transcribe

The First AI Answer From Space Was a Haiku | Lightwork

Lightspeed Venture Partners · 24m · transcribed 14d ago
More from Lightspeed Venture Partners Business
𝕏 Share ▶ YouTube 📥 PDF 🤖 .md

Section Insights

# 0:00

AI in Orbit: Project Star Catcher

What is Google's Project Star Catcher and its significance?

Google's Project Star Catcher aims to establish AI data centers in orbit, marking a significant step in the integration of AI technology with space exploration. The project recently achieved a milestone by receiving its first AI response from a TPU in orbit, showcasing the potential for advanced computing in space.

  • Project Star Catcher represents a major technological advancement for Google.
  • The project involves collaboration with Planet and SpaceX.
  • The successful AI response indicates the feasibility of operating data centers in space.
# 4:53

Incremental Progress in Space Technology

What challenges remain for data centers in space?

While the first AI response from a TPU in orbit is a significant achievement, it highlights that the development of space-based data centers is incremental and faces various challenges. These include ensuring the durability of technology during launches and the complexities of operating in a vacuum.

  • The journey to space-based data centers is gradual and involves overcoming technical barriers.
  • Google's approach differs from competitors by focusing on a network of satellites rather than individual units.
  • Successful tests validate the operational capabilities of TPUs in space.
# 9:47

The Threat of AI in Cybersecurity

How is AI impacting cybersecurity and personal security?

AI advancements, particularly in deep fakes and voice synthesis, are exacerbating cybersecurity threats. As AI becomes more sophisticated, it increases the potential for psychological manipulation in hacking, making individuals more vulnerable to scams and phishing attempts.

  • Deep fakes and AI-generated content pose significant risks to personal security.
  • Human psychology plays a crucial role in the effectiveness of cyber attacks.
  • The intersection of AI and cybersecurity requires heightened awareness and new defense strategies.
# 14:40

The Rise of AI-Generated Content

What are the implications of AI-generated comments on social media?

The proliferation of AI-generated comments on social media platforms is leading to frustration among users. Many feel overwhelmed by the volume of low-quality, automated responses that dilute genuine engagement and discussion.

  • AI-generated content is increasingly saturating social media, impacting user experience.
  • Users prefer authentic interactions over automated responses.
  • Platforms are exploring ways to manage and report AI-generated content.
# 19:34

Consumer Behavior and AI Reviews

How is AI influencing consumer decision-making?

Consumers are increasingly relying on AI-generated reviews and summaries to make purchasing decisions, particularly for utility-oriented products. This shift reflects a desire for efficiency and reduced friction in the shopping experience.

  • AI is changing how consumers approach product reviews and decision-making.
  • There is a distinction between the use of AI for utility versus cultural commentary.
  • The demand for quick, reliable information is driving the adoption of AI tools.

Transcript

0:01 First AI response from a TPU in orbit from Google. >> I know, and it was a haiku. Keeps getting more sci-fi every time we talk about it. >> The biggest crypto theft in history. >> All of our passwords and cryptography are like are on the line. Claire, is the internet dying? >> I've had some deep fakes, so maybe. >> BuzzFeed journalism kind of just evaporated overnight. >> I think shopping decisions is probably in dangerous territory, but like cultural commentary I feel like hopefully is safe.

0:36 We're back with another big week on Light Work. Google's Project Star Catcher sent back its first AI answer from a satellite in orbit. Binance says attackers stole about $6.3 billion from the exchange. And in another cyber security attack, hackers remotely warmed supermarket freezers across Europe. And in a discussion around dead internet theory, Josh and I discussed how Reddit CEO says that Google isn't sending people to Reddit the way it used to. And why he thinks AI search is one of the biggest reasons. Let's get into it.

1:09 So some big things happening in orbit. >> Yeah, data centers in space could be real soon. >> Yeah. we just had our first AI response from a TPU in orbit from Google. >> I know, and it was a haiku. I thought it was pretty. >> Yeah, yeah, it was really nice. I think we've been hearing a little bit about, you know, Project Star Catcher from a while now, and I'd say it's probably one of Google's most sci-fi projects that they have going.

