Google’s Reasonable AI Pause - DTNSB 5033  ## Metadata - Author: [[Daily Tech News Show]] - Full Title: Google’s Reasonable AI Pause - DTNSB 5033 - Category: #podcasts - URL: https://share.snipd.com/episode/a0c2bbc3-a034-47ba-a570-15f78a18112f ## Document Notes ## Highlights - **Google Pauses AI Photo Search** - Google paused its Ask Photos AI feature rollout due to latency, quality, and UX issues. - The pause allows improvement while users who already have it keep access, showing responsible AI rollout practices. Transcript: Tom Merritt Google said it has paused its rollout of the Ask Photos feature of Google Photos. Product manager Jamie Aspinall wrote on X that it is, quote, not where it needs to be in terms of latency, quality and UX. For those not in the know, that's user interface. He said that Google will ship an improved version in about two weeks. So it's not going to be paused for long. And I guess if it's rolled out to you, you get to keep it. That brings back the speed and recall of the original search is what he hopes to do in that two week pause. He was actually responding to complaints from a Roblox engineer, Yousef Amari and David Lieb. David Lieb co-founded BumpTech, which became Google Photos. So he's essentially the co-founder of Google Photos. He no longer works at Google now, but these are not people unknown to Aspinall. Ask Photos uses Google's Gemini models to interpret natural language search queries. So the idea is that you could say something like, which ramen restaurants have I been to? And instead of just giving you all the pictures of ramen, it would only give you pictures of ramen at ramen restaurants, or even pictures of the outside of the ramen restaurant, because It would know what that was. It would not give you pictures of instant ramen you made at home. It was described at Google IO in 2024 this way. Gemini's multimodal capabilities can help understand exactly what's happening in each photo and can even read text in the image if required. Ask photos then crafts a helpful response and picks which photos and videos to return. Now, the keyword part of this is apparently working. It can actually read the text in the photos if you put it in quotes as a keyword search. It's the more natural language stuff that is laggy and doesn't always work as you want. It was launched in early access in September. So not everybody has it. You might have it since September. You might not. But apparently people have complained that it's slow and not always accurate. It's yet another example of how these tools are good, but maybe not always shipping product good. And Jen, the other thing that caught my eye about this story was the friendly exchange of the complaints that led to a reasonable admission and action. Jenn Cutter Yeah, like they were valid complaints. When it's not working, it's not working. They had asked what a lot of people like me ask was, how do I turn this off? And yeah, and immediately got a response. Now, of course, it does help when you all know each other. You're all in the same circle and you all kind of know what you're talking about when it comes to tech products. So it didn't start as a very high animosity request and was dealt with as a non-high animosity request. Tom Merritt And I think it was Lieb who said in a follow-up, he's like, it's always good when users complain. It means they care. What you don't want is silence, which is, of course, easy for him to say because he's not the product manager. But I get what he's saying. And it reminded me of the times that I've seen a reply on social networks and started to get my back up a little, you know, at it and then saw who it was from and was like, oh, well, they mean Well and totally changes your opinion of a criticism. And I think that's definitely what happened here. I'm not even sure where to go with that analysis, but I was like, I bet you get a different reaction from Aspinall if it's people he doesn't know complaining in the same way. Jenn Cutter Yes. And if you didn't already know him, you're not going to get a text about when it's fixed. Tom Merritt You're definitely, you're probably not in his contacts. No, you're not going to get his mass text. So yeah, I think a lot of people are jumping on this story because it is very buzzy. I'm seeing everybody writing about it today because it is an admission by Google that its AI tool is not up to stuff and they have to pause it and make it good. But I think it's the right thing to do to say like, yeah, this isn't working right. Give us two weeks. We're going to pause the rollout. We're not going to take it away from anyone, but we're going to try to make it better before we start rolling it out, which is why you do slow rollouts of stuff in the first place, right? ([Time 0:02:47](https://share.snipd.com/snip/8b8ef0ab-d710-4487-bc79-3bb34152feae)) - **Friendly Feedback Spurs Fixes** - Complaints about Google's Ask Photos came from industry insiders familiar with the product. - The polite, constructive feedback led to a quick and respectful