kkanishᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @universalanish

تريلر جديد لمسلسل نتفليكس الرعب والغموض Below عن صياد يعيش في بلدة نائية تبدأ حياتها بالانقلاب بعد ظهور وحش بحري غامض يهاجم السكان.. الأجواء رهيبة المسلسل من 6 حلقات وقادم في 8 أكتوبر

INSTEAD OF WATCHING NETFLIX TONIGHT. Spend 2 hours with this. Claude AI FULL COURSE that teaches you how to BUILD and AUTOMATE anything. The people who watch this tonight will wake up tomorrow with a new skill. Watch it and bookmark it now.

Just choose one and enjoy the show…

🚨 BREAKING: Jain Temple Was the Primary Target The NIA chargesheet reveals that Jain Temple Lal Mandir in old delhi was Umar Un Nabi’s primary target, not the Red Fort. Traffic and heavy police presence foiled his attempt to park the explosives-laden car near the temple, forcing him to shift his target to the Red Fort. The chargesheet further reveals that Jasir Bilal Wani used ChatGPT and YouTube to research IED and rocket construction, while Flipkart was used to procure components for explosive trigger mechanisms. Wani was the “in-house engineer” of the revived AGuH module, testing IEDs in Jammu & Kashmir and developing the technical capabilities used in the wider terror conspiracy.

🚨 US top companies hiring in India. Companies: Meta, Amazon, Apple, Microsoft, Netflix, and Google. Headcount of these companies in India: 2020: 104,000 2021: 110,000 2022: 147,000 2023: 157,000 2024: 180,000 2025: 212,000 Almost 100% increase in the last 5 years.

Kristen Habacht @kristenhabacht

8 hours ago

AI broke hiring. Candidates use AI to mass apply, recruiters use AI to mass deny. Introducing Elly: the first AI Recruiter as a Service.

98 68 336 441K 174

𝗛𝘆𝗱𝗲𝗿𝗮𝗯𝗮𝗱 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗮 𝗠𝗼𝗿𝗲 𝗪𝗼𝗿𝗸𝗲𝗿-𝗙𝗿𝗶𝗲𝗻𝗱𝗹𝘆 𝗖𝗶𝘁𝘆 🛵💙 Hyderabad’s growth story isn’t just about new investments and technology. It’s also about the people who keep the city moving every day. The Telangana government has signed an MoU with Zepto to create dedicated resting spaces for gig workers across Hyderabad, offering delivery partners a place to take a break, access basic facilities and recharge between deliveries. For thousands spending long hours on the city’s roads, these spaces can make a real difference to their everyday working conditions. #Hyderabad #Telangana #GigWorkers #DigitalEconomy #InclusiveGrowth #UrbanDevelopment

Carrie takes place in the same universe as IT: Welcome to Derry.