1:35 They have a lot going, but >> A lot, but they worked with Planet and they used SpaceX to >> Yes. >> partner, which was interesting. >> Yeah, big SpaceX partnership here. So for those who are a little less familiar, Project Star Catcher is Google's moonshot bet to bring AI data centers into orbit and run them on near constant sunlight. And so we've sort of heard this pitch before. We discussed it extensively on the show around the idea that data centers can now run in space from a climate standpoint and energy standpoint, grid standpoint, there's reasons why we might want to move some of these off terrestrial land.

2:10 >> Yeah, all the jokes on X are it's going to be the only place you can build a data center is space. >> I actually think from the midterms it was like I think there were like 41 pieces of legislation on data centers and all of them went to against and I think half of them were like total moratoriums of like total banning of data centers being built. So I think that's one part of it, but I also think there's actual reasons why you might want to from an economic and an energy standpoint put these things in space. So >> makes a lot of sense and we've covered it a lot in the show, but like maybe break down for us like what what are the reasons you want a data center in space and and you know this is just one small little leap. This isn't saying we can actually put one there.

2:50 >> Yes, yeah and we'll get into the economics of it all, but there are a couple reasons. For starters, the fact that you have near constant sunlight. >> Mhm. >> That's one of the issues right now is when you have solar panels on Earth, you are pretty much limited by weather, you're limited by the cycles of our sun. I think another part of it is the land issue, the grid, but there are also issues that come with it, right? Like how do you solve for thermal and radiation? Those two are still largely unsolved and I think Sun Capture is a big initiative to start testing some of that. But a big part of this was just figuring out how to get it up into space. So what happened on October 1st and why it was such a big breakthrough was they had a prototype on one of SpaceX's Transporter 18 ride shares and it was basically a Falcon 9 that was carrying about 100 plus different other spacecraft and different items on its payload. So in this case, Google didn't have its own rocket, rented a piece of it and what they were able to do is put a refrigerator size satellite that carried four Google TPUs and these are the same chips that Google runs on its ground data centers that are built for running inference and AI. They had about 3 weeks of commissioning and what that entailed is basically this checkout period where after the satellites reached orbit, they were making sure that, you know, one, that the solar panels were generating power, that the satellites were pointed correctly, that the radios could even reach them, and that the main computer was healthy.

4:18 And then afterwards, they actually powered up the TPU trays to run the small experiment jobs. and as you mentioned, the haiku, that was one of the tests was the engineers uplinked a prompt, basically, I believe saying, write a haiku about being in space. And then from there, a Gemma 4 model on board that was stored and running on the satellite itself ran it. And then once the chip generated the answer, it had to wait for the next ground station to pass, and then it downlinked it and then gave us that little haiku that we saw.

4:50 >> I mean, that was the best explanation of what happened. I feel like it was better than the video that Google released there. >> I feel like it's like it's honestly pretty simple. It was just basically a test run for what was possible, but it goes to show that all of this stuff is extremely incremental. Like we are not necessarily waking up to data centers in space overnight. >> It's It's still a headline, right? And it's something that Elon Musk says every single day on Twitter, but that doesn't mean that it's here tomorrow. There's still quite a few little barriers, and this was one of them.

5:21 >> Yes. And I I think the big thing here is that it was the first time you had Google's own TPUs, which are the ones that normally sit in, you know, these liquid-cooled racks in Google data centers actually sit in orbit and return an answer. So that's a pretty big must and we'd never had that before. >> enormous. So cool. >> And I think it tested a bunch of things, right? Like the fact that the chips were able to survive you know, the shaking of a launch, that they could power on and off in a vacuum, the software works up there, and you know, the full round trip works, right?

5:52 Like prompt up, compute in orbit, answer down. That's not, you know, we don't take all of that for granted. You know, that full work. >> is complicated. So Claire, what's like the big next step that needs to happen for data centers in space. >> One thing to note with this initiative is that compared to a lot of other companies that are putting AI chips in space, they're mostly focused on individual satellites with their own little computer.

6:18 So for example, you have these companies like Star Cloud, Cowboy Space, they each want their own satellite to handle their own small jobs. The difference here is Google's vision is for it to actually be many satellites that work together as one big computer. And so you have these big AI jobs like training a model, and that actually typically on the ground gets split across many chips that are networked together and connected through wires. Their vision is to fly all these satellites together in a constellation of sorts and then actually connecting them with lasers. So actually having satellite beams beam data across different neighbors as opposed to using interconnect and traditional data center networking.

