Guru's Tech Bytes

Guru's Tech Bytes

Download on the App Store

Guru's Tech Bytes episodes

  • AI has access to a vastly larger working memory than the human brain | EP #133
    Good morning, it's Sunday. This is Guru's Tech Bytes, episode one hundred thirty three. And boy, do I have a pile of stuff for you today. Heh. Hhh, okay, that's something. We're talking about machines that remember everything, a drug that might save your brain, a bike that charges your phone, and a chip that should've seen the writing on the wall. So grab your coffee, settle in, and let's chew the fat for a minute or two. I promise it's worth your time.
    First up, and this one blew my mind, AI now has a working memory that's way bigger than a human brain. We're talking like, a hundred times more tokens, or whatever they call it. It can hold an entire book in its head while it's talking to you.
    Second, and this is a weird one, that diabetes drug, semaglutide, you know, the weight loss one, it's now linked to a lower risk of dementia. They did a big study, and folks taking it had a way lower chance of getting that awful disease. Now, I'm not a doctor, so don't go chugging it, but it's a heck of a side effect. It's like finding out your work boots also fix your back. The brain and the gut, they're more connected than we thought, I guess. Pretty wild stuff.
    Third, we got the Asus Bike Booster. This is a gadget for your bicycle that uses the spinning of your wheel to generate power. So you're pedaling, and it's charging your phone or your lights. It's not a full battery, just a little boost, but for a guy like me who gets lost on trails, that's a lifesaver. No more dead phone when you need the map. It's simple, it's mechanical, and it just makes sense. No fancy AI, just good old-fashioned physics doing a job. I love that.
    And finally, we gotta talk about RISC-V. This is a computer chip design that's open source, and the big guys, they should have known better. They ignored it for years, and now it's everywhere, from your router to your car. It's like they had a free blueprint for a better engine and said, nah, we'll stick with the old one. Now they're scrambling.
    That's your daily byte. Have a great day. Until next time.
    Stories covered
    • AI has access to a vastly larger working memory than the human brain — Hacker News
    • Semaglutide linked to lower predicted dementia risk — Hacker News
    • Asus Bike Booster — Hacker News
    • RISC-V: They Should Have Known Better — Hacker News
    • 2 min
    • Qwen 3.8 27B | EP #132
      Good morning, it's Saturday. This is Guru's Tech Bytes, episode one hundred thirty two. And boy, do we got a full plate today, folks. I mean, we got everything from giant brains to browser wars, and even some spy stuff that’ll make you wanna check your locks. So grab your coffee, maybe a donut, and let’s just dive right in, because there’s a lot to chew on and I ain’t got all day, even if it is Saturday. Heh. Hhh, okay, that's something.
      First up, we got this new AI model called Qwen three point eight, and it’s got twenty-seven billion parameters. Now, I ain’t gonna pretend I know exactly what all that means, but the folks online are losing their minds over it, over eleven hundred upvotes. It’s supposedly this open-source thing that can run on your own computer, not some big cloud server. That’s huge for us regular folks, you know? It’s like having a smart intern in your basement, but one that don’t eat your sandwiches.
      Second, and this one hits close to home, Firefox is now the last big browser that still supports uBlock Origin. You know, the ad blocker that actually works. Chrome and Edge already kicked it to the curb, and now it’s just Mozilla holding the line.
      Third, Google is doing some wild stuff with private AI using something called homomorphic encryption. Basically, it means the computer can work with your data without ever actually seeing it. It’s like giving a mechanic a sealed box and telling him to fix the engine inside, without ever opening the hood. It’s still early days, only three hundred eighty-nine upvotes, but if they pull it off, that’s the holy grail for privacy. You get all the smart features without handing over your whole life story.
      And finally, we got this story about “Going Dark,” and it’s about the cops getting into your phone. Law enforcement is getting real good at hacking into devices, and it’s a whole new era of digital lockpicking. It’s a scary thought, but it’s also a real conversation we gotta have about where the line is between safety and privacy. It’s not just about the bad guys anymore, it’s about all of us.
      That's your daily byte. Have a great day. Until next time.