𝐈𝐧𝐝𝐢𝐚’𝐬 𝐧𝐞𝐱𝐭 𝐞𝐜𝐨𝐧𝐨𝐦𝐢𝐜 𝐜𝐡𝐚𝐩𝐭𝐞𝐫 𝐢𝐬 𝐛𝐞𝐢𝐧𝐠 𝐛𝐮𝐢𝐥𝐭 𝐭𝐨𝐝𝐚𝐲, 𝐚𝐧𝐝 𝐉𝐞𝐟𝐟𝐞𝐫𝐢𝐞𝐬 𝐬𝐞𝐞𝐬 𝐚 𝐧𝐞𝐰 𝐢𝐧𝐝𝐮𝐬𝐭𝐫𝐢𝐚𝐥 𝐫𝐞𝐯𝐨𝐥𝐮𝐭𝐢𝐨𝐧 𝐭𝐚𝐤𝐢𝐧𝐠 𝐬𝐡𝐚𝐩𝐞. 📌 Space: India’s space economy is targeted to grow 5x to US$40–45 billion by 2030, with private players driving innovation. 📌 Semiconductors: US$20 billion has already been invested, with a new US$13 billion incentive plan strengthening the chip ecosystem. 📌 Data Centres: Capacity could grow 5x to 10GW, driven by cloud adoption, digitisation and data localisation. 📌 Electronics: India is moving from assembly to greater value addition, targeting 50% of mobile component value. 📌 Solar: Already the world’s second-largest solar PV manufacturer, India is targeting 90% localisation of the solar manufacturing value chain by 2030. 📌 Aerospace: Indian companies are increasingly supplying global aerospace giants, with exports already around US$2 billion. 𝐉𝐞𝐟𝐟𝐞𝐫𝐢𝐞𝐬’ “𝐈𝐧𝐝𝐢𝐚’𝐬 𝐍𝐞𝐰 𝐈𝐧𝐝𝐮𝐬𝐭𝐫𝐢𝐚𝐥 𝐑𝐞𝐯𝐨𝐥𝐮𝐭𝐢𝐨𝐧” 𝐫𝐞𝐩𝐨𝐫𝐭 𝐡𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬 𝐭𝐡𝐞 𝐬𝐜𝐚𝐥𝐞 𝐨𝐟 𝐨𝐩𝐩𝐨𝐫𝐭𝐮𝐧𝐢𝐭𝐲 𝐚𝐜𝐫𝐨𝐬𝐬 𝐭𝐡𝐞𝐬𝐞 𝐞𝐦𝐞𝐫𝐠𝐢𝐧𝐠 𝐬𝐞𝐜𝐭𝐨𝐫𝐬. ⬇️

🚨 RUMORAZO 🚨 Grant Gustin sería el BARRY ALLEN aka FLASH del DCU, y hasta podría tener una aparición en 'THE PEOPLE VS GRODD' que actualmente se encuentra en pleno rodaje. (Vía: @bestever23)

keyboard_arrow_left Previous keyboard_arrow_right Next

This will never go viral on Instagram Elderly couple in Bengal lived without electricity after ~80 yrs of independence Now, new BJP govt has given her an electricity connection See her smile Dear Mahua Moitra, This couple is not fascist. They are victims of TMC's liberalism.

Harness Engineering is the part of agent building that starts to matter once the demo actually has to survive real work. A strong model can write good code and reason through difficult problems, but that does not automatically give you a reliable coding agent. The moment it has to navigate a repository, choose between tools, run commands, inspect what happened, remember progress, respect architectural boundaries, and continue after something goes wrong, most of the interesting problems move outside the model. That surrounding system is the harness. One useful framing comes from control systems. Before an agent acts, you can shape its behavior with task specifications, repository instructions, relevant context, examples, tool definitions, permissions, and architectural guidance. After it acts, you need signals from the environment: compiler errors, tests, type checks, linters, logs, screenshots, traces, runtime metrics, code review. Those signals give the agent a chance to notice that the thing it just did was wrong and try something else. This is where the difference between guidance and enforcement matters. Writing "do not create dependencies from this layer into that layer" in AGENTS[.]md is useful. A structural test that fails when that dependency appears is much stronger. OpenAI describes doing this while building with Codex, supplementing documentation with custom linters and structural tests for important architectural rules. They also exposed browser state, screenshots, application logs, metrics, and traces so Codex could inspect the consequences of its own changes. The environment became easier for the agent to reason about. OpenAI found something else that is easy to miss. More instructions did not keep making the system better. Their AGENTS[.]md stayed relatively small and acted as a map into structured repository documentation. Architecture, plans, product information, design decisions, and technical debt could live in versioned files and be pulled into context when useful. That is a much better model than stuffing every possible instruction into the context window and hoping the model pays attention to all of it. Long-running agents make the state problem even clearer. Anthropic found that context compaction by itself was not enough for agents working across multiple context windows. Their solution included an initialization script, a progress file, Git commits, structured task state, and explicit handoffs between sessions. A new agent session could inspect those artifacts and reconstruct where the work had reached. The important part is not the specific files they used. It is that project state had moved outside the conversation. Mitchell Hashimoto arrived at harness engineering from a much more practical direction. In his own coding-agent workflow, when an agent made the same kind of mistake repeatedly, he would spend time changing the environment so that mistake became less likely. Sometimes that meant adding guidance to AGENTS[.]md. Sometimes it meant adding tooling that allowed the agent to detect the problem itself. I like this framing because it turns repeated prompting into an engineering smell. If you keep telling the agent the same thing, some of that knowledge probably belongs in the harness. The same reasoning shows up everywhere. An agent repeatedly sending bad arguments to a tool may indicate a poor tool contract. Losing progress during a long task suggests useful state is living in the wrong place. Declaring a feature complete before it works usually means the completion signal is too weak. An agent that cannot diagnose a UI failure may simply lack access to the browser state, screenshots, or logs that a human engineer would inspect. Broad permissions are better handled through scoped access, sandboxing, or explicit approval than through another paragraph in the system prompt. There is a trap here too. A harness should not become an ever-growing pile of compensations for yesterday’s models. Anthropic has written about removing scaffolding when newer models no longer needed it. Every extra workflow, instruction, checkpoint, reviewer, and constraint adds complexity of its own. So the useful question is not 'how much harness can we add?" It is whether the harness is making the right state discoverable, the important rules enforceable, progress durable, risky actions bounded, and failures visible enough to correct. A newer model can obviously help. But quite a lot can change even when the model does not. Change the environment around it, and the same underlying model can behave very differently.