6:58 >> Wow. Keeps getting more sci-fi. Like every time we talk about it, it just feels like anything, right? >> So the next big step with this is like, okay, now we know it works. You can, we sort of had this like hello world moment for AI in space. The next big initiative is actually being able to have two satellites to test whether they can talk to each other in space. So now that we know it works from ground to orbit, can you actually have two of them in a mini cluster of sorts communicate with each other?

7:27 As of last week, we've had the biggest crypto theft in history. >> Crazy, right? I think most people by now know Binance. It's one of the largest crypto companies out there and how a lot of people secure their crypto wallets they were broken into. >> this was the biggest theft in history at $6.3 or to $6.4 billion from Binance's hot wallets, which are basically the wallets that are online where customers can have their own withdrawals and where those are processed. And it happened across 10 blockchains. Most of it surprisingly went down in a matter of 20 minutes, which is a crazy timeline. As of right now, the FBI is blaming a group called Trade Trader, which is a North Korean state hacking unit, which is also actually the same group that did the previous largest hack. But to your point, compared to I think what's been rumbling around, you know, OpenAI Hugging Face or what happened in the Australian government, I actually think a lot of these cybersecurity incidents that we'll get into for the rest of this segment are all actually human driven.

8:31 >> Right? This this this feels like this was just a silly mistake. It seems like they took a fake job interview and the thing opened up and clearly it was malware, but then it worked its way through their system into a few others, which is how they were able to gain access. And that seems like they should have better protections against that at Binance, but at the same time, there's lots of companies where you could see this being a very easy thing to do. And AI tools actually just only make it easier to do this kind of stuff and start to like put these kinds of plans in action. And so I feel like this is just like the the tip of the iceberg of what we're going to see.

9:12 >> Yeah, for sure. What happened here was basically a Binance engineer was job hunting at the same time, and he had received a role that turned out to not exist. They had a fake CEO on a chat app. And then as part of it, that person was given a assignment to download as part of their hiring process. Completed it. That included malware on the work laptop. What is interesting though is you have these groups that are now armed with agent swarms that work 24/7 and they don't sleep. And so these groups are much more capable. And on top of that, it's not only the cyber part, it's actually also the sort of like human parts of the psychological parts of tricking someone, right? Like now deep fakes are getting that much better. AI voice is getting that much better.

10:00 >> many of these now. >> so I expect it's like yes, the actual like swarm itself is difficult to defend against, but there's also all the like human psychological parts of hacking that I feel are sort of exacerbated in this era. >> Yeah, and the deep fakes are getting so good. And we haven't even gotten to the fact that as AI continues to accelerate, all of our passwords and cryptography are like are on the line and there's a lot of money as we just learned from Binance sitting in wallets.

10:28 >> Yeah, and remembering that like as much as there is a lot of discussion around AI and hacking, like people are still often the weak point, but AI does make creating fake people and fake storylines and making the sort of like pressures around oh my gosh, I need to recover my password now or I I have to fill this form out right now, that much more believable depending on, you know, what modalities you're using. >> Grocery stores were hit for the second time.

10:55 >> Yes. >> and they're all from one manufacturer, which is kind of interesting. >> Yeah. >> What's going on? How are these connected? Is this yet another human story? >> Yeah, this is mostly happening in Europe. So completely separate. I would say Binance is like a story of an engineer within a firm. In this case, it's a little bit more state linked is what people are saying and you know, we we can only speculate today, but a lot of AI security and and research engineers are saying that this is likely state linked given the fact that there is no ransom and there's no sort of like outcome demand, right? I think in the case of Binance money withdrawal, here it seems like the goal is just sabotage. And so a couple weeks ago, the first wave of attacks happened, I believe, in mid October in France, mostly in stores like Intermarché, Carrefour, about 700 stores across France. And what happened was basically they had these units that were heated up to 50 Celsius, which is about 122 Fahrenheit, pretty hot. These were freezers and so obviously everything in there melted and their controllers were at their past end of life. So they no longer had updates.

