2022: the ChatGPT shock Generative AI LLM Hallucination Chatbot Prompt engineering Diffusion / Stable Diffusion / Midjourney RLHF 2023: the tooling and fear year GPT-4 Multimodal Foundation model Fine-tuning Open-source LLMs (Llama) Plugins and Custom GPTs Copilot Embeddings and vector database RAG Context window Jailbreak AI alignment / AI safety / AGI Deepfake 2024: from chat to reasoning RAG (peak hype) Function calling / tool use Small language models Long context (1M tokens) Text-to-video (Sora) Chain-of-thought Reasoning models (o1) Test-time compute Synthetic data AI slop Computer use MCP (Model Context Protocol, launched in November) Evals 2025: the agent year Agentic AI / AI agents MCP going mainstream Vibe coding DeepSeek moment and distillation Context engineering Deep research Agentic coding (Claude Code and similar) Sub-agents and multi-agent systems Agent skills A2A Agent memory Sycophancy Spec-driven development World models AI bubble 2026: harness and beyond Agent harness / harness engineering System One models / Jev the buzzwords moved from the model itself (LLM, hallucination), to how you talk to it (prompt engineering), to what you feed it (RAG, context engineering), to what it does (agents, MCP), to the system around it (harness). Jev now questions whether every decision needs a chat-style LLM at all.

This 15-sec add was built on GPT 6

Claude ChatGPT Grok Gemini Mistral Kimi Deepseek Qwen and more CAN BUILD APPS CAN ANYONE NAME A APPLICATION That is build completely and maintained by these AI tools Used by consumers/Business user worldwide Like Uber, Ola, Swiggy, Zomato, Skyscanner, Salesforce, SAP, TALLY etc We are still at the VIBECODING ERA !!!!!!

Why does the major AI tools don't show the token usage and its limit Chat products sell a flat subscription that is supposed to feel like just talk. Sell an experience, meter compute in the background, reveal the constraint only when it bites. Is there a strategy behind Just like no clocks placed in Restobars Just like milk placed at last shelves in supermarkets __________________________________________ Knowledge share: use OPEN AI Tokenizer tool below to understand how a piece of text might be tokenized by a language model platform.openai.com/tokenize…

Three LLMs, three different games I asked this to most of the AI MODELS "I use multi AI tools for my day to day work & planning to purchase a monthly subscription help with the 3 to 4 tools that you wish I would subcribe to" The response shook me away Grok Says : go with Claude-ChatGPT-Perplexity-Google gemini Claude - says in the same order ChatGPT- asked me to choose ChatGPT Plus followed by the same order None of the AI models suggested Grok or any Chinese AI models We are now at a track of SI- super intelligence

THIS SHOOK ME AROUND I asked this to most of the AI MODELS "I use multi AI tools for my day to day work & planning to purchase a monthly subscription help with the 3 to 4 tools that you wish I would subcribe to" The response shook me away Grok Says : go with Claude-ChatGPT-Perplexity-Google gemini Claude - says in the same order ChatGPT- asked me to choose ChatGPT Plus None of the AI models suggested Grok or any Chinese AI models

A family of GPT models by OPENAI - that has turned our lives smarter, brighter and worthier GPT models of ChatGPT GPT‑3.5 - Powered ChatGPT’s original November 2022 launch. GPT‑4 family - Included GPT‑4, GPT‑4 Turbo, and GPT‑4 with vision. GPT‑4o family - GPT‑4o and GPT‑4o mini; later specialized audio and realtime versions also appeared. o-series reasoning models - o1, o1 mini, o1 pro, o3 mini, o3, o3 pro, and o4 mini. GPT‑4.1 family - GPT‑4.1, mini, and nano. These models appeared in ChatGPT and the API GPT‑4.5 - A research-preview model GPT‑5 family - GPT‑5, mini, nano, and Pro, followed by GPT‑5.1, GPT‑5.2, GPT‑5.3, GPT‑5.4, GPT‑5.5, and GPT‑5.6 variants. GPT‑6 Astra - Introduced in ChatGPT in September 2026 with limited rollout at launch.

If my net salary is I lakhs rupees per month, can I buy a 15 lakhs worth SUV car Give me an answer from the investment perspective angle

Here is the list of terms and phrases that have been significantly declining in usage since ChatGPT launched in late 2022: Google fu: The skill of using advanced search operators and keywords to find obscure information on the web. LMGTFY (Let Me Google That For You): A sarcastic acronym and link shared when someone asks a question they could have easily searched themselves. Spinning content / Article spinner: Automated software used by SEO writers to rewrite articles by replacing words with synonyms to avoid plagiarism detectors. Template-driven copywriting: Pre-written, fill-in-the-blank text frameworks used for basic marketing emails and sales landing pages. "Google it": The universal, default advice given to someone looking for a direct explanation or step-by-step instructions. Stock response: Generic, pre-written customer service scripts used to reply to common user inquiries. Lorem Ipsum: Traditional placeholder text used by designers to demonstrate the visual form of a document without relying on meaningful content.