🚨🔥💥 𝗘𝗨𝗥𝗢𝗣𝗘 𝗕𝗘𝗚𝗚𝗘𝗗 𝗠𝗢𝗗𝗜 𝗙𝗢𝗥 𝗗𝗜𝗘𝗦𝗘𝗟! 🔥💥🚨 They flew to Delhi in expensive suits. Gave 𝗜𝗻𝗱𝗶𝗮 a moral lecture: Don’t buy 𝗥𝘂𝘀𝘀𝗶𝗮𝗻 oil. You’re funding the war. Then their traders called 𝗚𝘂𝗷𝗮𝗿𝗮𝘁. 𝗣𝗹𝗲𝗮𝘀𝗲 send diesel. Even 𝗨𝗸𝗿𝗮𝗶𝗻𝗲 took Indian diesel. Claim on the table: 𝟲𝟬–𝟳𝟬% of Europe’s diesel now comes from 𝗜𝗻𝗱𝗶𝗮𝗻 𝗿𝗲𝗳𝗶𝗻𝗲𝗿𝗶𝗲𝘀. Not Texas. Not Qatar. 𝗝𝗮𝗺𝗻𝗮𝗴𝗮𝗿. They killed their own refineries. Banned the cheap Russian barrel. Then bought the same molecules back — 𝗰𝗿𝗮𝗰𝗸𝗲𝗱, 𝗰𝗹𝗲𝗮𝗻𝗲𝗱, 𝗜𝗻𝗱𝗶𝗮𝗻. 𝗝𝗮𝗶𝘀𝗵𝗮𝗻𝗸𝗮𝗿’s line still stands: Substantial transformation. It is no longer Russian crude. It is an 𝗜𝗻𝗱𝗶𝗮𝗻 𝗽𝗿𝗼𝗱𝘂𝗰𝘁. They wanted to teach the Global South a lesson. India sent a 𝘀𝟯𝟬-billion masterclass instead. ⚡🇮🇳 𝗝𝗮𝗶 𝗛𝗶𝗻𝗱 🇮🇳✊

🚨 Airtel has partnered with Duolingo, the world’s leading learning platform, to offer 12 months of Super Duolingo for free to all its 376 million subscribers.