      Stories covered
      • Qwen 3.8 27B — Hacker News
      • Firefox is now the last major browser that still supports uBlock Origin — Hacker News
      • Google is making private AI practical with homomorphic encryption — Hacker News
      • Going Dark, and the era of law enforcement hacking — Hacker News
      • 2 min
      • Gemini 3.7 Flash | EP #131
        Good morning, it's Friday. This is Guru's Tech Bytes, episode one hundred thirty one. And boy, do we got a packed show for ya today. I mean, we’re talkin’ four stories that’ll make your head spin, and I ain’t even had my second coffee yet. Heh. Hhh, okay, that's something. So let’s just jump right in, because there’s a whole lotta buzz out there, and I gotta tell ya, the future’s lookin’ mighty weird and wonderful.
        First up, we got Gemini three point seven Flash, and this thing’s got eight hundred ninety three upvotes on the boards. That’s a big number, folks. People are goin’ nuts over how fast it spits out answers, like, blink and it’s done. I heard it’s leaner, meaner, and cheaper to run, which is music to my ears. You know, for a guy who just wants his phone to not lag when I’m lookin’ up a torque wrench.
        Second, DeepSeek’s got a Harness developer preview out, and that’s got six hundred eighty eight upvotes. Now, I ain’t no coder, but from what I gather, this thing lets you build AI agents that actually do stuff, not just talk. Like, you give it a task, and it goes off and figures out the steps, grabs the tools, gets it done. It’s like havin’ a foreman who don’t need no sleep.
        Third, we got somethin’ called Spaghettifying DRAM, and that’s six hundred sixty one upvotes. Sounds like a pasta dish, right? But no, it’s about memory that gets all tangled up under pressure, like when you’re runnin’ too many apps at once. They figured out a way to untangle it, make it run smoother. It’s like cleanin’ out your garage, but for your computer’s brain.
        And finally, GLM five point three, and this one’s a doozy. Six hundred sixty five upvotes, and they’re talkin’ frontier coding with emergent cyber capabilities. Now, that sounds scary, but basically it means this AI can write code that’s real good, and it’s learnin’ new tricks on its own. It’s like a kid who’s not just doin’ homework, but inventin’ new math. We gotta keep an eye on it, but man, the potential’s huge. Just don’t let it near your bank account, alright?
        That's your daily byte. Have a great day. Until next time.
        Stories covered
        • Gemini 3.7 Flash — Hacker News
        • DeepSeek Harness developer preview — Hacker News
        • Spaghettifying DRAM — Hacker News
        • GLM-5.3: Frontier coding with emergent cyber capabilities — Hacker News
        • 2 min
        • Tracking down the 16-year-old WAL-reset SQLite bug | EP #130
          Good morning, it's Thursday. This is Guru's Tech Bytes, episode one hundred thirty. Man, oh man, do we got a good one for you today. I was scrolling through the feeds last night, coffee in hand, and I just kept saying "no way" over and over. We got bugs, we got big models, we got stuff that's gonna make your head spin. So buckle up, buttercup, because we're diving straight into the deep end. Heh. Hhh, okay, that's something. Let's get to it.
          First up, we got a story that's pure detective work. Some poor soul tracked down a sixteen-year-old bug in SQLite. Sixteen years! That thing was older than my truck. It’s the WAL-reset thing, and it got over a thousand upvotes because it’s the kind of bug that makes you want to pull your hair out. It’s the kind of thing that only shows up when you’re doing something weird, and this guy just dug and dug until he found it. I love that. That’s real grit, right there.
          Second, we got DeepSeek V4 Pro, the oh-eight-thirteen version. Now, I ain't gonna pretend I understand every little tweak, but the buzz is real. Nine hundred twenty-eight upvotes says people are excited. It’s faster, it’s smarter, it’s the next big thing in open models. And the best part? It’s not locked behind some fancy paywall.
          Third, we got Qwen, the three-point-eight, two-point-four-trillion parameter beast. Six hundred twenty-six upvotes. Now, that number, two-point-four-trillion, that’s just stupid big. I mean, my brain can’t even wrap around that. But what matters is what it can do. It’s pushing the envelope on what’s possible, and it’s making the big boys sweat. It’s like watching a heavyweight fight, and the underdog is landing some serious punches. I’m telling you, the future is now.
          And finally, we got Delta, and it’s got five hundred fifty-two upvotes. Now, this one’s a little different. It’s not about raw power, it’s about something else. It’s about efficiency, or maybe a new way of doing things. I didn’t get all the details, but the community is buzzing, and that’s always a good sign. It’s a reminder that there’s always something new around the corner, always a fresh angle. Keeps things interesting, you know?
          That's your daily byte. Have a great day. Until next time.
