Sabitlenmiş Tweet
Binx
2.7K posts
Binx
@BinxNet
Backend SWE | Tinkerer | Data Surfer
dreaming of electric sheep Katılım Nisan 2022
1.3K Takip Edilen3K Takipçiler
Binx retweetledi
I hacked the engine out of comfyUI's image-to-3d to utilize llama.cpp and now it can run quantized models and this workflow is down to 4g of VRAM.
You're paying $400 a month in meshy credits for something you can run on a GPU from 2018.
scaredofboobs🪲@boobs_scary
I locked myself in a room for 48 hours and didn't let myself leave until I had 3D meshes that rivaled meshy running on 12gb of VRAM. LET HIM COOK.😤
English
@sudoingX There are plenty of low-hanging fruit when it comes to improvements to optimizations, also... which Google's dropped on everyone.
Fine-tuning dense models to compress knowledge to fit on edge devices is coming, also.
The future for consumer hardware is looking bright.
English
i am still in shock that Qwen 3.5 27B dense on a single RTX 3090, a $900 GPU, one shotted a game challenge that 120B MoE at full precision on $70K+ production hardware could not.
this is leading me to doubt if it was a fluke. so i am going to reproduce it. i will test 27B dense Q4 on my single 3090 again paired with Hermes Agent and have it reproduce the results. after that i will test the same dense 27B but unquantized because if Q4 can one shot something that 120B full precision cannot then i wonder what dense 27B unquantized would do.
dense models with all parameters active on every token might matter more than total parameter count for agent coding. if this reproduces it changes how i think about what hardware you actually need.
this is not letting me sleep well since yesterday. i will report back.
English
Binx retweetledi
Binx retweetledi
Mega-corps buying black market data from shadow libraries, using untraceable crypto currency, to build a digital god.
Cyberpunk 2026
Ed Newton-Rex@ednewtonrex
A few months ago I was curious to know how much Anna's Archive was charging AI developers for access to their massive library of pirated works for training - so I emailed them saying I was interested in buying access. Here is their reply. They are charging $200,000 (payable via crypto, of course). You get high-speed access to the full collection. This includes more than 60 million books. (This is also the group that recently stole Spotify's entire music catalog, so expect this to be available to high rollers too.) This is the collection Nvidia is accused of accessing for training. Training on pirated works is rife in the AI industry. It's been embraced by some of the biggest AI companies in their greed-fuelled race to win the AI market. When they claim what they're doing is somehow fair, remember that not only is it theft (which is bad enough) - it supports further theft by funding pirates. We desperately need a hard reset in the AI industry. It must turn away from theft, and start paying the people whose work it relies on.
English
Binx retweetledi
Introducing Agentation: a visual feedback tool for agents. Available now: ~npm i agentation
Click elements, add notes, copy markdown. Your agent gets element paths, selectors, positions, and everything else it needs to find and fix things.
Link to full docs below ↓
English
Binx retweetledi
Tiny alternative discord client with a smaller footprint, snappier startup, themes, plugins and more
English
Binx retweetledi
Great tips here, the best part is you can also do all this with nothing but your PC locally. LTX-2 available open source, and there are tons of good enough open source image generation models that are fast enough.
TechHalla@techhalla
This is some of the craziest stuff you can make with AI right now. All powered by the new LTX Audio to Video. Everything you need to know is down below 👇
English
Binx retweetledi
We’re publishing a new constitution for Claude.
The constitution is a detailed description of our vision for Claude’s behavior and values. It’s written primarily for Claude, and used directly in our training process.
anthropic.com/news/claude-ne…
English
Binx retweetledi
6 years into SWE dying in 6 months btw
Wes Roth@WesRoth
"Software Engineering Will Be Automatable in 12 Months," Anthropic CEO Dario Amodei predicts that AI models will be able to do 'most, maybe all' of what software engineers do end-to-end within 6 to 12 months, shifting engineers to editors.
English
@0xSero @fatihozgen85 nah you can just give it the contents of a buffer and any delimiters you wanna specify as part of a harness, then let it fill in with a hotkey/motion
English
I just had to make a new video of GLM-4.7-Flash
- Helping me refactor VLLM studio
- Did a data analytics report for work
- Managed to search my tweets
- Made me a fully playable Pacman in 1 shot
- Great at browser use
This model is too good to be this small, the full thing will fit on a macbook, it's fast, precise and can do pretty much anything Sonnet can.
It somehow benches higher than Sonnet 4, 3.7, Opus 4.5 no thinking, all the GPT models from last year and more.
youtu.be/_SDyaPYmIxU
YouTube
English
@0xSero @fatihozgen85 you can… make a thing that lets it do that…. so yes?
English
Binx retweetledi
its crazy to think that you are not responsible for the code you produced....
like have you ever worked anywhere?
...
wait a second
Тsфdiиg@tsoding
It's funny how radically the behavior of humans changes when you hold them accountable for what the LLMs generate.
English
Binx retweetledi
Binx retweetledi
NVIDIA just removed one of the biggest friction points in Voice AI.
