Anas

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

@claudeai - github.com/anthropics/cla… Open Web UI - github.com/open-webui/ope… @deepseek_ai - github.com/deepseek-ai/De… @GeminiApp - github.com/google-gemini/… Rag Flow- github.com/infiniflow/rag…

Top AI Repositories of 2026 @openclaw - github.com/openclaw/openc… @opencode - github.com/anomalyco/open… @n8n_io - github.com/n8n-io/n8n @ollama - github.com/ollama/ollama @LangChain - github.com/langchain-ai/l… Langflow - github.com/langflow-ai/la… Dify - github.com/langgenius/dify More...

@buildwithshyam Ctrl+L

What command do you use most in the terminal?

@nikhilkamathcio Hey nikhil, recently i had a confusion related to T&C of DPPI facility which is now clear, But i have always seen you and admire you in a stock market so i just want to ask is the DPPI facility safer & better than TPIN? Is it same across all brokers?

@DhanHQ Dear Team, I would like to enable DPPI facility for my account but i dont want to agree to all the T&C as i dont invest in MF, i dont use leverage so i dont be needing the pledge and repledge facility. If i only agrees to 1st T&C in the document will i be able to use it?

@sarlloc I have never seen his portfolio, neither GitHub nor anything which made me listen to him

"India's dev community is shit" says the guy who uses half naked women to promote his "tech content". Blud cant use git cli to save his life, he is one of the worst creators in the tech domain

@mahabank Today i sent an email regarding my issue with KYC to the email mentioned on your webpage and after sending the email i recieved a notification saying that domains have been migrated. Kindly update the communication data on your website.

Which Linux distro do you use? I use BTW #Linux

I have decided to post articles on @X But for that I need to buy premium. Is it worth it guys be honest?? and if yes then lets Connect. I will mainly write around tech and sometimes post some random comments/posts. #Connect

@OpenAI @sama I just got the email saying that my account is deactivated and even i have not used it that much and created my account 4 to 5 weeks ago. Since i am an Indian i have also subscribed to the Go plans you offer for free.

@Windows What the hell did you do with the charging indicator 😭😭

@kunalstwt I have also seen your rolex too you have a great collection to be honest ❤️ one day i will become capable to buy those also

These two watches ended up getting the most wrist time for me last year, and if I had to recommend just one watch to buy at retail and keep for life, this would be it. I’ve made a lot of great friends in tech who are also watch enthusiasts, would love to connect with more of you.

The big challenges? Token costs, noisy retrieval, and avoiding contradictions over time. Smart context engineering often beats heavy RAG setups for speed and reliability. Overall, well-designed memory turns basic chatbots into adaptive agents. 5/5

Implementation wise, you can store memories as simple text strings, structured JSON for easy updates, or even knowledge graphs for complex relationships though graphs add overhead. Tools like Mem0 make it straightforward to add and search these. 4/5

Building AI agents that feel truly smart isn't just about bigger models it's about giving them memory so they can remember you, learn from past chats, & act consistently over time. Without it, every interaction starts from scratch, like talking to someone with total amnesia. 1/5

So which part of that pipeline feels the most surprising, or the most broken, to you? And Sorry for All this long thread actually i dont have premium but anyways i wanted to write and share my learning here Thanks a lot❤️

In the end, modern frontier LLMs are simply the result of huge data, enormous compute, and smart engineering stacked on top of next-token prediction. There’s no hidden magic - just scale, iteration, and a lot of tuning. 10/10

Large language models don’t really “learn” the way humans do. They’re trained on absurd amounts of internet text with one deceptively simple goal: predict the next word (or token) in a sentence. 1/10