from the internet, including social networks like Twitter and Reddit. Assuming the same applies to ChatGPT, the company may be violating the GDPR across blocks, as it doesn’t appear to notify the owners of the data it reuses to train its AI. .

The GDPR is just one of many potential legal hurdles facing AI, especially generative AI (AI that generates text and art like ChatGPT). With every installation, it becomes clear that the dust takes time to settle. But that doesn’t scare VCs who continue to pour capital into technology as if there’s no tomorrow.

Will they prove to be smart investments or liabilities? But don’t worry, we’ll let you know if anything happens.

Other notable AI headlines from the past few days include:

• Ads appear in Bing Chat. Microsoft said last week that it was “exploring” inserting ads into responses from its search agent, Bing Chat, which uses OpenAI’s GPT-4 language model. As Devin points out, sponsored answers are clearly labeled as such, but it’s a new, potentially more destructive form of advertising that can easily be delineated or ignored. You can’t. Moreover, you may further undermine trust in your language model. The language model has already made enough factual errors to sow doubt on the veracity of the response.
• Pause request: A letter signed by more than 1,100 people, including Elon Musk, was released Tuesday, calling on “all AI labs to immediately suspend training AI systems stronger than GPT-4 for at least six months.” . However, the situation surrounding it was darker than expected. Some of the signatories have since retracted their positions, but reports have revealed that other prominent signatories, such as Chinese President Xi Jinping, were found to be fake.
• And the response to the suspend request: A prominent AI ethicist points out that worrying about distant hypothetical problems is dangerous and self-defeating if we don’t address the problems AI contributes to today.
• Twitter reveals its algorithm. as promised repeatedly According to Twitter CEO Elon Musk, Twitter is Open Portions of our source code, including the algorithms we use to recommend Tweets on users’ timelines, are subject to public inspection. Interestingly, Twitter appears to use in part a neural network continuously trained on tweet interactions to rank tweets and optimize for positive engagement such as likes and replies. But it has many nuances. researcher dig to the codebase notes.
• AI meeting summary: Following companies like Otter and Zoom, meeting intelligence tool Read has introduced a new feature that cuts hour-long meetings into two-minute clips with key pointers. The company says it uses large-scale language models, which are combined with video analytics to extract the most notable parts of meetings. This is a useful feature.

---

Other machine learning

AI enabler Nvidia’s Bionemo is an example of their new strategy. The advances aren’t all that new, but they’re becoming more and more accessible to businesses. The new version of this biotech platform adds a sleek web UI and better tweaks to the suite of models.

Amgen’s Peter Grandsard, who leads research using AI technology, said: “We are looking to make our operations more efficient in research, just as we are in manufacturing. With the acceleration that technologies like Nvidia provide, what we were able to do in one project last year can now be done in the same technology. You can do 5 or 10 things with an investment of .”

This excerpt from Meredith Broussard of Wired is worth reading. She was intrigued by her AI model used to diagnose cancer (which she’s fine with), but it’s very tedious and frustrating trying to own and understand that data and process. I was. Medical AI processes clearly need to consider patients more.

In fact, malicious AI applications pose new risks, such as trying to influence discourse. We’ve seen what GPT-4 can do, but whether such a model can produce effective persuasive texts in a political context has been an open question. Research suggests that. When people were exposed to essays arguing about issues such as gun control and carbon taxes, “AI-generated messages were at least as persuasive as human-generated messages on all topics.” These messages were also perceived as more logical and factual. Will AI-generated text change someone’s mind? It’s hard to say, but it seems very likely that people will use it more and more for this kind of agenda.

Machine learning is being used by another group at Stanford University to better simulate the brain, including the tissue of the organ itself. The brain is not just complex and heterogeneous, but “much like her Jell-O, which makes both testing and modeling of physical effects on the brain very difficult,” she says, Professor Ellen Kuhl News. explained in her release. Their new model selects among thousands of brain-modeling techniques, combines and combines them, and identifies the best way to interpret or project from given data. This won’t reinvent brain injury modeling, but it should make that research faster and more effective.

In nature, a new Fraunhofer approach to seismic imaging applies ML to existing data pipelines processing terabytes of output from hydrophones and airguns. Typically, this data must be simplified or abstracted, losing some precision in the process, but new ML-powered processes can analyze unsummarized datasets.

Interestingly, the researchers noted that while this would normally be a boon for oil and gas companies looking for deposits, moving away from fossil fuels could also help identify potential CO2 sequestration sites and potentially damage gas. It could be aimed at more climate-friendly ends, such as identifying potential giveaways, he said. build up.

Forest monitoring is another important task for climate and conservation research, and measuring tree size is one of them. However, this task requires you to manually check each tree. A team from the University of Cambridge built an ML model that uses smartphone lidar sensors to estimate trunk diameter and trained it on a large number of manual measurements. Just point your phone at the trees around you and boom. The system is more than four times faster than hers, yet more accurate than expected, said Amelia Holcomb, lead author of the study. Sometimes I like to venture into particularly dense forests and oddly shaped trees, and I don’t think there’s a way to make it work, but it does. ”

Fast and requiring no special training, the team hopes it will be widely released as a way to collect data for tree surveys or to make existing efforts faster and easier. . Android only for now.

Finally, enjoy this interesting investigation and experimentation by Eigil zu Tage-Ravn. See how a generative art model creates his famous painting of the Spouter-Inn, described in Chapter 3 of Moby-Dick.