How Disinformation Agents Are Using ChatGPT

0:00Today on the AI Daily Brief, how disinformation agents are using AI right now. Before that in the headlines, Google attempts to explain why it told us to go eat rocks. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:24Welcome back to the AI Daily Brief Headlines Edition, all the AI daily news you need in around five minutes. We kick off today with Google discussing its pretty problematic rollout of AI overviews. The Verge writes, Google defends AI search results after they told us to put glue on pizza. Google blames data voids and edge cases for its bad AI-generated search results. Basically, after a couple weeks of bad headlines, Google posted a blog post on Thursday by their head of search, Liz Reed. The Verge writes, the tech giant blames data voids for the inaccurate results, along with people making up odd questions, and double down by claiming that AI results are leading to higher satisfaction with search.

0:59Reed argues that AI overviews generally don't hallucinate, they just sometimes misinterpret what's already on the web. Reid writes, There's nothing quite like having millions of people using the feature with many novel searches. We've also seen nonsensical new searches seemingly aimed at producing erroneous results. She also tried to argue that many of the screenshots that have been shared around are fake. Now, I understand how frustrating it can be for a product company to roll something out, have millions and millions of people use it really positively and successfully, but then to have everyone focus on the edge cases that are bad.

1:27Unfortunately, however, that's sort of part of the cost of doing business. Especially for the global leader in search, you have to expect that some of that is going to happen. What's more, thinking adversarially, there's no universe in which people aren't going to try to break systems. Anyone who's building any sort of AI technology just has to understand and accept that that is a key constraint, that that will be a factor in how they roll things out. Now, Google says that they've built better detection mechanisms for nonsensical queries, updated their systems to limit how much user-generated content comes in responses that could offer misleading advice.

1:58But ultimately, the proof will be in the pudding, and it's yet another challenge for Google to overcome that's contributing to a broader sense of where they sit in the AI ecosystem. Next up, one of the big questions for 2024 is the extent to which predictions of AI election apocalypse via misinformation will come true. According to a new survey, people certainly seem to think that it will. An Elon University poll found that 73 % believe that AI will be used to manipulate social media to impact the election. 70 % said that it's likely that the election will be affected by the use of AI to generate fake information.

2:2862 % said the election is likely to be affected by the targeted use of AI to convince certain voters not to vote. Overall, 78 % of Americans expect at least one of these types of AI abuses will impact the election, with over half thinking all three are at least somewhat likely to occur. I think these results actually show something interesting, however. One of the big questions for me is whether people's anticipation of being targeted with deepfakes will make them less susceptible to deepfakes. It's hard for me to imagine that being on guard doesn't help. But at the same time, 69 % of those surveyed said that they aren't confident that most voters will be able to detect fake media.

3:03And 52 % aren't confident in their own ability to detect fake media. Whatever happens this year will set the baseline for our expectations for AI in elections going forward. Over in markets, a really interesting thing this week, which probably deserves a full episode soon. Reuters writes Salesforce plummets as weak forecasts sparks concerns of AI competition. Salesforce is down about 20%. Anominally, the cause was its lower ever quarterly revenue growth forecast, but the underlying cause were fears that high interest rates and rival AI offerings were hampering demand. Indeed, Morgan Stanley analysts said, Weak bookings in Q1 further test investor patience as the Gen.AI innovation cycle has yet to inflict top-line results and now increasingly becomes a point of competitive concern.

3:45In other words, so far, Wall Street isn't necessarily seeing huge impact of AI on the bottom line, at least not with these professional services companies. It's showing up more in cloud companies. And on top of that, AI competition against traditional SaaS is becoming more of a focus. This could be a big and important theme going forward, so it's something we're definitely going to watch. Lastly today, something that is absolutely going to get a full episode next week. You might remember those viral AI-generated South Park clips from last year. Well, now the studio behind that, Fable, has announced a streaming platform they call Showrunner that allows users to create their own content.

4:20The Hollywood Reporter sums up the feeling of the entertainment industry about this with a headline that reads, Hollywood Nightmare? I think that this is actually an early piece of evidence of something that I am tracking closely, which is a bet that in the future, entertainment becomes hyper-personalized and the ability to create more things faster leads us to have an even more segmented content and entertainment landscape where we are in control of basically endlessly remixing the IP that we like into a never-ending playlist of stories that suit our interests and moods. We'll dig more into that next week, but for now, that is going to do it for the headlines.