response from Google, showing the value of friendly tech critiques. Transcript: Tom Merritt And action. Jenn Cutter Yeah, like they were valid complaints. When it's not working, it's not working. They had asked what a lot of people like me ask was, how do I turn this off? And yeah, and immediately got a response. Now, of course, it does help when you all know each other. You're all in the same circle and you all kind of know what you're talking about when it comes to tech products. So it didn't start as a very high animosity request and was dealt with as a non-high animosity request. Tom Merritt And I think it was Lieb who said in a follow-up, he's like, it's always good when users complain. It means they care. What you don't want is silence, which is, of course, easy for him to say because he's not the product manager. But I get what he's saying. And it reminded me of the times that I've seen a reply on social networks and started to get my back up a little, you know, at it and then saw who it was from and was like, oh, well, they mean Well and totally changes your opinion of a criticism. And I think that's definitely what happened here. I'm not even sure where to go with that analysis, but I was like, I bet you get a different reaction from Aspinall if it's people he doesn't know complaining in the same way. Jenn Cutter Yes. And if you didn't already know him, you're not going to get a text about when it's fixed. Tom Merritt You're definitely, you're probably not in his contacts. No, you're not going to get his mass text. So yeah, I think a lot of people are jumping on this story because it is very buzzy. ([Time 0:05:00](https://share.snipd.com/snip/d87a50bf-c2db-4866-8898-ba5320a4a7b9)) - **AI Photo Search Capabilities** - Google's Gemini AI for Ask Photos can detect text in photos and identify context like actual ramen restaurants. - However, it struggles with natural language queries and is slower than expected. Transcript: Jenn Cutter That was not a helmet. But I also have a lot of web searching from the phone stuff disabled. So it's not getting any assistance from any, you know, IQ stuff that's not on the phone. Tom Merritt Yeah, but at least with Apple, anyway, you're supposed to be able to do that on the phone at some point. And so yeah, it would be nice if that came sooner than later, but it doesn't seem like it is there. Jenn Cutter In fairness, taking out like the liner of the helmet probably made it harder to recognize as a helmet. Tom Merritt Yeah, but these things are supposed to be smart. They should be able to get there. ([Time 0:07:09](https://share.snipd.com/snip/1f6e0fb9-8731-47d9-a6ec-422d268b4b3f)) - **Protect Against Salesforce Phishing** - Beware of phishing scams targeting Salesforce users via phone calls posing as IT support. - Always verify requests for credentials to protect company data from social engineering attacks. Transcript: Tom Merritt Google's Threat Intelligence Group also warned of a phishing scam against Salesforce users. So if your company uses Salesforce, pay attention. Attackers have been calling employees, pretending to be IT support, and then convincing them, social engineering, to hand over credentials that they then use to steal data from the Company's Salesforce installation. So beware of those kinds of calls. Finally, Microsoft is offering the 27 member states of the European Union free access to a new European security program. Among the benefits are increased AI-based threat intelligence sharing, strengthening security and resilience of government systems, and expanding efforts to disrupt and dismantle Bad actors. Jenn Cutter Social engineering, the danger that never goes away, never goes out of style. Tom Merritt That's one where awareness is a big help in preventing it from happening. ([Time 0:15:52](https://share.snipd.com/snip/59126800-e3e9-4ecc-b4fc-7fcae4a1cec2)) - **Privacy Violations by Big Tech** - Meta and Yandex used Android app loopholes to gather web cookie data, breaching user privacy. - Meta removed the code silently without a good justification, revealing unethical data practices. Transcript: Tom Merritt Easily convinced that it's real than if you're like, wait, you're trying to get my Salesforce credentials? I heard about a social engineering thing. How do I know this is real? Jenn Cutter Computer scientists affiliated with Spain's Mdia Networks, Radboud University in the Netherlands, and Belgium's KU Leuven discovered that Meta and Yandex were using their Android Apps to monitor local host ports to gather web cookie data and help link users to browsing data. Local hosts lets the device make requests to itself. By using this port, apps were able to evade privacy protections, which are focused on outbound network traffic. After the disclosure, Meta stopped sending data to local hosts and largely removed the code. Tom Merritt I didn't see anything about Yandex stopping this, but that's a Russian company now, so who knows what they're up to. I imagine this has got to violate some kind of Google provision. And Meta said