The AI Daily Brief: Artificial Intelligence News and Analysis - The 5 Biggest AI News Stories This Week

Episode Date: May 27, 2023

AI news this week - here are the 5 most important stories!  AI regulation and safety - OpenAI, Microsoft and more AI helps discover antibiotic for superbug Adobe generative fill Microsoft Windows Copilot ChatGPT Browse with Bing for all users The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI Breakdown weekly recap, the top five news stories from AI this week. We cover the latest regulatory news, Adobe's generative fill, plus all the announcements from Microsoft Build. The AI breakdown is a daily video and podcast all about the most important news and discussions in AI. Like, subscribe and share, and go to Breakdown.network for more information. Welcome back to the AI breakdown. Today we are doing the weekly recap, and we are counting down the five most important stories in AI this week. and we start with a set of stories that are all focused on AI safety and AI regulation. We kick it off with OpenAI's comments on the EU, which have shifted pretty dramatically over the course of the week. Interviewed during his world tour, Sam Altman, the CEO of OpenAI, had earlier said that if the EU proceeded with its AI act,

it might create a situation where OpenAI and their chat GPT had to leave European shores. That created quite a backlash among EU lawmakers and led ultimately to Sam Altman tweeting, very productive week of conversations in Europe about how to best regulate AI. We are excited to continue to operate here and, of course, have no plans to leave. Microsoft also threw their hat in the AI regulatory ring. Microsoft President Brad Smith gave a long speech in Washington, D.C. this week where he laid out a five-part plan for regulating AI. Smith said that his biggest concern when it came to AI was deepfakes and the potential havoc that they could wreak on public discourse, and he said that Microsoft supported the type of AI licensing regime that Sam Altman had talked about in the recent congressional hearing.

Describing those licenses to deploy quote unquote highly capable AI, he said, that means you notify the government when you start testing. You've got to share results with the government. Even when it's licensed for deployment, you have a duty to continue to monitor it and report to the government if there are unexpected issues that arise. Now, on the one hand, Microsoft here might be showing that they are a responsible corporate citizen with an extremely powerful new technology, but on the other hand, this is, of course, igniting more accusations of corporate ladder pulling and attempts at regulatory capture. Now, beyond the big existential risk questions of AI, there are also smaller questions that still matter, such as questions of political and racial bias. Open AI has launched a new million-dollar grant program for what they call democratic inputs to AI.

They'll be awarding $10,000 grants for people who come up with compelling answers to questions such as, under what conditions should AI systems condemn or criticize public figures, given different opinions across groups regarding those figures? How should disputed views be represented in AI outputs? Should AI by default reflect the persona of a median individual in the world, the user's country, the user's demographic, or something entirely different? The point they say is no single individual, company, or even country should dictate these decisions. Now, when it comes to existential risk, there are also growing conversations in that domain as well. Elon Musk and Eric Schmidt both discussed AI X-risk at the recent Wall Street Journal CEO Council Summit with Elon Musk making very widely

reported comments that there was a non-zero chance of AI going Terminator. 2018, Turing Award winner, Joshua Benjillo also came out with a new essay called How Rogue AIs May Arise. Now, if Benjillo is sounding the warning, other people are thinking about how to actually put into practice some solutions. Google DeepMind, for example, just released research that they call an early warning system for novel AI risks. They write, to pioneer responsibility at the cutting edge of AI,

research, we must identify new capabilities and novel risks in our AI systems as early as possible. As the AI community builds and deploys increasingly powerful AI, we must expand the evaluation portfolio to include the possibility of extreme risks from general purpose AI models that have strong skills in manipulation, deception, cyber offense, and other dangerous capabilities. Now, importantly, this new research involves not only Google DeepMind, but also academics from the University of Cambridge, the University of Oxford, the University of Toronto, as well as contributions from OpenAI, Anthropic, and many more. Number four, on our list of the most important AI stories this week,

is AI's use in helping discover a new antibiotic that could help treat a dangerous superbug. So the bacteria in question, and forgive my pronunciation, was Astyne Tobacter Bhamene, which is the bacteria that tends to cling to surfaces in hospitals and other healthcare settings like doorknobs encounters. The WHO has classified it as among the most serious risks to humanity because it has properties where it's able to, as CNN puts it, grab bits of DNA from other organisms it comes into contact with, thus incorporating their best weapons, genes that help them resist agents, doctors used to treat them.