4:52Up next, the main episode. Today's podcast is brought to you by Plum. If you're a startup building AI features for your customers, you're probably feeling the pain of hallucination, prompt testing, unstructured responses, subpar queries for embeddings, and of course, the mind-numbing process of general iteration and refinement when your engineers have to make every change by hand. That's where Plum comes in. Plum is a no-code AI app builder designed for product teams who care about quality and speed. What is taking you weeks to hand code today can be done confidently in hours. Check out useplum.com, U-S-E-P-L-U-M-B.com, or reach out to me for early access.

5:32Hello, friends. Before we get back to the episode, I want to tell you about something special I'm doing on Super Intelligent this June. Super is, of course, our platform for AI learning, and I've heard from a lot of you that you really want something for a true AI beginner, someone who's really just getting their feet wet with these tools. So what I'm going to do is put together basically a course that sits on top of and uses super intelligent tutorials and lessons, but where I hand guide you through around 10 different lessons and how-tos that I think once you complete them will have you ahead of 80 % of the other people who are just starting to use AI right now.

6:02If you are interested in this learning experience, go to besuper.ai and sign up using code June. You'll get 25 % off your first month and I'll automatically add you to that AI for Beginners group. That's bsuper.ai, discount code JUNE. See you there. Welcome back to the AI Daily Brief. Today we kick off with a story in the geopolitics of AI. If you've been listening to this show for any amount of time, you will know that one of the really interesting things about artificial intelligence right now is that it is not just a cool new technology that some businesses are adopting and consumers are talking about.

6:35It is something that is being seen as a key geostrategic imperative. AI is, for example, having huge impact on U.S. policy vis-a-vis China, and by extension, U.S. policy in the Middle East. Well, interestingly, OpenAI has just reported that a number of different states have been using AI for fairly nefarious purposes. The New York Times reports OpenAI says Russia and China used its AI in covert campaigns. TLDR, OpenAI announced on Thursday that it had identified five different online campaigns that had used its tools to, quote, deceptively manipulate public opinion and influence geopolitics. These efforts were run by both private companies as well as state actors connected to Russia, China, Iran, and Israel.

7:13So what were they doing? Well, they were using OpenAI to generate social media posts, translate and edit articles, write headlines, basically to create propaganda in so many words. This will probably not be surprising to any of you. This seems like an obvious use case for someone who's engaged in a propaganda effort, which is of course as old as time. But as the Times points out, OpenAI's report is the first time that a major AI company has revealed how its specific tools were used for such online deception. One of the interesting things though, is that the report also suggests that people are just still sort of fumbling around with it.

7:43Said Ben Nimmo, a principal investigator for OpenAI, while the campaigns were used to post political content, OpenAI had a hard time figuring out if they were trying to target specific elections or more just generally about getting people all riled up. He also noted that they hadn't been very effective. Said Nimmo, these influence operations still struggle to build an audience. So what are some of the examples? In one case, the campaign used OpenAI tech to generate anti-Ukraine comments and then translate them into English, French, German, Italian, and Polish. In another case, they were used to translate and edit articles supporting Russia in the war in Ukraine.

8:14They were also used to convert anti-Ukraine articles into Facebook posts. In a different campaign, people in the Ukraine, Moldova, Baltic states, and the United States were targeted via Telegram, focusing once again on the war in Ukraine. Writes the Times, The political comments received few replies and likes, and the efforts were also unsophisticated at times. At one point, the campaign posted texts that had obviously been generated by AI. said one post, As an AI language model, I am here to assist and provide the desired comment. The report also talked about how China sought advice on how to analyze social media and research current events, as well as to generate social media posts that disparaged people who had been critical of the CCP.

8:50Iran used ChatGPT to translate long-form articles that were aimed to spread pro-Iranian, anti-Israel, and anti-U.S. sentiment. Our allies were at it as well, though. For example, an Israeli firm that manages political campaigns used ChatGPT to generate fictional personas and biographies that were used in Israel, Canada, and the United States to post anti-Islamic content. Ultimately, the report was sort of a mixed bag. It showed that a thing that people have been worried about, which is AI-enabled influence operations, are definitely happening, but they're not really sophisticated yet. I don't think that that will necessarily make anyone feel more comfortable, given that ultimately, if this is just a learning curve issue, as well as a technological capacity issue, both of those things are likely to get nothing but better in the time to come.