Naruto returns to the past as Hokage to protect his mother and father. 🔥❤️

As an AI Engineer, you must build these projects. Systems that prove you can retrieve, reason & ship. 1.) Production RAG Pipeline Build: Ingestion → chunking → hybrid search → reranking → cited answers over 1000+ docs. Why: Retrieval quality decides if your LLM product lives or dies. 2.) Structured Output Engine Build: Pydantic schema enforcement with retries, validation, graceful fallbacks on malformed JSON. Why: Apps break on free text. Schemas make LLM output usable. 3.) Context Assembly Service Build: Dynamic context builder that budgets tokens across memory, docs & tools per request. Why: Context engineering beats prompt tweaking in production. 4.) LLM Evaluation Harness Build: Golden dataset + LLM-as-a-judge + CI gate that blocks merges on quality regression. Why: You cannot ship what you cannot measure. 5.) Semantic Cache Layer Build: Embedding-based cache returning stored answers for similar queries, with hit-rate tracking. Why: The cheapest token is the one you never generate. 6.) Model Routing Gateway Build: Complexity-based router across 3 model tiers with fallbacks and per-request cost tracking. Why: Unit economics decide if your AI product survives. 7.) Multi-Tenant LLM API Build: Per-tenant keys, rate limits, token budgets, isolated data access. Why: Enterprise customers demand isolation and spend control. 8.) Fine-Tuning Pipeline Build: LoRA/QLoRA fine-tune of a small model on synthetic data, benchmarked vs prompt-only baseline. Why: Know when adaptation beats prompting, and when it doesn't. 9.) Agent Memory System Build: Working + episodic + semantic memory with compression and eviction policies. Why: Agents without memory fail by turn five. 10.) Guardrails Middleware Build: Input/output filtering, PII redaction, injection detection as a reusable service layer. Why: One leak destroys trust. Safety must be infrastructure. 11.) Streaming Response Infrastructure Build: SSE streaming with backpressure, reconnects, time-to-first-token tracking. Why: Perceived latency is the UX of AI products. 12.) Prompt Versioning and A/B System Build: Registry for prompts and configs with rollback, traffic splitting, outcome tracking. Why: Prompts are production code. Treat them that way. 13.) LLM Observability Stack Build: Tracing for prompts, tokens, latency, cost per request with alerting on anomalies. Why: You cannot debug what you cannot see, especially non-determinism. 14.) Tool-Calling Framework Build: Typed function schemas with discovery, retries, sandboxed execution. Why: Tools turn chatbots into systems that actually do work. 15.) Self-Correcting RAG Agent Build: Query rewriting, retrieval critique, confidence-based fallback to web search. Why: Production retrieval must repair itself, not hallucinate. Most people watch tutorials. Builders ship systems. Bookmark & Repost.

Suraj Sharma @suraj_sharma14

a day ago

If I had 6 months to become an AI Engineer (Applied LLM). I'd do this. Stage 1: Python and LLM Foundations Python, async, APIs, tokens, embeddings, context windows, Anthropic/OpenAI SDKs, reading model docs like an engineer. Stage 2: Prompting and Context Engineering System

26 112 794 132K 2K

Matt Reeves has reupped his overall deal with Warner Bros. TV to produce more projects for the studio “Matt has a true knack for getting into the psyche of his characters and he and his team have delivered some of the most compelling, thought-provoking stories over the years”, says CEO of WB TV Channing Dungey (via @Variety)

keyboard_arrow_left Previous keyboard_arrow_right Next

🚨 SHOCKING SCENE IN ARGENTINA 🌊 - The sea reportedly moved 23 KM away within an hour, leaving boats stranded and fish dead. What caused this? 👀

🚨 BREAKING: Tensions rise in Meerut’s Kesarganj after residents and Mandi Samiti members OPPOSED the construction of a large MOSQUE in the Hindu-majority area. Protesters alleged that Friday prayers were causing problems for local women and girls and urged the administration to STOP illegal construction and take action. x.com/SudarshanNewsT…

FOODS THAT DETOX YOUR BODY: 1. Lemon water – 1 glass daily 2. Beets – 1/2 cup daily 3. Green tea – 2 cups daily 4. Garlic – 2 cloves daily 5. Leafy greens – 2 cups daily 6. Turmeric – 1 tsp daily 7. Apple cider vinegar – 1 tbsp daily

@tarunvats33 @mahendrantheiva Offline store said 1,34,000+17000(apple care) Is it fair price?