they are. The only thing they've said about it is we are talking to Google about a misunderstanding about the blah, blah, blah, blah. They have not admitted to this, but the code is gone. And I am one, if you've followed me throughout the years, who constantly is telling people like, hold on, there might be a good reason of this. I've been in a company when they do something and everybody thinks it's malicious, but it's not. This is malicious. Okay. Okay. I'm here to tell you, like, there is not, ([Time 0:16:48](https://share.snipd.com/snip/2d75a7ac-8da2-4cb1-b050-db7e310556e8)) - **AI as Creative Partner** - ABBA's Bjorn Ulvaeus uses AI as a creative partner rather than a replacement. - AI sparks creativity by providing fresh ideas, though it's poor at writing full songs or lyrics alone. Transcript: Tom Merritt He did say he's about three quarters of the way through. And here's a quote. It is like having another songwriter in the room with a huge reference frame. It is really an extension of your mind. You have access to things that you didn't think of before. Now, he also said it's lousy at writing a whole song and very bad at lyrics. So he's not universally praising it. He says he used it to prompt him to think in different directions rather than to write things whole cloth. So it didn't replace him as a writer. He's using it as a way to inspire his creativity and mold his creativity. And composers, and he's involved in the negotiations to ensure that music generators pay royalties to the original creators of music who are used to train the systems. So he's not giving companies a pass, but he's also not boycotting the tools. He's like, the tools are great. They're not great at everything. They're good at some things, and they should pay us for the training that they use to make the tools. But that doesn't mean the tools are bad. Anyway, I thought that was really interesting. Also, in related Hollywood meets AI news, AMC Networks, which has BBC America and AMC TV, has reached a deal with Runway to use Runway's generative tools to generate marketing images And help pre-visualize shows that have yet to be produced. So the idea there is if you're a show runner, you can try out some ideas without having to shoot material. So it gives you a chance to do things you wouldn't have the budget to do otherwise. And of course, create a wider range of marketing materials and promos around stuff. Jenn Cutter So I get the pre-vis work. Pre-vis is very expensive. And this also lets people, even in, like the early pitching stage to, to help nail things down and implement feedback almost instantly. The marketing though, we've seen some very not great stuff from Amazon Prime for shows that exist that have marketing materials and making terrible, terrible thumbnails. Tom Merritt Yeah, you're going to have to be careful how you use this, but it's still it's like it's it's basically what Bjorn is saying is like, I it would be wrong for me to have it try to write the whole Song. But if I use it right, it actually speeds up my creativity. So what you saw with those thumbnails was like, yeah, that's the bad way to do it. But it can still speed up generation and allow you to create wider marketing materials if you don't let it get out of hand, you know, and you look at the thumbnail and go, yes, that one's Good. No, that one's not. Jenn Cutter Yeah, I liked the ABBA approach to, yeah, I'm using it as a sounding board because he's probably like other creative people who don't necessarily work at the other hours that their co-writers Want to be awake for. So you have something to bounce off of, but you're still the one doing the work. And to me, that is important. Tom Merritt I love that he seemed very enthusiastic about it while also not compromising and saying, oh yeah, no, they need to pay us, but it's a good tool. I thought that was a really interesting take. You don't see that often. ([Time 0:22:14](https://share.snipd.com/snip/fa5fb544-fb8c-4f30-a57b-8d711065d578)) - **Ethical AI in Filmmaking** - Natasha Lyonne's film uses an ethically sourced AI model for augmented reality scenes. - The AI model's traceability and ethical licensing set new standards for responsible generative AI in film. Transcript: Jenn Cutter Lyonne has ventured into AI filmmaking with Uncanny Valley, a sci-fi project blending human creativity and generative AI to explore virtual reality's impact on identity. Andy Beach tells us more. Tom Merritt Andy, thanks again for joining us, especially because you brought a story with one of my favorite actors, probably people know most from Orange is the New Black or maybe Poker Face on Peacock. But tell us what Natasha Lyonne is up to here. Andy Beach Yeah, I absolutely love her too. Natasha Lyonne has recently announced that she's working with, I want to get the name right, Britt Marling, and collaborating with technologist Jaron Lanier to create a new independent Film. It's called Uncanny Valley, and it really tells the story of a teenage girl whose world and life is shaped by this augmented