kravisw

[Psalter @Psalteric]

Your ad blocker could fit behind your router.

meet ESP32-C3 AdBlock, an open-source DIY DNS ad blocker by ZedAxis, built around a roughly $2 ESP32-C3 Super Mini board

inside: 400 KB of RAM, 4 MB of flash, built-in Wi-Fi and a USB adapter, tucked into a printed case

no subscription

no browser extension

no separate power brick

plug it into your router’s USB port for power, connect over Wi-Fi and set it as your DNS server to block listed ad and tracker domains.

compact 40-bit hashes save RAM. a web dashboard lets you check blocked requests and add domains,

while open-source code and case files let you build your own.

follow for more hardware you can see working

[kravisw @kravisw]

@NanospeedGamer en la version de firmware de ps5 digo

[kravisw @kravisw]

@PizzaL0sViernes como ves un mac neo

[kravisw @kravisw]

@ssocialtrading @_Nam_Libertas_ no crees que la noticia de que claude ha descubierto otro enzima de crispr le puede sentar mal?

[kravisw @kravisw]

@ssocialtrading que opinas de $SECZ con el tema de la tokenizacion.

[Alfonso C. Suárez @AlfonsoCSuarez]

Quiche de cebolla caramelizada, queso feta y jamón cocido

Lo habitual suele ser ver quiches con queso de cabra o algún queso azul, pero el feta es un ingrediente perfecto para este plato. Tanto por el sabor (ese salado tan característico) como por la textura, ya que al no derretirse por completo queda perfecto en la mezcla de huevo, leche y nata.

Y un punto importante de esta receta: el feta pega muchísimo con la cebolla caramelizada. Me habría gustado caramelizarla más, pero no tenía tiempo. Pese a ello, el sabor dulce estaba ahí.

El jamón cocido completa el trío ideal para una quiche. Y solo hay que añadirle un toque.

Eso sí, me pasé un poco con el feta (pero no me arrepiento 😂).

Receta demasiado sencilla para un resultado que no lo parece. Y nunca te olvides de la nuez moscada, clave en esta elaboración.

¿Tienes alguna combinación favorita para la quiche? 👇

[Brais Moure @MoureDev]

Han creado un catálogo abierto de 302 ejercicios con 906 ilustraciones y paquete npm para desarrollo web/mobile.

Si estabas pensando en crear una app de fitness, entrenamiento o proyecto educativo, te permite buscar ejercicios por músculo o equipamiento

→ https://t.co/9HQw2xPB5f

[RoboRocks @malagojr]

Delete 99.8% of your online presence from the internet.

Here’s a step-by-step guide:

[@goldengames7890 @RobinsonPauli18]

Goldengames PS5 WebKit Autoloader v1.0.4

Auto Jailbreak
etaHEN 2.5B
Manual Payload Launcher
Kstuff Lite
Payload Manager
WebSrv
ShadowMountPlus
Goldengames persistent payload sender

Full credit and special thanks to itsPLK for creating the original project that made Goldengames PS5 Autoloader possible. ❤️

Goldengames PS5 WebKit Autoloader is a fork and modified version of itsPLK’s PS5 WebKit Autoloader / Auto Installer v0.3.0.
https://t.co/GMXbREpFBR

[Tom Dörr @tom_doerr]

OpenGym is a self-hosted fitness tracker built with React and Docker. It automates weekly training plans with animated demos and charts body weight progress, ensuring all sensitive workout data remains local rather than stored on third-party servers.

https://t.co/UcNtIdyYoO

[Javi López ⛩️ @javilop]

Gente que se enfada porque ese producto me parezca una mierda... si te enfadas tienes dos problemas: enfadarte y desenfadarte 😂😂😂

No esponsorizado (no me pagan por esto, ni los conozco), pero si queréis probar algo BUENO, mismamente en Vacaymuu tienen esta delicia:

[kravisw @kravisw]

@nextbigtrade Price target?

[Jordy @couldbeautism]

You can now jailbreak your consoles from 9.00 - 12.00 from webkit. Im sure some other devs will setup there own github pages, dns servers etc https://t.co/sUOPeiiYME enjoy! also if you don't mind supporting please help cyber. Credit > @egycnq @Sonic_Iso

https://t.co/HtXLOaIAMA

[kravisw @kravisw]

@ElRincondeVaro @xVaaro Yo compré hace 4 días la ultima palit 5080 de Amazon por 1249

[Bad theory labs @Badtheorylabs]

Sonnet-class model. 9.96 GB. Your hardware.

Today we shipped BTL-4 and BTL-4 Compact.

BTL-4 scores 78.4% on SWE-bench Verified frontier-class agentic coding, from a 35B open-weight model you can download.

It's a mixture of experts with ~2.1B active parameters per token, so it costs a large model's memory and a small model's compute. Trained only on trajectories whose code actually ran and passed its tests. +4.3 on BFCL v4 AST over base, paired. 262K context.

BTL-4 Compact is that model in a single 9.96 GB file. 2.30 bits per weight.

It fits on a 16 GB card — with the full 262K context still resident, because this architecture's KV cache is only ~20 KB per token.

And no custom runtime: it loads in stock llama.cpp, Ollama and LM Studio. No fork, no patched backend, no reconstruction step. One file, one command, a running agent.

Compressing 35B this far normally destroys a model. We expected to need the whole stack measured precision islands, behavioural bisection, a repair adapter. We measured first, and almost none of it mattered.

At two bits you get four levels, and one outlier in a group of 128 wastes three of them. Replacing min/max ranging with a per-group clip search moved behavioural retention from 77.1% to 95.8% at an identical byte budget. Twelve seconds of compute. No training, no calibration data. Range selection was the whole problem.

Two things we were wrong about, and the measurements say so. Protecting the output head the standard advice, and our own prior recommendation — is worth nothing: head and embedding at 4-bit retained 118 of 118. And sub-2-bit is unreachable this way: binary experts scored 0 of 118.

Then we tested whether the packed file survived. On a sealed 120-item gate, Compact reproduced 111 of the 118 behaviours the full-precision model got right 94.1%. 95.0% short-form factual, 100% grounded extraction, 87.2% false-premise rejection.

The gate and the harness ship with it. Reproduce every number in three lines.

BTL-4: https://t.co/DZY9RVHL4E Compact: https://t.co/b68dgtzGxK Apache-2.0.

[kravisw @kravisw]

@vinates Tiene el gpt 5.6 no?

[Cyril 🏁 @SlamingDev]

https://t.co/HQrYGfP2hG

[Phoenix Yin @Phoenixyin13]

今天CNBC直播直接原地爆炸。

Palantir CEO Alex Karp 上午参与节目，聊着聊着突然精神失控，近20分钟全程情绪拉满，主持人几次想打断都打断不了。

他疯狂输出，就一个核心意思：

现在OpenAI、Anthropic那些大模型根本就是个坑货，企业花大钱按token付费，交出去的数据和核心竞争力全被大厂偷去训练模型，等于自己花钱养对手，拉完了。

他的意思很明确，就是说这些美国的闭源模型被irresponsibly over-sold，即一种不负责任地过度吹捧，闭源大模型的核心就是把美国企业和军方的命脉外包给几家实验室。

注：Palantir长期给美国军方、情报部门、战场提供数据分析和AI工具，是能影响生死和国家机器的至高层级。

[Javier Costas Franco @javiercostas]

El Tesla Model Y Tieso Edition puede mejorar mucho con AliExpress 😱