In a new study, researchers used AI as part of the process to try to discover a drug that could actually deal with a dangerous pathogen. They exposed colonies of the bacteria to some 7,500 agents including drugs and the active ingredients in drugs, and then used information about the 480 compounds that blocked the growth of the bacteria to train a new AI model. John Stokes, one of the researchers then said, once we had our model trained, what we could do then is start showing that model brand new pictures of chemicals that it had never seen, right? And based on what it had learned during training, it would predict for us whether those molecules were antibacterial or not. Over the course of a couple hours, AI was able to then screen more than 6,000 new molecules.

From there, they were able to narrow the search to 240 chemicals which were tested in the lab, and from there they found nine that had antibacterial properties. One of those, which researchers named Abosin, seemed to work only with this specific bacteria, exactly the type of property they wanted out of an antibiotic. Now, the reason that this matters outside of just being genuinely interesting is that when people talk about slowing down or turning back the tide on AI, often they leave out this type of potential capacity. Sam Demutrio says, AI has helped discover a new superbug killing antibiotic. Anyone calling to slow down progress in AI should remember there are massive cost to inaction. Third on our list of the biggest stories in AI this week

is Adobe Photoshop's new generative fill feature. Adobe announced earlier this week that Firefly was being integrated into Photoshop, and the main feature that they discussed was generative fill. Generative fill is a feature that allows you to select something in Photoshop. It could be an object in a photo, it could be an entire scene or setting, and either add or subtract elements in a precise way using text to image generation. Now, this can allow for image enhancement. It can allow for image modification. It can allow people to create entirely new worlds and scenes with just a few clicks. There are very basic use cases, such as removing objects from a photo, but then there are also more advanced use cases like combining images to create new worlds. Now, the reason I have this at number

three is that I think a lot of what takes AI mainstream is use cases that happen day and day out and don't require some specific experimental task. Millions and millions of people already use Photoshop for photo enhancement and modification. And the fact that this tool now lives natively inside that program means that a huge, huge number of artists and creators and marketers and designers are going to now use generative AI for the first time. Number two on our list of Most important AI news of the week is the announcement of Windows copilot. Microsoft held its build conference this week, and just like Google's I.O. a couple weeks ago, it was filled chockerblock full of AI announcements.

And I think one of the most significant announcements was the fact that copilot, a feature which had been debuted in GitHub, was coming in a broader application to Windows 11. Nvidia AI scientist Dr. Jim Fan says three months ago, I said Windows will be the first AI-first operating system. Surely Microsoft delivers with a sharp vision and a steady hand. To me, Windows co-pilot is already a way bigger deal than BingChat. It's becoming a full-fledged ailent that takes actions on the OS and native software level, given your computer state and local files as input. Windows is on the right track, the first Jarvis is around the corner.

Now, in much the same way that I said that Photoshop was such an important announcement because it would introduce millions and millions of regular normal people to using generative AI tools, Windows 11 copilot is like that but on steroids. This is a major operating system, one of the two biggest operating systems that people use day and day out. So to have integrated AI that just sits at the core of your operating system experience and that natively integrates with all of the existing files and information that you interact with on your computer is a pretty big deal. Samo Berger writes, The first LLM-centered operating system will be out in June and it's Windows 11.

Forget learning an interface, just ask copilot to have the computer do it. you want. People underestimate Microsoft at their own peril. It seems set to beat Apple. Now, while by and large, people agreed with the assessment and were very excited to see Windows copilot, that wasn't universally true. Sully Omar wrote a threat about why he thinks that co-pilot is underwhelming and why it suggests to him that Apple is likely to win. Sully writes, Microsoft just announced that they are adding Windows co-pilot their AI operating system. It looks great at a first glance, and you'd think Apple is in trouble. But that's far from the truth. While Windows AI OS might be cool, Apple's Mac OS may have already won.

