BrokeIt - Daily AI News · All episodes: ↗

US water hacked, Gemini agent speed, AI learns your channel

2026-08-15 · 7 min

Listen · Apple Podcasts Listen · Spotify

Stories covered

Transcript

Intro

Ivy: An AI that can process an entire codebase in under a minute. The intern you never have to pay.

Marcus: I'm Marcus.

Ivy: And I'm Ivy. It's Saturday, August 15th, and this is The AI Daily.

Marcus: And we have a huge show lined up for you.

Marcus: On the docket today: Google's new Gemini Flash model is so fast, it could finally make true AI agents a reality.

Ivy: Then, we're looking at how Twitch is using your channel to train Amazon's AI—whether you know it or not.

Marcus: And finally, we'll dig into alleged Iranian cyberattacks on U.S. water utilities. Spoiler: it's not good.

Gemini 3.7 Flash: Fast Enough for Real Agent Workflows

Marcus: Alright, let's talk speed. Google just dropped Gemini 3.7 Flash, and the name says it all.

Ivy: Right. Another model, another version number. We get these every other week. What's different here?

Marcus: It's all about 'agentic workflows.' We're talking high-volume, multi-step tasks. This model is fast and cheap enough for an AI to plan, act, check its work, and repeat—all in real time.

Ivy: So it's not about being the smartest model, it's about being the most industrious? The blue-collar model?

Marcus: You got it! It's optimized for speed and cost. For a developer, it's like a junior partner who can summarize code reviews, run tests, and find dependencies, without you ever feeling like you're waiting.

Ivy: But what's the trade-off? 'Flash' sounds like you're losing something. Is it accuracy? Nuance? Is it going to confidently write buggy code faster than ever before?

Marcus: Google claims it keeps high quality for retrieval and instruction following. It's just hyper-focused. And the cost makes running millions of queries a day actually viable.

Ivy: Which, for a company like Google, translates to a lot of money saved on their own internal infrastructure.

Marcus: Totally. But for everyone else, too. The AI tools you use every day are about to get way less laggy and start doing actual work, instead of just chatting.

Ivy: Great. So my AI assistant will now be able to book the wrong flight for me with breathtaking efficiency.

Marcus: You have to see the potential, Ivy! This is the engine for the agentic future everyone's been promising.

Ivy: We'll see. The 'agentic future' also sounds like a great way to automate mistakes at scale. I remain unconvinced.

The Ghost in the Stream: When Your Channel Becomes AI's Classroom

Ivy: Alright, next up: let's talk about where AI models get their data. Because right now, it might be from you, while you're streaming on Twitch.

Marcus: Right, the Twitch mess. Amazon, their parent company, is using creator content to train its generative AI models.

Ivy: And to put it mildly, they weren't very open about it. By default, every single creator is opted in.

Marcus: Wait—so you have to manually go into your settings and tell them not to use your content?

Ivy: You do. It's the classic opt-out versus opt-in fight. And for your voice, your face, your actual work... consent needs to be explicit. You should have to check a box saying 'yes,' not hunt for a box to uncheck.

Marcus: Okay, that's... not a good look. What are they even training? A model that can tell you the best way to defeat a boss in Elden Ring?

Ivy: That's the thing—they won't say. 'Amazon's generative AI' could be anything from Alexa to AWS to content moderation tools. Creators have no clue what they're helping to build.

Marcus: This feels like a huge breach of trust with the creator community. Their entire business is built on these people.

Ivy: Exactly. So if you're a streamer, listen up: you need to go to your Twitch dashboard, find your 'Content' settings, and look for the 'Use for AI Training' toggle. You can decide for yourself, but the point is, now you know to look.

Marcus: Wow. The digital ghost in your own stream... being used as a classroom.

Ivy: It raises the fundamental question: when you upload content to a platform, do you still own it? Or are you just renting space in their AI's brain?

What we know about the alleged Iranian hacks on US water utilities

Marcus: Alright, our last story today is a serious one: hackers are targeting U.S. water utilities.

Ivy: And they're not just targeting them—they're getting in. The FBI and CISA are pointing the finger at hackers backed by Iran's Islamic Revolutionary Guard Corps.

Marcus: How are they getting in? This is critical infrastructure. You'd expect Fort Knox.

Ivy: You'd think so, but you'd be wrong. In at least one case in Pennsylvania, they got in through a device still using its default, factory-set password. We're probably talking '1-2-3-4'.

Marcus: No. Are you serious? For a water utility? That's not a hack, that's walking through an open door.

Ivy: They're targeting things called PLCs—programmable logic controllers. Basically, the industrial computers that manage things like water pressure. A lot of them are decades old and were never meant to be connected to the internet.

Marcus: So we have ancient, insecure hardware controlling our water supply, and it's exposed online. That's... terrifying.

Ivy: So far the disruptions have been minor—the hackers seem more interested in sending a message. But the potential for real-world harm is enormous. They could alter chemical levels, shut off water flow... it's a public safety nightmare.

Marcus: And this is a two-way street for AI, right? You can use AI to scan for these weak spots, but the attackers can use it to find them even faster.

Ivy: Exactly. And this isn't about your laptop's security. It's a reminder that our physical world runs on digital systems, and a lot of them are shockingly fragile.

Marcus: A wake-up call for thousands of small, under-funded utilities across the country.

Marcus: Alright, let's recap the headlines.

Ivy: Google's new Gemini Flash model promises to make AI agents faster and cheaper, but the jury's out on whether they'll be better.

Marcus: Twitch is using streamer content by default to train its AI, forcing creators to manually opt-out.

Ivy: And Iranian-backed hackers are reportedly exploiting embarrassing security flaws at U.S. water utilities, putting critical infrastructure at risk.

Marcus: Alright, before we go, I saw a great one. Someone asked an AI to generate 'the most boring picture possible.'

Ivy: And? A photo of drying paint? A lecture on tax law?

Marcus: It came back with a single, slightly out-of-focus, beige pixel. That's it. The AI's definition of boring was... almost nothing.

Ivy: I think I've attended corporate meetings that were less interesting than a beige pixel.

Marcus: And that's our show for Saturday, August 15th. We'll be back Monday with more of The AI Daily.

Ivy: Until then, try not to be a beige pixel. I'm Ivy.

Marcus: And I'm Marcus. Talk to you then.

This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.