          Stories covered
          • Tracking down the 16-year-old WAL-reset SQLite bug — Hacker News
          • DeepSeek V4 Pro 0813 — Hacker News
          • Qwen3.8-2.4T — Hacker News
          • Delta — Hacker News
          • 2 min
          • Stealing Reasoning Traces from Proprietary LLM APIs | EP #129
            Good morning, it's Wednesday. This is Guru's Tech Bytes, episode one hundred twenty nine. And boy, do we got a weird one for you today. I mean, we talk about artificial intelligence a lot on this show, but this story, it’s like something out of a spy movie, except instead of secret codes, it’s about stealing the *thinking* part. The quiet little notes the machine makes before it talks. You know, like when you scribble on a napkin before you order? Yeah, it’s like that, but for a computer brain.
            First up, some smart folks figured out how to swipe the private reasoning traces from those fancy, paid AI models. You know, the ones you gotta subscribe to? They’re not stealing the answers, no sir. They’re grabbing the behind-the-scenes scratchpad, the chain-of-thought.
            Second, we got some genuinely good news for a change, and it ain’t about robots. England is on the verge of wiping out hepatitis C. Can you believe that? A whole disease, gone. They used to think it was a lifelong sentence, but with these new miracle pills and a massive push to test everybody, they’re talking about being one of the first countries on Earth to just... end it.
            Third, let’s get a little philosophical on you. There’s this idea floating around that compression is prediction. Sounds fancy, but here’s the deal: if you can squish a whole bunch of data into a tiny little file, you’ve basically figured out the rules of that data. And if you know the rules, you can guess what comes next. That’s the whole secret behind these AI chatbots, really. They’re just super-powered zip files that learned how to talk. It’s a beautiful, simple way to look at the whole mess.
            And finally, in a move that surprised absolutely nobody, the head of ethics over at OpenAI packed his bags and left. Less than a year on the job. I mean, you gotta wonder what he saw in those board meetings that made him say, “Nope, I’m out.” It’s like hiring a lifeguard and then telling him the pool is full of sharks. It’s a real head-scratcher, but it sure makes for good gossip in the tech world. Heh. Hhh, okay, that's something.
            That's your daily byte. Have a great day. Until next time.
            Stories covered
            • Stealing Reasoning Traces from Proprietary LLM APIs — Hacker News
            • England set to be one of the first countries to eliminate hepatitis C — Hacker News
            • Compression is prediction — Hacker News
            • OpenAI’s head of ethics leaves less than a year after joining — Hacker News
            • 3 min
            • Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows | EP #128
              Good morning, it's Tuesday. This is Guru's Tech Bytes, episode one hundred twenty eight. And boy, do we got a full plate today, so let's just jump right in, shall we? I mean, the stuff coming out lately, it’s like every single day there’s something new that just blows my mind wide open. You know, back in my day, we were happy just to get a calculator that could do square roots. Now we got machines that can think, or at least fake it real good. Heh. Hhh, okay, that's something.
              First up, we got Muse Glimmer. That’s a thirty-billion parameter model, and it’s built for always-on local agent workflows. What that means for you and me is you can run this thing right on your own rig, no cloud, no waiting, no handing your data to some big corporation. It’s got eleven hundred and fourteen upvotes, and I get it. It’s like having a little genius living in your basement, always ready to help, never asking for a raise. It’s the future of personal computing, I swear.
              Second, Mark Zuckerberg is out there attacking the closed AI rivals, saying Meta is going back to open models. And honestly, I gotta hand it to him on this one. I mean, I’m not always his biggest fan, but when he’s right, he’s right. Five hundred and five upvotes for that story, and it makes sense.
              Third, we got the UK’s war on anonymity, and it’s come to America. That’s a scary one, folks. Five hundred upvotes, and it should be higher. They want to know who you are, all the time, everywhere. And that just rubs me the wrong way. Privacy is a right, not a privilege. If you can’t be anonymous, you can’t be free. It’s that simple.
              And finally, Sonic Pi version five is out. Three hundred and seventy-six upvotes, and it’s a gem. It’s a live coding music tool, and it’s just pure fun. You write code, and it makes music. It’s like being a DJ and a programmer at the same time. It’s creative, it’s educational, and it’s a great way to get the kids interested in tech. It’s not all doom and gloom, you know? Sometimes it’s just about making something cool.
              That's your daily byte. Have a great day. Until next time.
