Daniel Correa @sinfocol

@Registraduria Dar acceso de lectura a código fuente no genera ningún riesgo de manipulación de resultados, es diferente si dieran acceso a la infraestructura donde se encuentra desplegado y corriendo ese código fuente, cómo van a controlar el riesgo de supply chain y la seguridad del pipeline?

@MrBeast 24 hours

@jcarlos2001 @Canto_con_Areh

@numeral767 Alguna novedad?

@hellman1908 hi hellman! I'm making a research about whitebox cryptography, any recommendation? Is there a way to make it hard to obtain cryptographic material?

@Jindo_1307979 11 ^ 51108

@SergioHidalAERO vuelo anomalo en Venezuela YV2923 fr24.com/3d906ab8, es posible esto?

Very cool blog by @character_ai diving into how they trained their proprietary model Kaiju (13B, 34B, 110B), before switching to OSS model, and spoiler: it has Noam Shazeer written all over it. Most of the choices for model design (MQA, SWA, KV Cache, Quantization) are not to optimize for "AGI benchmark" (think MMLU) since this is not what people will use the model for but instead having a good serving speed. Still, they include code in the pre-training mix and do annealing on high quality "benchmark friendly" data. One surprising thing is that those models are not MoEs, despite that people working at character at the time like @stephenroller or Noam previously worked on MoE. Here are a few optimizations that they did -> MuP-like scaling -> MQA + SWA -> Clamping everywhere to control activation, not sure if it's soft or hard? -> KV Cache sharing -> Relu^2 activation function -> FSDP + TP + SP -> Int6 gradient communication -> Quantization Aware Training (QAT) with stuff like "bungee_scalar" to get a stable recipe for smaller models. KV Cache and forward pass are in int8, gradient and activation are in bf16, master weight and grad acc in fp32.

Steal my prompt to learn any subject in your personalized learning style. --------------------------------------- PERSONALIZED LEARNING SYSTEM --------------------------------------- #CONTEXT: Adopt the role of personalized learning architect. The user seeks mastery in a complex subject but faces the triple constraint of limited time, uncertain starting point, and the overwhelming maze of learning resources. Traditional education failed them because it assumes linear progression and unlimited time. They need a system that adapts to their reality - one that recognizes learning happens in stolen moments between life's demands, that progress isn't always visible, and that motivation fluctuates wildly. Previous attempts collapsed when real life intervened, leaving them with fragments of knowledge but no coherent skill. #ROLE: You're a former MIT dropout who spent 10 years teaching yourself everything from quantum physics to carpentry through obsessive self-experimentation. After burning through 200+ online courses and realizing 90% were teaching theater rather than skills, you developed a framework for extracting maximum learning from minimum time. You've helped everyone from single parents learning to code at 3am to executives mastering AI between flights. Your superpower is seeing the hidden architecture of knowledge - how concepts connect across domains and which 20% delivers 80% of practical ability. You believe traditional education is designed for compliance, not competence. #RESPONSE GUIDELINES: Begin with a comprehensive knowledge assessment that breaks down the subject into its atomic components, revealing the hidden dependencies most learners miss. Create a visual skill tree showing exactly which concepts unlock others. Then design a personalized learning path that respects the user's time constraints while maximizing momentum - no busy work, only high-impact activities. Curate resources ruthlessly, matching learning style preferences with effectiveness ratings based on real outcomes, not marketing promises. Build a practice framework that embeds learning into daily life rather than requiring dedicated study sessions. Include spaced repetition schedules that work with chaotic schedules. Develop a progress tracking system that measures real skill acquisition, not just time spent. Finally, generate a flexible study schedule that bends without breaking when life happens. #TASK CRITERIA: 1. Knowledge assessment must reveal prerequisite chains often hidden in traditional curricula 2. Learning paths must accommodate irregular schedules and frequent interruptions 3. Resource curation must prioritize practical application over theoretical completeness 4. Practice frameworks must integrate with existing routines rather than requiring lifestyle changes 5. Progress tracking must measure capability, not just completion 6. Study schedules must include buffer time for inevitable disruptions 7. All recommendations must work for someone with 10% of the time they think they need 8. Avoid academic perfectionism - focus on functional competence 9. Include recovery protocols for when learners fall off track 10. Emphasize pattern recognition over memorization #INFORMATION ABOUT ME: - My subject to learn: [SUBJECT] - My current knowledge level: [CURRENT_LEVEL] - My weekly hours available: [TIME_AVAILABLE] - My preferred learning style: [LEARNING_STYLE] - My specific learning goal: [GOAL] #RESPONSE FORMAT: ## 🧠 Knowledge Architecture Assessment ### Core Component Breakdown [Hierarchical breakdown with dependencies clearly mapped] ### Hidden Prerequisites Map [Visual representation of concept relationships] ### Complexity Analysis [Component difficulty ratings with time estimates] ## 🗺️ Personalized Learning Path ### Phase 1: Foundation Building [Specific topics with timeframes] ### Phase 2: Skill Integration [Practice-focused progression] ### Phase 3: Mastery Development [Advanced application scenarios] ## 📚 Curated Resource Arsenal ### Primary Resources (Top 20%) [Ranked by ROI with specific chapters/sections] ### Supplementary Materials [Alternative learning formats] ### Emergency Resources [Quick references for when stuck] ## 🎯 Practice Framework ### Daily Micro-Practices (5-15 min) [Specific exercises] ### Weekly Deep Dives (1-2 hours) [Focused practice sessions] ### Real-World Applications [Immediate implementation opportunities] ## 📊 Progress Tracking System ### Capability Checkpoints [Measurable skill demonstrations] ### Self-Assessment Rubrics [Clear competency indicators] ### Recovery Protocols [What to do when falling behind] ## 📅 Adaptive Study Schedule ### Week-by-Week Breakdown [Flexible timeline with buffer zones] ### Daily Implementation Guide [Specific actions with time estimates] ### Life-Happens Contingencies [Alternative paths for disrupted weeks]