9:28I mentioned at the top of the show the way that AI is influencing US foreign policy and how it's expanding from China to the Middle East. Bloomberg recently published a piece called U.S. is slowing AI chip exports to Middle East by NVIDIA and AMD. So one of the things that's happened over the last couple years is in addition to chip export restrictions targeted at China, some of those same chip restrictions have found their way to the Middle East. That's not necessarily because the U.S. is concerned about rogue Gulf states or anything like that having access to AI technology, but that they're concerned that it could be a backdoor way for China to get access to the technology through their relationships in the region.

10:00Because of that, they've been in a licensing regime, and apparently the issuing of licenses has slowed down significantly. Apparently, there is a national security review of AI development in the Gulf region going on right now. Writes Bloomberg, it's unclear how long the review will take, nor is there a concrete definition of what constitutes a large shipment. Officials are particularly focused on high-volume sales, as countries including the UAE and Saudi Arabia look to import massive quantities of the chips used in AI data centers. Bloomberg continues, U.S. officials have delayed or not responded to license applications in the past several weeks.

10:30That includes attempts to sell to customers in the UAE, Saudi Arabia, and Qatar, according to a source. Maybe the best example of the U.S.'s growing focus on this area came in April, when the Department of Commerce facilitated a$1.5 billion investment from Microsoft into Abu Dhabi-based G42. G42 had been at the very epicenter of this concern around the Middle East playing both the Chinese and U.S. sides of the AI equation, and had come under increasing pressure to pick a side. Towards the end of last year, G42 started to basically say that they couldn't remain neutral and they had to pick a side and that they were going to pick the U.S.

11:01And this minority investment seems to have been part and parcel of that. In other words, it feels like a way to give Microsoft, a U.S. corporation, access to G42's books, which is probably, frankly, a way for the U.S. government to keep tabs on what G42 is doing. It's not just the U.S.-China Middle East access that has AI at the center of its politics, though. CNBC this morning published a piece called France is Aiming to Become a Global AI superpower. The article talks about how France has produced a number of big high-profile AI companies, including Mistral, as well as the firm H, which recently raised a$220 million seed funding round.

11:33But even with this, it shows just how integrated US-based big tech companies are into the entire world. We just talked about that G42 deal, which again was facilitated by the Department of Commerce, but these big tech companies also are making major investments in places like France. Microsoft, for example, committed 4 billion euros into the country. Said French finance minister Bruno Le Maire, France is the leader on artificial intelligence in Europe. And he also tried to clarify that while they wanted to work with U.S. tech companies, they, quote, want to have our own artificial intelligence being created and being developed in France.

12:03For example, as to the Microsoft deal, he said, Microsoft is much welcome in our country, but the challenge for us is to have our own devices, our own scientists, and we are working very hard for that. French President Emmanuel Macron put his ambitions for his country in global terms. He said, it's insane to have a world where the big giants just come from China and the U.S. Finally, there is, of course, another dimension to this as well, which is the military and defense dimension. The U.S. Army is no stranger to AI, and just this week, Palantir announced a new nearly half-billion-dollar deal from the Pentagon for an AI-powered intelligence platform.

12:34Said Shannon Clark, Palantir's head of defense growth, this is taking what has been built-in prototype and experimentation and bringing this to production. Users are going to span everyone from intel analysts and operators and some of the remote island chains across the world to leadership at the Pentagon. The system is called the Maven Smart System, which started deploying as a prototype back in 2021. Writes Defense One, the prototype will focus on key areas like battle space awareness, global integration, contested logistics, joint fires, and targeting. For example, Maven is able to aggregate data on U.S.

13:02forces locations from hundreds of sources into a digital map that leaders can use. Describing the difference between this and what they have now, Palantir's Enterprise Defense Lead Andrew Locke said, Typically how this is handled is through static update briefs, typically PowerPoint, on a 12 or 24 hour cadence. This actually becomes challenging for a leader to be able to either visualize, describe, or provide direction to forces when information is either outdated or they can't actually see it in a synthesized manner. So tons going on in the world of AI and geopolitics, but for now, that is going to do it for the AI Daily Brief.

13:30Appreciate you listening or watching as always. And until next time, peace.

13:44You