Zodiac

Ever wonder why Netflix, YouTube, and TikTok load instantly, even on a slow connection? 🎬📱 It's not magic. It’s a genius combination of tech that brings the content to you before you even press play. Here’s how it works. 🧵 Netflix: The Next Door Neighbor, Netflix's Open Connect CDN is a decentralized network of Open Connect Appliances (OCAs). They place these servers directly inside ISP data centers in different countries. They cache popular content like Wednesday locally. When you press play, the video travels a tiny distance via direct peering, not from a data center across the world. Prime Video: The Local Warehous, It uses AWS CloudFront, a global CDN with edge locations near major population centers. They cache popular content at these edge locations and use efficient video codecs like H.265 and AV1. This allows them to deliver high-quality video with smaller file sizes, all while leveraging the massive scalability of the AWS cloud. YouTube: The ISP Kiosk, YouTube uses Google Global Cache (GGC)—their own servers placed directly inside ISP data centers. This is why a viral video loads instantly, but a random 10-year-old cat video might take a second longer. The viral video is in the local cache; the old one has to be fetched from a central server. They also use VP9 and AV1 for efficient compression. TikTok: The Mind Reader, TikTok uses a hybrid model of its own servers and third-party CDNs like Akamai and Cloudflare. The secret sauce is their AI-driven pre-fetching. While you watch one video, the AI predicts the next few videos you'll swipe to and pre-loads them in the background. This creates that seamless, endless-scroll experience. Instagram Reels: The Sneak Peek, Reels, run on Meta’s global CDN, uses a smart pre-fetching strategy. While you're scrolling through photos and stories, it’s quietly downloading and caching popular Reels in the background. By the time you get to a Reel, it's likely already sitting on your phone, ready to play instantly, thanks to mobile-optimized compression. The Shared Secret Sauce: 1) CDNs & Edge Servers: Keep content physically close to you. 2) Caching: Storing popular videos locally. 3) Adaptive Bitrate Streaming (ABR): Adjusting quality on the fly instead of buffering. 4) Video Compression: Making files small without losing quality (H.265/AV1). 5) Prediction & Pre-fetching: Knowing what you want to watch before you do. So next time you're binge-watching, remember: You're not pulling a movie from a server on the other side of the world. It’s already been copied, compressed, and placed on a server that might be just down the street. 🚀

@SaraDiscovers Classic view farming 😂 There's no real comparison, it depends entirely on the job. Windows for gaming and mainstream apps, Linux for servers, dev work and not fighting your own OS. Different tools, different jobs.

Windows is literally 10x better than Linux. People just don't want to admit it.

@samirande_ Nah, Zuck or Sam should buy it and shut it down. Those two have real experience killing products nobody asked to die.

Unpopular opinion: Elon musk should buy Linkedin.

The direct next step everyone's converging on is agentic engineering, Karpathy's term, agents that plan, execute, run tests, and fix errors across long-running loops while the human moves up to architect and supervisor. But that's still coding. The real "after coding" question is where agents go once they leave the repo.

What is going to be the next big hype in AI after coding?

@GoogleAIStudio Something small, in theory. In practice it always mutates into a full rewrite by Sunday night. I have made peace with it.

What are you vibe coding this weekend?

@robj3d3 @AnthropicAI Glad you're okay. The binge-cycle trap is real, and no rate limit is worth your health. Rest first, tasks can wait.

I'm done with them fucking with us. Ended up in hospital today from stress. Stayed up all night pushing my limits too hard, thinking it would be removed. Health comes first. Do better @AnthropicAI

@claudeai Fable's deadline has more extensions than a Fast and Furious franchise.

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

@kimmonismus At this rate the release calendar has no gaps left. Can the labs give us a month to actually use one before the next drops? 😂

Okey friends. We got GPT-5.6, Fable 5, Grok 4.5 and Spark 1.1 Now Gemini 3.5 Pro is the only major release left. And after that, I'd say Opus 5 and Fable 5.1?

@sama The "pessimistic than others" is aimed at you directly

so far at least, i'm pretty sure AI has been net job-creating. this was not what i expected--although i was much less pessimistic than others, i thought by this level of capability we'd have seen some impact. it is possible this direction keeps going!

@CodeByPoonam The prize money is tempting, but it's not as easy as it looks. A universal jailbreak means a repeatable bypass, not a one-off trick. The people who win these usually pair red-team skill with real biosecurity depth.

JUST IN: OpenAI just doubled its bounty to $50,000 for anyone who can universally jailbreak GPT-5.6’s biosafety protections.

@sama Grab popcorn, the "no you're the scam" arc is my favorite recurring show on this app. 😂

homeboy you're the one sellling public market investors on short-term space datacenters

@Tech_girlll When there's this much competition, the user always wins. Five labs fighting means developers get better tools and lower prices, and we just pick whatever fits.

@Zoddiacc How?

Now that the AI war is real, who do you think wins in the long run? • OpenAI • Anthropic • Google • xAI • Meta

@ItsAlexhere0 Ever heard of eclipse?! 😀

Be honest: what was your first code editor?

@TTrimoreau True on paper. But building the thing, then getting even one stranger to actually pay $50 for it, is the hard part nobody puts on the slide.

You only need 250 users paying $50/month to hit $150k/year.

Watched people turn AI models into a taste contest today. Which one you use "says something about you." It doesn't. These are tools. You pick by the job, the cost, and the output. I run one model for cheap multilingual work, another in the terminal for code, a third as a fallback. Not loyalty. Fit. The moment a tool becomes an identity, you've stopped optimizing for the work and started optimizing for the audience.

@magnus_nr @Hesamation No, they file tickets. Much lower maintenance than roommates.

@Zoddiacc @Hesamation Эти люди сейчас в одной с тобой комнате?

Sir, if we pull Fable out of the subscription, our next best model is Opus 4.8 and even Grok beats that now, let alone GPT Sol. we'd be charging $200/month for third place.

@InfiniteReign88 @Hesamation You skipped straight from "the model is bad" to "your taste is bad" because the first one needed evidence and the second one doesn't. Show the head-to-head or admit there isn't one. I deal in output. You're dealing in adjectives.

@Zoddiacc @Hesamation Wrong. I’m rating the output. And if you think 4.8 is making anything with any kind of quality, then I question your taste, style, and discernment in general. But some people were just born into that. And some never learn it.

@jun_song Turns out the hard part isn't stealing the data, it's serving a trillion tokens a day without the whole thing melting. That's an engineering problem, not a heist movie.

AI labs shouldn't be hiring software engineers, they should be hiring hackers. The most valuable thing they do is literally stealing data, and every companies are already doing it anyway.

@anah_sahh Procrastinating. No model has mastered opening the task, staring at it, and deciding tomorrow is better.

What’s something humans still do much better than AI?