Do we have a competitor to this business idea

The era of writing 2,005 lines of Python to spin up a custom AI agent is officially over. You no longer need a CS degree or a complex codebase to build autonomous systems. Visual drag-and-drop frameworks and zero-code builders are letting anyone prototype end-to-end workflows in under an hour. The barrier to entry has completely collapsed. The only limit left is your imagination. Amplify rapid prototyping for everyday operational tasks. Remove traditional coding bottlenecks from your product cycles. Reframe AI agents as configuration tasks rather than engineering projects This 10 min video focuses on building autonomous AI agents without writing traditional code, msn.com/en-in/technology/art…

Follow this 9 steps to make your CLAUDE better Learn more here - how-to-ai.guide

An AI Engineer & Applied Scientist has built a research agent CLI Its called jevgrep is a CLI tool that helps coding agents quickly find and understand relevant code by asking natural language questions instead of manually searching through files. The tool is designed to help coding agents find code by asking what it does it discovers relevant files and source context. To install, just send claude/codex this exact repo: github.com/dzhng/jevgrep

Jev by @typesafeai is all around the X, reddit, Linkedin Etc,. It has 3 elements Choice- Select one option from a defined set Score- Rate the state against ordered levels Noul- Evaluate a yes/no statement whereas a typical LLM Tokenizer-> Splits input text into discrete pieces or tokens. Transformer Layers (Encoder/Decoder) -> Uses self-attention mechanisms to understand context and relationships between words. Autoregressive Generator -> Predicts and outputs the next sequential token based on prior context. Decoding/Sampling Parameters -> Controls creativity and randomness (like temperature and top-p) LLMs are designed to generate text, while Jev is designed to evaluate pre-declared, structured decisions LEARN MORE : huggingface.co/blog/sora-2/j…

This might be an AI video, but the fictional characters are doing their role exactly the same Watch the video till the end

Chinese researchers have published a paper called “The Last AI Built by Humans” An AI should not merely perform better on a task -it should identify its own weaknesses, make a validated and persistent improvement, and then use that improvement to make its future improvement process better. 1. L1: AI executes an improvement humans specify. 2. L2: AI chooses how to improve. 3. L3: AI creates its own future learning experiences. 4. L4: AI learns from real deployment environments. 5. L5: AI improves the system that generates and evaluates future improvements. The key distinction is L5 changing the “improvement engine” itself. The paper argues that most current systems automate only parts of this loop, so genuine end-to-end recursive self-improvement remains largely unproven. It is primarily a survey and roadmap, rather than a demonstration that fully autonomous recursive improvement already exists. arxiv.org/abs/2609.11873

This one is Built by Grok 4.7 @threejs Interactive Turbofan Engine Visualization

How the names changed, in order -Then: RLHF Reinforcement Learning from Human Feedback People pick which answer they like. The model is trained to sound like what people prefer. - Next: RLAIF Reinforcement Learning from AI Feedback Same idea as RLHF, but another AI does the judging instead of humans. Cheaper, faster. - Also around then: DPO Direct Preference Optimization Skip the complicated RL setup. Just train the model on “this answer is better than that one.” - Old RLCD (2023) Reinforcement Learning from Contrastive Distillation No human labels. Make two answers on purpose - one “good prompt,” one “bad prompt” -and treat that as preference data - Then: RLVR Reinforcement Learning with Verifiable Rewards No opinions. A program checks the answer: math is right, code passes tests, format is valid. Correct = reward. How they run RLVR: GRPO, then DAPO GRPO = compare several answers to the same question and keep the better ones. DAPO = a later fix so that training doesn’t collapse on long reasoning. - Now: RLCD (2026) Reinforcement Learning for Calibrated Decisions Not “people liked it” and not “the checker said yes.” Reward = the confidence number matches reality. If it says 80%, it should be right about 80% of the time. One-line map

🚀 Apple has a spaceship. 🏢 Microsoft has a campus. 💻 Linux just had code to finish. Real estate never built the internet. Builders did. Campus or kitchen table where would you rather ship?

Time to save your bill on photo-shop subscription I'm not just saying it just to grab your attention Really some one built an open source tool to get rid of Adobe photoshop If Adobe see this post in their timeline, this guy would be sued 😄😄😄