12:06 >> like no longer warranted refrigerators like that people just hadn't gone and bought a new model, which you can understand. Like grocery stores aren't exactly keeping up with the latest technology every year. >> and many of them are still using their default passwords. But I I think the goal largely here was disruption. I do think that >> clear why though. Like I understand like yes, scary, disruptive. Like whether it's against the owner of the company that makes the refrigerators, is it against Europe overall because it was kind of everywhere?

12:36 >> Well, a lot of folks are saying specific sort of like nation states that are engaging in this like sort of shadow war. Like if if you think about it, attacking a food supply is technically a an act of >> terrorism for sure. >> and so that claim could be made. we of course don't know what is happening here. There's a lot of investigations. I think there was a similar event that happened in military commissaries in the US and the Pentagon as well. So I think the French government and the US military are also coordinating to figure out what's happening. But as of even a few days ago, there has been more of an escalation beyond food, which was France's blood service said that attackers actually tried to raise temperatures at three plasma cold storage sites. Plasma is typically used for transfusions and making medicines.

13:25 They had an alarm that caught it pretty quickly and the staff moved and secured all their products. So nothing was lost, but you can see how it like what happens when you raise temperatures at all these places that have frozen embryos and and people's biological tissue and medical facilities. And then what happens if you screw around through cyber attacks in water systems and the grid? And so like right now we're only here and you're sort of seeing some with the blood banks, but sort of scary to think and abstract like what happens at a larger scale.

13:53 >> abstract where we're heading with all of these AI enabled cyber attacks that we need to be even more vigilant. >> And that the internet is so much more connected than we think and the internet itself has physical consequences. And as we think about the importance of cyber, this is why it's so important to have strong cyber security and why even, you know, with our Cyber 60 event last week, it is even more consequential because it is binary code, but at the end of the day, it has very real consequences on our food, on our medicine, on our grid, on our energy.

14:27 >> Claire, is the internet dying? >> I don't know. I don't know. I've had some deep fakes, so maybe. >> Right? I also feel like X is now littered with AI slop posts. >> And LinkedIn. I It is so incredibly frustrating to go on any Yes. You know, I I I understand not breaking through, but it's like incredibly frustrating to see just you post something and you have thoughts you want to put out into the world and it's 200 AI generated bot comments.

14:53 >> I know. It's funny. I feel like it was getting slop-ish when we had AI where it wasn't agentic. But now I feel like so many people just have all these automations running of like I I see a trending like a story and I'm like, oh, this is like an old article recapped into some trending piece created by either Grokbot or by Muse or something has done this and it's so clear, but there's so much of it now.

15:23 >> It's tough. There's a lot of bots out there. I think also just seeing your point on blah blah blah is really good, but the quiet part is this. Honestly, at this point, if you just commented like love this or like something really quick and not very sophisticated, I'd almost appreciate that more than like a block of random AI generated characters. >> commenters. I know. Engagement farming or whatever it is that they're doing. >> But I think LinkedIn actually has a button where you can report something as AI.

15:51 >> it says you can report it as AI slop, I think, right? Or something like that. >> Yeah, and I think there was finally a couple days ago, Andrej Karpathy had published a website called Hacker 2, I believe, which is basically his initiative to remove the slop and sort of the clickbait and hype of short form media on his YouTube algorithm. And so he actually built this himself with Claude code and created a YouTube clone with only videos that Hacker News had recommended and only high-quality things. And so, I think one, people are craving cleaner versions that are sort of a little less noisy but a little less AI of software.

16:31 And that two, personal software now enables that, which is really interesting. It's it's definitely, I think, one of the frustrations I've been feeling navigating the internet. But beyond that, I think on the other side of it is not as a consumer, but as a publisher, I think there's also some growing pains of this potential dead internet theory we are experiencing. >> Because it makes it even more difficult to get good written or video content seen, heard, listened to, all the things. But speaking of that, we also had Reddit this past week announced earnings. They're starting to see a little bit of a decrease in traffic due to all the Google AI summaries.