reality game. So within the film, while the film itself is not an AR or VR piece, you're going to be going in and out of that world and that environment a lot. And for those scenes, they're working very specifically with an ethically sourced generative AI model called Mary from a company, Moon Valley. And so this got a lot of buzz this week, I think in part because Poker Face is back on the air, but also because this really hits into a lot of areas we're talking about. It's the use of generative AI in a mainstream film, even though this is an independent production. But it brings up that fact that everything that went into creating Moon Valley's Model Mary was licensed and has traceability and is ethically sourced. And this is the sort of the convergence of a lot of conversations of what we talk about in this current age of AI. Tom Merritt Yeah, it just strikes me as funny that what was true once for coffee beans is now being applied to generative models. But yeah, so the Mary model, they have the receipts, I guess, where they can say like, this is everything that went in the training data. We're transparent about the training data. Andy Beach Yeah, absolutely. And I think importantly, you know, Adobe had a little bit of a gaffe early on where they were touting Firefly as an ethically sourced AI model and then had to acknowledge that, you know, The mid-journey content that they had licensed and used didn't necessarily have provenance to that. So I think people are being even more explicit now about not only did they license the content, but what is the provenance and the background of that content that they've traced and can They provide that it itself was ethically created as part of the process? Tom Merritt Is it as good though? Like that's the other side of it is like, great, you're using a different model, but not all models are as good at generating this sort of thing. Like, can you tell? Yeah. Andy Beach You know, it's interesting. And in this case, I'm not sure that it, that it matters because it is, uh, and, and I'll, I'll be honest, I haven't gotten a chance to play with the Mary model yet. So I don't know the answer of what the quality is. But given that it's playing a role of play of being the sort of the augmented reality pieces of a game that even is supposed to be from sort of the little bit in the past, I think they get a Little bit of a pass on what that quality is. So it's probably good enough for the scenario they're using it in. My personal, you know, I suspect personally that even as good as it is now, it's likely to get much better in the next sort of six months to 12 months, because that is the speed and the rotation That we sort of see on quality changes coming to models as they get updated. Tom Merritt Yeah, it's a clever use of generative models because you made it part of the story that is, it's almost like diegetic sound, right? Like, right, like this, this is not a model pretending to be something else. This is model playing itself. It is a character in the story. Andy Beach It's kind of a perfect first use case of it in in a big production because it gives it the guardrails of keeping it in that environment where we're not trying to say that the entire scene Or the entire thing is shot and is supposed to look like real life. It's supposed to look like a game. Yeah. ([Time 0:25:34](https://share.snipd.com/snip/12fa4d6f-589a-483c-99ea-500a3abdcd88)) - **AI and Sustainability Challenges** - Major cloud providers aim for carbon neutrality by 2030 despite AI's energy demands. - Sustainability efforts in AI infrastructure will be critical as model usage scales. Transcript: Tom Merritt And that makes it easier. On the ethical side, though, I mean, I know we talked a lot about the sources of the data, but what about the power usage? Because that's another thing that people object to a lot of times with these models is how power hungry they are. Yeah, absolutely. Andy Beach And, you know, I think that ultimately becomes the issue primarily for the cloud that is storing it. And this is a thing that both all Google, Oracle, Microsoft, and AWS have all made commitments to being carbon neutral by certain dates. Generally, they're all at a sort of 2030 timeline. Microsoft had actually been ahead of that and had been carbon neutral for quite a while until AI took off and proliferated and it sort of set them back a few more years. I know from my time at Microsoft that they also have publicly announced future dates where they plan to be carbon positive as well, actually putting back to the community. So I think it will have to be the platforms and the data centers that really focus on pushing for that. But we do need people talking about this and we need groups pushing for the sustainability and the green side of AI because it's going to be critical to help get it to, to the scale that, That we want, uh, at the, uh, sort of at the environmental costs that, that we are comfortable with. ([Time 0:29:54](https://share.snipd.com/snip/87b99e2b-fd43-42d2-be61-867adeef5efd))