Here's why. He says first up the new Windows AI OS. From the demo video, it's co-pilot like OS, basically clipy, lets you edit, summarize, and create documents across Windows, can use a chatbot to change settings like dark mode or turn Bluetooth on, send images to teams. Microsoft has done a killer job at adopting AI at an absolutely insane speed. They were the first on OpenAI investment, AI and Bing, AI tooling inside Office 365, and way more. But here's the issue.

Microsoft starts to fall flat. The biggest problem with them can be boiled down to user experience, hardware and privacy. Microsoft has a killer software team, no doubt, but they're pretty questionable when it comes to those three areas. Now let's take a look at Apple for a second. The most powerful thing they have is Apple Silicon. Powerful chip means they can run local LLMs directly on your computer, no latency or privacy concerns. Why does that matter? Well, firstly, latency is a huge factor when using LLMs, so having an extremely optimized chip in over a billion devices worldwide probably helps. You can can run a local model for each user. Minimal latency, no network needed, and most importantly, private. What about the average Windows user? Well, for the most part, 99% don't have gaming PCs. So if you

want quick inference, it has to be done through the internet. And there's one problem with that. Even if Microsoft fixes latency, they'll never have true privacy because they have to send the request to their servers. They will have your data no matter what somewhere, even though it's very secure and adheres to all regulation. Apple, on the other hand, dominates privacy from a brand perspective. They spend billions doing it and have built trust with their users. Eventually, we're going to have OS-level LLMs that can read every email and text everything. Would you trust Apple or Microsoft? And Sully goes on, but I'll stop there just because it's getting a little bit long.

But this is a great thread, really interesting. And even if you disagree with Sully's conclusions, you will definitely come out having a better sense of how the operating system wars might play out over the next couple years. Number one, last up on our list of most important AI events this week is the transformation of search. And there are two big stories that relate to this topic. The first is that as part of that same Microsoft event, Microsoft Build, they also announced that Bing would be natively integrated with ChatGPT. Now, you remember that up until about two weeks ago, the average ChatGPT experience ended on data trained in the end of 2021.

So if you had more contemporary use cases, you had to use some other tool that was actively connected to the internet. When ChatGPT started rolling out its browse feature to its plus users that changed things for people who are willing to pay $20 a month. But there are lots of people out there who either weren't willing or just aren't able to pay that fee. Microsoft's announcement was that Browse with Bing would be rolled out to all users of chat GPT, not just plus users, making chat GPT natively connected to the internet for all 100 million plus people who are using it on a daily basis. If you've used chat GPT with browse and chat

GPT without browse, you will definitely understand just how many more use cases there are available to you when you have connection to the existing internet. But of course, as fast as Microsoft's moving, they're not the big player when it comes to search. That player is Google. At their IEO developer conference a couple weeks ago, we learned that Google was in the midst of the biggest transformation of search in the company's 20-plus year history. What was clear was that generative AI was going to reshape how they thought about search. We were moving past the world of just 10 blue links into something that had a much more curated set of answers on top of those links. And what that meant for everything from SEO to marketing to content discovery has remained up in the air. While all those questions haven't been answered, Google is now rolling out access to search labs, which is their labs portfolio for new search.

search experiments. Users can sign up to be on the wait list, and some are already starting to get access to what they call their search generative experience or SGE. In their blog post announcing the waitlist, they give a number of different benefits of SGE, including easier and more contextual search for products, and being able to get up to speed much more quickly than you might otherwise be able to when it comes to a new topic you're trying to learn about. Now, as people get access to the tools, they're frantically trying to figure out what it means for all those domains I just mentioned, like SEO and marketing. But whatever they uncover and whatever the experience ultimately ends up being, there is no doubt that this is the biggest shift in the search experience on the internet since

Google first came out in the first place. That's why between Google's SGE and chat GPT, adding browse with Bing for all users, the transformation of search is the most important AI story this week. All right, guys, that's it for this AI breakdown weekly recap. If you're enjoying the AI breakdown, please go like, subscribe and share it, as well as following the podcast and the newsletter. wherever you are, I hope you are having a great Memorial Day weekend. And until next time, peace.