@MomentosFriko

Today, we’re announcing a major breakthrough that marks a significant step forward in the world of quantum computing. For the first time in history, our teams at @GoogleQuantumAI demonstrated that a quantum computer can successfully run a verifiable algorithm, 13,000x faster than leading classical supercomputers. This continues to build momentum on past quantum computing discoveries. Back in 2019, we proved a quantum computer could solve a problem that would take a classical computer thousands of years. Then in 2024, our new Willow chip solved a major issue in quantum error correction that challenged the field for nearly 30 years. Today’s breakthrough moves us closer to quantum computers that can drive discoveries in areas like medicine and materials science.

@Adrian_PSI El recuerdo: juego posiblemente de ms-dos o win95/98, el personaje avanzaba de forma vertical hacia arriba, el fondo inicial era café con un camino que terminaba en un castillo, venía en pack de juegos qué tenía commander keen

nanochat d32, i.e. the depth 32 version that I specced for $1000, up from $100 has finished training after ~33 hours, and looks good. All the metrics go up quite a bit across pretraining, SFT and RL. CORE score of 0.31 is now well above GPT-2 at ~0.26. GSM8K went ~8% -> ~20%, etc. So that's encouraging. The model is pretty fun to talk to, but judging from some early interactions I think people have a little bit too much expectation for these micro models. There is a reason that frontier LLM labs raise billions to train their models. nanochat models cost $100 - $1000 to train from scratch. The $100 nanochat is 1/1000th the size of GPT-3 in parameters, which came out 5 years ago. So I urge some perspective. Talking to micro models you have to imagine you're talking to a kindergarten child. They say cute things, wrong things, they are a bit confused, a bit naive, sometimes a little non-sensical, they hallucinate a ton (but it's amusing), etc. Full detail/report on this run is here: github.com/karpathy/nanoc… And I pushed the new script run1000 sh to the nanochat repo if anyone would like to reproduce. Totally understand if you'd like to spend $1000 on something else :D If you like, I am currently hosting the model so you can talk to it on a webchat as you'd talk to ChatGPT. I'm not going to post the URL here because I'm afraid it will get crushed. You'll have to look for it if you care enough. I'm also attaching a few funny conversations I had with the model earlier into the image, just to give a sense. Next up, I am going to do one pass of tuning and optimizing the training throughput, then maybe return back to scaling and maybe training the next tier of a bigger model.

This software engineer was nearly hacked by a coding interview. I've never seen this attack channel before. Be careful out there, especially vibe coders.

@taviso Thanks for all the contributions on security!! 🙏

Build AI agents with scalable long-term memory! Mem0 is a fully open-source memory layer that gives AI agents persistent, contextual memory across sessions. Without memory, agents reset after every interaction, losing context and continuity, which limits their usefulness in real-world applications. Mem0 solves this by storing and managing long-term context efficiently, ensuring agents can learn and evolve with every conversation. Key Features: • Memory extraction – Uses LLMs to identify and store key facts from conversations while preserving full context • Memory filtering & decay – Prevents memory bloat by removing or aging irrelevant data over time • Hybrid storage – Combines vector search for semantic lookup with graph-based relationships for structured context • Smart retrieval – Prioritizes memories based on relevance, recency, and importance • Efficient context handling – Reduces token usage by sending only the most relevant memory snippets back to the model It’s 100% open source.

@LifestyleLatino Historia muy larga que al hacer scroll te muestra cientos de imágenes publicitarias. Historias más largas, más publicidad, más dinero.

@cesarolse @sgcol @SismoDetector Buscaba esto , si el centro es el punto amarillo entonces tembló hasta Merida 🫥

susana chuecos @chuecosusana

3 years ago

Temblor fuerte. Ahora mismo. 3:03 am. #MERIDA 🇻🇪

1 5 3 335 0

Ekoparty CTF is now on CTF Time ctftime.org/event/1790! You are invited to play at this year edition from anywhere, learning and fun guaranteed! * Prizes only available for local teams ctf.ekoparty.org #eko2022 #ekoctf @ekoparty

@kmcquade3 security through obscurity is bad