17:10 >> Mm. >> And so basically, if you go to Google something and that first result would have been a Reddit post, it is instead a summary of that. And so it's You know, you can kind of see like, why am I going to click through to Reddit? Like, I don't owe Reddit a click. >> Mm-hmm. >> And I think we're starting to see the early innings of maybe people not clicking through. >> There were some pretty stark numbers here. US daily uniques drop by 2% year over year on 50 million plus folks. And their search referrals fell by more than a million visits a week on average. And I actually find this even with myself, right? Like, I used to >> Yeah.

17:51 >> through a ton of Reddit posts and maybe for the sake of efficiency, now I use AI overviews, which has saved me a lot of time. But sometimes I'm a little bit nostalgic to the time when I like got sucked into a Reddit rabbit hole and could see the discourse. Now I feel like the main way I'm interacting with that information is in a overview summary. >> I know. I also feel like Google was a big referral for that. And I wonder how that will change cuz there was this whole era of companies that were sort of built around when Facebook was like popping, that all these publishers that were basically built on getting the click out of Facebook. And then Facebook killed that whole mechanism and many of those companies from like little things to Lad Bible, like all though that like era of of >> BuzzFeed >> BuzzFeed journalism >> Yeah.

18:42 >> kind of just evaporated overnight. And so it'll be interesting to see. The great thing with Reddit is they already have like an incredible user base. And I feel like so much of AI answers, YouTube, like all of these things are all heavily fueled by expertise if you will or whatever you want to call it that is >> opinion. >> that is available on Reddit and for some reason nowhere else. And so I think they have a very clear advantage in the garden and the walled garden that they've built.

19:17 But will they be able to bring in new people because Google was such a high referrer? >> What do you think they'll do in reaction to this? >> Oh gosh. Do you owe Reddit a click? I I don't owe Reddit a click. I I just wanted the information. I know Reddit is often the best place to go for it. But like if I have to wait for my browser to switch over to the Reddit app and then load the post to then read the post, I think I just prefer the AI summary.

19:46 And so it's like I'm just thinking about it in terms of pure consumer behavior, which is like the consumer just wants whatever's the least friction. >> Yeah. >> And so if they are an avid redditor, they're kind of already there. >> Yeah. It's interesting cuz I always think of like the idea of these like places for centralized commentary and I think of similar things of like comments on YouTube videos or like reviews on Amazon products and what happens to all those.

20:13 And I think for me at least for Amazon products and like utility oriented things, I find myself relying more on AI reviews or just using like whatever AI version of bot being like is this actually you know is this >> Rufus. >> I I find myself asking Rufus. >> Really? >> Yeah yeah cuz I'm like I don't want to scroll through 20 comments. Instead I want to just be like is this vacuum good at carpets or is this vacuum good at bathrooms? At least for me I find myself switching behaviors more towards these AI reviews for utility oriented commentary.

20:45 >> Makes sense. >> But I think still for entertainment and like being part of the zeitgeist >> Yeah or understanding how to be in on the joke. All that stuff is is still kind of go to go to the platform for that. >> Like when I wanted to hear more discourse around that. >> So maybe this isn't all their traffic, right? Cuz it's it's not just shopping decisions. >> Yes, exactly. I think shopping decisions is probably in dangerous territory, but like cultural commentary I feel like hopefully is safe and we'll see.

21:10 >> you kind of want to click through to the Reddit thread and read all the replies to it. It's not just about that. >> Yeah. >> And so I I I do think that that that gives them a little bit of an edge, but I don't know. That's that's a lot of traffic. Like if you Google like best robot vacuum and it shows you a Reddit thread that gives you the answer you're not going to click through. So it's it's a it's this is there's a reason the stock fell. Like this is a material piece of their business that is in for the first time jeopardy.

21:40 >> Yeah. As someone who spent a lot of time in the marketing world like how are some of your peers and and folks in the industry thinking about all of this? >> I mean the thing is is I actually think that marketers should spend more time advertising working with Reddit because if you think about AEO and the value that Reddit has on all of these platforms >> Yeah. >> it actually is like one of the most valuable things you can spend your time on is like how do you make sure that your product or whatever it is that you're selling >> Mhm.

22:16 >> has the kind of like word of mouth on Reddit. And so I am pretty bullish on Reddit as a marketer. I think the hard part is just like not thinking that it's now like the destination, that it's like this place that is maybe now becoming a little bit more of a CRM for these other systems, for agents, for for the models, and that a lot of discourse will happen there, but the utility of it is different.