              Stories covered
              • Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows — Hacker News
              • Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models — Hacker News
              • The UK's war on anonymity has come to America — Hacker News
              • Sonic Pi v5 — Hacker News
              • 3 min
              • How I use LLMs to learn complex topics | EP #127
                Good morning, it's Monday. This is Guru's Tech Bytes, episode one hundred twenty seven. And boy, do I got a pile for you today. I mean, I was scrolling through the feeds over my coffee, and I kept yelling at my wife, "Honey, listen to this!" She don't care, but you will. We got four stories that'll bend your brain, from learning stuff to, well, your brain itself. So buckle up, grab your thermos, and let's dive into the digital ditch. Heh. Hhh, okay, that's something.
                First up, and this is the big one, how I use these big language models to learn stuff that used to make my head spin. I mean, I ain't no scholar, right? But I asked this AI to explain quantum physics like I was fixing a carburetor. And it worked! It broke it down into pistons and fuel lines. You can ask it to dumb it down, then dumb it down again, then ask it to quiz you.
                Second, and this one's a little creepy, everything you do is being recorded. Not just your texts, but your car, your fridge, your smart TV. They know when you sneeze, man. I read this story about a guy who got a coupon for baby stuff before he even told his wife. It's not a conspiracy, it's just data. And they're hoarding it like gold. So, you gotta think, every click, every pause, every "Hey Siri" is a little note in your permanent file. It's wild, but it's real.
                Third, and this one's a real head-scratcher, taxi drivers rarely die of Alzheimer's. I swear, it's true. All that navigating, remembering streets, that constant mental map-making, it builds up a bigger hippocampus. That's the memory part of your brain.
                And finally, we got a classic from way back in ninety-eight. Cool URIs don't change. That's the rule, and it's still golden. It means if you put a link on the internet, you better make it a good one, because once it's out there, people are gonna bookmark it, share it, and link to it. If you break that link, you're breaking their trust. It's like moving your house but not leaving a forwarding address. The internet remembers, and it don't forgive. So keep your addresses clean, folks.
                That's your daily byte. Have a great day. Until next time.
                Stories covered
                • How I use LLMs to learn complex topics — Hacker News
                • Everything you do is being recorded — Hacker News
                • Taxi drivers rarely die of Alzheimer's — Hacker News
                • Cool URIs Don't Change (1998) — Hacker News
                • 3 min
                • Fastmail offers EU data region | EP #126
                  Good morning, it's Sunday. This is Guru's Tech Bytes, episode one hundred twenty six. And boy, do we got a good one for you today. I mean, we got four stories that just, they cover everything from where your email lives to, get this, a phone that thinks it's a server. Heh. Hhh, okay, that's something. So grab your coffee, settle in, because we are diving headfirst into the weird and wonderful world of tech, and I ain't gonna lie, I'm pretty pumped about this lineup.
                  First up, Fastmail is offering a European data region, and it got four hundred sixty-six upvotes. Now, for the average Joe, that might sound boring, but listen, it's a big deal. It means your emails, they stay in the EU, which is all about that privacy and those strict data laws. No more wondering if your stuff is bouncing around some server in who-knows-where. It's like keeping your mail in a local post office instead of shipping it across the ocean.
                  Second, there's this story called "My server is a phone now," and it got four hundred eighteen upvotes. And I gotta tell you, this one just blew my mind. Some clever fella turned an old Android phone into a full-blown server. We're talking web hosting, file storage, the whole nine yards, all running off a device you could fit in your pocket. It's not gonna replace a big data center, but for a hobby project or a home lab? That's pure genius.
                  Third, we got a deep dive into improving heuristics for A-star pathfinding, three hundred six upvotes. Now, I know that sounds like pure math nerd stuff, and it kinda is, but hear me out.
                  And finally, we got dithered QR codes, two hundred eighty-nine upvotes. You know those ugly black and white squares? Well, someone figured out how to make them look like actual art. They use dithering, which is a fancy way of saying they scatter the dots so the code still scans, but the pattern makes a picture. Imagine a QR code that looks like a mountain landscape or a portrait. It's functional and beautiful, and honestly, it's the coolest thing I've seen in a while. Makes me want to print one on a t-shirt.
                  That's your daily byte. Have a great day. Until next time.
                  Stories covered
                  • Fastmail offers EU data region — Hacker News
                  • My server is a phone now — Hacker News
                  • Improving Heuristics for A* Pathfinding — Hacker News
                  • Dithered QR Codes — Hacker News
                  • 3 min
                  • What happens if an entire class of workers loses faith in their careers | EP #125
                    Good morning, it's Saturday. This is Guru's Tech Bytes, episode one hundred twenty five. I swear I was just telling the wife the other day how these smart programs are changing everything, right? I mean, you walk onto a job site, guys are using their phones to calculate lumber cuts instead of using the old tape measure. It's crazy. But I love it. Anyway, let's get into the news.