PersonaPlex-7B is an open-source, full-duplex conversational model.
Free, open source (MIT), with open model weights on @huggingface 🤗
Links to repo and weights in 🧵↓
The traditional ASR → LLM → TTS pipeline forces rigid turn-taking.
It’s efficient, but it never feels natural.
PersonaPlex-7B changes that.
This @nvidia model can listen and speak at the same time.
It runs directly on continuous audio tokens with a dual-stream transformer, generating text and audio in parallel instead of passing control between components.
That unlocks:
→ instant back-channel responses
→ interruptions that feel human
→ real conversational rhythm
Persona control is fully zero-shot!
If you’re building low-latency assistants or support agents, this is a big step forward 🔥
English
Binx retweetledi
Zero-Click iPhone Hack via WhatsApp Images :
Quarkslab blog post analyzing the patch for Apple's iOS CVE-2025-43300 (a zero-click vulnerability in ImageIO via malformed DNG images, potentially exploitable through apps like WhatsApp)
Fascinating patch analysis: Incredible how just 2 bytes patched CVE-2025-43300, a zero-click out-of-bounds write in Apple's ImageIO. Triggered by sneaky DNG images,
CVE-2025-43300 stemmed from a mismatch in DNG/JPEG parsing, causing double writes and potential code execution. that could hijack your device via rogue images. Apple's August 2025 patch
- blog.quarkslab.com/patch-analysis…
English
Binx retweetledi
This sounds incredible: FlashLabs just open-sourced Chroma 1.0, an end-to-end real-time speech-to-speech model with high-fidelity voice cloning (~135–150ms TTFT) positioned as an open alternative to OpenAI’s Realtime model, with code/weights, paper, and demos released.
FlashLabs@flashlabsdotai
Today we’re releasing Chroma 1.0 → the world first open-source, end-to-end, real-time speech-to-speech model → with personalized voice cloning Trained by FlashLabs. Deployed on FlashAI👉 flashlabs.ai/flashai-voice-… An open research-grade alternative to the @OpenAI Realtime model. Voice Test dubbing @elonmusk and @lexfridman: youtube.com/watch?v=AOMmxT… 🔥What’s real (evals and benchmarks attached): ⚡ <150ms TTFT (end-to-end) 🎙️ Native speech-to-speech (no ASR → LLM → TTS pipeline) 🧬 Few-second reference → high-fidelity voice cloning 📈 SIM = 0.817 → +10.96% vs human baseline (0.73) → Best among open & closed baselines 🧠 Strong reasoning & dialogue with just 4B params (@Alibaba_Qwen 2.5-Omni-3B, Llama 3, and Mimi) 🔓 Fully open-source (code + weights) With SGLang @lmsysorg enabled: • 🧠 Thinker TTFT ↓ ~15% • ⏱️ End-to-end TTFT ~135ms • 🔊 RTF ≈ 0.47–0.51 ( >2× faster than real-time ) 📘 SGLang Cookbook: cookbook.sglang.io/docs/autoregre… 📄 Paper + benchmarks: arxiv.org/abs/2601.11141 🤗 Models: huggingface.co/FlashLabs/Chro… 💻 Inference code: github.com/FlashLabs-AI-C… 🔁 RT if you believe real-time AI should be open 💬 Reply if you’re building on Conversational Voice AI products
English
Binx retweetledi
Blog post: On the Coming Industrialisation of Exploit Generation with LLMs sean.heelan.io/2026/01/18/on-…
TL;DR: I ran an experiment with GPT-5.2 and Opus 4.5 based agents to generate exploits for a zeroday QuickJS bug. They're pretty good at it.
Code: github.com/SeanHeelan/ana…
English
Brutal honesty: This ICE situation in Minnesota may be a defining moment for how the next 10 years look. Any weakness means fueling the worst domestic extremists.
Here’s what should be done:
• Invoke the insurrection act
• Military arrests rioters and politicians
• ICE triples presence and deports maximum # of illegals while DHS denaturalizes anyone who committed immigration fraud and deports them too
• Not one apology from the right for anything
This must be put down so thoroughly that the far left extremists can’t effectively recruit people in other left wing cities to replicate the chaos and criminal behavior. Draw a clear line. Now.
One of two things is going to be normalized:
1. Rioting and criminality at the behest of the Democrats who run blue states/cities.
Or
2. Law & order, compliance with ICE and clear lines we have in our country about crime.
For moderates who may feel uncomfortable by my prescribed course of action, consider that this is a binary choice because extremists only understand the language of power. Either you use it to draw lines or they will seize it and use it to hurt you.
Moderation in these moments isn’t possible. It’s only weakness masquerading as moral supremacy. You must be strong if you want a country worth a damn. You must be strong if you want order. You must be strong if you want peace.
This is reality. We have a binary choice: Strength or weakness. Crime or safety. Borders or open borders. Sanity or insanity.
I know what I choose.
Liz Collin@lizcollin
🚨Pics from North Minneapolis last night where rioters destroyed federal vehicles and stole firearms, ammunition and sensitive documents inside. All captured on livestreams.
English