22:44 >> Yeah. At least for me, it feels like there is a lot of value in like where people want to gather on the internet. Like I actually think I'm more bullish on >> I also don't see a lot of slop on Reddit, I have to say. I mean, I see some, but not as much as I see on X and and and but and I'm not sure that they're doing anything to prevent it. I don't think they are.

23:04 >> Mm. >> Slop's getting harder and harder to detect. >> Yeah. I don't know. I I mean, I do think there is this like sort of like bifurcation of the internet. Like I think the Times actually has a separate version just for AI bots to navigate. And I think most websites will start having that. >> I think so, too. >> The question is like for Reddit, for their consumers versus like being fed to Google Overviews, those are different models. And how they figure out like where the value sits.

23:31 I also think for me, it just goes to show like how much more value there is in just like the curation layer. And that's always sort of a bet I've made in my own content creation is information is so cheap now, right? Like everyone can output content. And so now it's a matter of like in this world where the value is no longer in like who can give you information, it's who can actually curate it. And so I think at least for me, that feels like what we're trying to do here with this podcast or even in my own channels, it's like that's the initiative is the value gets abstracted to a higher level of who can actually help you navigate this information and all this noise. But we'll see. I'm very curious about the future of Reddit because I do love the platform.

24:07 And things do need to change. And And maybe it is that sort of like bifurcation of like what is the value prop for the actual like agent navigating the web versus like a human consumer like you and me. >> Yeah. There's this whole layer that's still to be built. >> Yes. >> But feels like we're on the road to it. >>

Summary

Google's Project Star Catcher successfully received its first AI response from a TPU in orbit, marking a significant milestone in space-based data centers. Meanwhile, Binance experienced the largest crypto theft in history, with hackers stealing $6.3 billion, and grocery stores in Europe faced cyberattacks that compromised freezer temperatures, raising concerns about food supply security.

- Google's Project Star Catcher aims to establish AI data centers in orbit, leveraging constant sunlight for energy.
- The first AI response from a satellite was a haiku generated by a TPU, demonstrating the feasibility of AI operations in space.
- Binance suffered a $6.3 billion theft attributed to a North Korean hacking group, highlighting vulnerabilities in cybersecurity.
- A Binance engineer fell victim to a fake job interview that led to malware installation, emphasizing human error in security breaches.
- Cyberattacks on grocery stores in Europe involved manipulating freezer temperatures, suggesting a potential state-sponsored effort to disrupt food supply.
- Reddit's traffic is declining as Google AI summaries replace direct links to its content, raising concerns about its future engagement and revenue.
- The rise of AI-generated content is leading to a "dead internet" theory, with users frustrated by the prevalence of low-quality, automated posts.
- The need for stronger cybersecurity measures is underscored by the interconnectedness of digital systems and their real-world implications.

Questions Answered

What is Google's Project Star Catcher and its significance?

Google's Project Star Catcher aims to establish AI data centers in orbit, marking a significant step in the integration of AI technology with space exploration. The project recently achieved a milestone by receiving its first AI response from a TPU in orbit, showcasing the potential for advanced computing in space.

What challenges remain for data centers in space?

While the first AI response from a TPU in orbit is a significant achievement, it highlights that the development of space-based data centers is incremental and faces various challenges. These include ensuring the durability of technology during launches and the complexities of operating in a vacuum.

How is AI impacting cybersecurity and personal security?

AI advancements, particularly in deep fakes and voice synthesis, are exacerbating cybersecurity threats. As AI becomes more sophisticated, it increases the potential for psychological manipulation in hacking, making individuals more vulnerable to scams and phishing attempts.

What are the implications of AI-generated comments on social media?

The proliferation of AI-generated comments on social media platforms is leading to frustration among users. Many feel overwhelmed by the volume of low-quality, automated responses that dilute genuine engagement and discussion.

How is AI influencing consumer decision-making?

Consumers are increasingly relying on AI-generated reviews and summaries to make purchasing decisions, particularly for utility-oriented products. This shift reflects a desire for efficiency and reduced friction in the shopping experience.

© transcribe · For agents Built with care and craft by Gokul Rajaram