                    First up, what happens if an entire class of workers loses faith in their careers. Nine hundred sixty eight upvotes on this one, so people are talking. And I get it, I really do. You spend twenty years laying pipe or turning wrenches, and suddenly some computer can do the math faster. But here's the thing, it can't physically turn the wrench. Yet. I always say, until a robot can fix a leaky toilet in the dark, my job is safe. But it makes you think.
                    Second, DeepSeek V4 Flash 0731. Seven hundred fifty four upvotes. Now, I don't understand all the technical jargon, but Flash means fast, right? And fast is good. My buddy Dave says this thing runs quicker than a cat on a hot tin roof. Heh. Hhh, okay, that's something. But seriously, if these programs keep getting faster, maybe they can help me order parts before I even know I need them. That would save me a trip to the supply house every Monday morning.
                    Third, DeepMind's WeatherNext model achieves breakthrough forecasting cyclones. Three hundred fifty four upvotes. This is huge, folks. I remember when we just looked at the sky and guessed. Now they're telling me a computer can predict cyclones? That saves lives. That saves property. That saves my buddy's fishing boat. If AI can tell me whether to bring a jacket or not, I'm sold. Weathermen better watch out though, right?
                    And finally, Denmark Requires Oral Defenses for Students' Written Work to Counter AI Cheating. Four hundred fifty two upvotes. You know, I actually like this. Back in my day, you had to stand up and explain your work. If you didn't know it, you didn't know it. Kids today, they can just ask the computer to write their essay. But if they gotta stand in front of the teacher and explain it, they better know their stuff. Good for Denmark.
                    That's your daily byte. Have a great day. Until next time.
                    Stories covered
                    • What happens if an entire class of workers loses faith in their careers — Hacker News
                    • DeepSeek V4 Flash 0731 — Hacker News
                    • DeepMind's WeatherNext model achieves breakthrough forecasting cyclones — Hacker News
                    • Denmark Requires Oral Defenses for Students' Written Work to Counter AI Cheating — Hacker News
                    • 3 min
                    • Mario Meets Pareto | EP #124
                      Good morning, it's Friday. This is Guru's Tech Bytes, episode one hundred twenty four.
                      First up, somebody on the internet figured out that Mario, like the Mario from the video game, follows the Pareto principle. You know, the 80-20 thing where 80 percent of the coins come from 20 percent of the levels or whatever. And I gotta be honest, I don't fully understand it, but it reminds me of the time I ate 80 percent of a pizza and only felt 20 percent bad about it, which Lois said was not the same thing, but I think it proves the science. Anyway, smart people are analyzing a plumber with a statistics framework and honestly, respect.
                      Second, AMD bought some company called Taalas, and what they do is they etch AI models right into the silicon, which — I don't know what that means, but it sounds like carving a turkey with the gravy already inside, which is insane and also delicious. You know what this reminds me of, that time Microsoft told me my Windows update would take five minutes and it took four hours and then my whole desktop looked different for no reason. Like, why would you change where the start button is, Microsoft? Why would you do that? Heh. Hhh, okay, that's something. But anyway, AMD says this makes AI inference faster, so good for them.
                      Third, GitHub Actions and Pages are having some kind of degraded availability situation, which is computer talk for "the thing is broken and we don't know why yet." I feel like every time I hear about GitHub it's either down or someone's arguing about tabs versus spaces, and honestly I don't know what either of those things are but I have strong feelings about both.
                      And finally, there's a piece called "Taste Is All That's Left," which is about how now that AI can do basically everything, the only thing humans still got is taste. And I think about that a lot, like when I pick a shirt and Lois says "Peter, that's a tablecloth," and I say "it's got a pattern, Lois, it's got style." That's taste. You can't teach that to a robot.
                      That's your daily byte. Have a great day. Until next time.
                      Stories covered
                      • Mario Meets Pareto — Hacker News
                      • AMD acquires Taalas to boost inference performance by etching models in silicon — Hacker News
                      • GitHub Actions and Pages are experiencing degraded availability — Hacker News
                      • Taste Is All That's Left — Hacker News
                      • 3 min

                      About Guru's Tech Bytes

                      From the publisher's feed

                      A daily AI-generated tech briefing. Top stories from Hacker News, distilled into a quick morning podcast by an automated pipeline.