Alex Vu

@adiix_official Anyone who actually doing works in local llm know these are pretty bad value. You can run, but too slow. I bash the dgx spark, but even it is better than this.

AMD CEO Lisa Su just killed Nvidia’s $4,000 AI box with a $1,499 lunchbox. She walked on stage, held it in one hand, and ran a 235 billion parameter model live. No data center. No cloud. No rented GPU. The chip inside is something nobody saw coming. AMD’s Ryzen AI Max+ 395 is the first x86 silicon where CPU and GPU share the same 128GB of memory. That single trick lets a desktop run models that used to need a server rack. Out of those 128GB, Linux hands the GPU 110GB to play with. For context, an RTX 5090 gives you 32GB. A 4090 gives you 24. This box gives you more than three times either of them, in a chassis the size of a thick paperback. The benchmark that broke the room: this chip beat an Nvidia RTX 5080 by more than 3x on DeepSeek R1 inference. A $1,499 lunchbox outrunning a $1,000 discrete graphics card on a real AI workload. Nvidia spent a decade convincing the world you needed their hardware for serious AI. AMD just put that on a desk for half the price. Here is what nobody is telling you. A heavy AI user right now pays $200 for Claude Code Max, $200 for ChatGPT Pro, $20 for Cursor, $20 for Gemini. That is $5,280 a year leaving your account. The box pays itself off in 9 months and then runs free for the rest of its life. Install Ollama. Pull Qwen3 235B. Point Claude Code at localhost. Same interface you already use, except now nothing leaves your machine, nothing costs per request, and no company throttles your usage at 3am when you finally have time to build. This is the moment every AI subscription becomes optional. Lawyers stop fearing OpenAI leaks. Developers stop watching the token meter. Founders stop renting H100s for prototypes that never ship because the bill scared them. The first thousand people to figure this out will own the next two years of private AI consulting. Save this, and read the full breakdown article below you are watching the next shift hit before everyone else does.

@noor36758 May?

the weekend really humbled him 💀

@Snixtp Not gonna happen but I would love for it to happen.

Mythos this, Fable that Watch it get mogged by Qwen3.7 27B

@iulianlita @0xTib3rius God forbid the ignorance, and the trolls. Luckily for humanity, most of these posts are trolls.

@0xTib3rius it's 2026 and people still use these garbage tests to assess the capability of a model... 🙄 next, you should ask it to count to infinity and back, twice.

My god, its incredible.

@github @AnthropicAI @code 1 prompt would cost 50 AI credits correct

📣 Claude Fable 5, the first in @AnthropicAI's Mythos model class, is now generally available and rolling out in GitHub Copilot. It is designed for long-horizon, autonomous coding and knowledge-work tasks. Try it out in @code or the GitHub Copilot app. ⬇️ github.blog/changelog/2026…

@ABralder @awnihannun The trick is to load all those 20b, but infer 3B at a time.

@robberviet @awnihannun But the trick is that they never load those 20B params, they have like a "first pass" that classifies the prompt and derive it to an "expert" that may well be 10B, 5B, or less.

It's very cool that Apple shipped a 20B parameter on-device. You can't put 20B parameters in RAM at any reasonable precision. To make it work they are using pretty exotic architecture by today's standards. A small model predicts from the query (or prompt) which experts to load from Nand into RAM. The key distinction from a typical MoE is that you do this once per query and then generate all the tokens with the same experts (instead of switching the experts for every token).

@Marcela252016 Enjoy other things

Es 1995. Te despiertas. No hay Wi-Fi. No tienes celular. No tienes computadora. No hay Facebook/Instagram/TikTok/X ¿Qué es lo primero que haces?

@haider1 Price

something's off if mythos is really as good as anthropic says, why release opus 4.7 and 4.8 first? both were barely ahead of 4.6 so they're telling us mythos is a massive leap, but they couldn’t make 4.7 or 4.8 even close? is the lineup going to be: mythos > opus > sonnet > haiku?

@NewsFromGoogle Wow that's good news. Great with stepping plan too at 1TB instead of jump straight to 2TB.

We're reducing the cost of our Google AI Plus plan and doubling your storage space. 💰 New price: $4.99 USD/mo or local equivalent (down from $7.99 USD) ☁️ New storage: 400GB (doubled from 200GB) Enjoy!

@pankajkumar_dev Why svg generation is a benchmark now? How many people work is generate SVG?

Mythos 5: We're Not Ready - Mythos is extremely strong at SVG generation, producing highly detailed outputs that can take several minutes to create - It is exceptionally good at creating graphics, games, websites, and complex UI designs - Mythos is expected to set a new bar for web development and frontend generation, especially for design heavy workflows - It can also generate surprisingly good music through code - Anthropic's data suggests Mythos can achieve up to 52x training code speedups in certain optimization tasks, compared to roughly 4x for skilled humans on similar workloads - Mythos is expected to be extremely expensive, the public version will likely be a nerfed version of the current testing model

@intheworldofai Not offical information then it's worth nothing.

DeepSeek GUI just dropped and it’s exactly what power users needed. A clean, local-first desktop agent workspace for DeepSeek with: • Code Mode - real file ops, planning, reviews, approvals & project context • Write Mode - excellent Markdown editor with smart AI assistance • Kun runtime - insane token efficiency (cache-first design, high hit rates) • IM integration (Feishu/Lark/WeChat) + background scheduled agents Local, private, cross-platform (macOS/Windows), and already feeling like a polished Codex-style agent environment: youtu.be/fRjWpcA40hs

@Taniyatweets_ As always SQL.

In your experience, which programming language is future proof?

@TheAhmadOsman Skill issue. I can but not mean I should.

I never understood the idea of putting a middleman like Tailscale between you & your local nodes WireGaurd exists, and in the age of agents even having a Dynamic IP or firewall issues could easily be solved So why not spend the time to learn and cut the middle man? Skill issue?

@vibeonX69 PHP

Name a programming language that deserves more respect

@Anaya_sharma876 The older you are the less game you play. And you can always play using more proper devices like PS.

Choose one for rest of your life Gaming or mackbook

@Rmcfmjay Most of the time we don't remember the 2nd. I don't remember any. Just pretty sure when Ronaldo or Messi won the the others probably the 2nd

His Ballon d'Or was so clear-cut that nobody remembers who came in second that year.

@usanewshq Compared to what? Human is stationary compared to earth, but?

Explain using physics: Is the drone stationary or moving at 50 kph?

@Surendar__05 dev tools: git, uv, docker, zed...

Finally installed Linux. What’s the first thing you’d install?

@yuhasbeentaken The web? No interests. The model? Yes.

the new version deepseek is comming...