Ksenia_TuringPost

OpenAI Agents SDK – an open orchestration layer for building multi-agent workflows It lets you define agents as LLMs with instructions, tools (APIs, functions, external systems), guardrails, and supports: • sessions with conversation history management • human-in-the-loop • tracing to monitor and debug how agents make decisions • voice agents An interesting feature is Sandbox Agents, which operate in a controlled environment with access to files, code repos and terminal commands. You can plug in 100+ models including open ones (but you may need an adapter layer)

To really understand embeddings, you need a few core ideas: - vectors and dimensions - dense vs sparse representations - vector and embedding spaces - what latent space means - semantic similarity importance - and how embeddings are formed These concepts completely change and explain how meaning is represented in AI. Full guide (including RoPE): turingpost.com/p/embeddings

9 New approaches to Multi-Agent Systems ▪️ RecursiveMAS ▪️ OneManCompany (OMC) ▪️ OrgAgent ▪️ CORAL ▪️ LLMA-Mem ▪️ Agentic Federated Learning ▪️ CASCADE ▪️ GRASP ▪️ Reinforced Agent These methods express truly interesting various ideas! Learn more about them here: turingpost.com/p/9masmethods

And he is here! Alpha called him Johnny Cash 🧑🏻‍🍼 #ReachyMini

Agents are already moving at machine speed, but security is still stuck on static, outdated rules. We can close the gap On May 5, @rubrikInc is hosting a technical webinar on AI security at scale – Building AI Resilience: Managing Agent Risk with Trust Infrastructure → bit.ly/4cZcbI3 They’ll cover: • Where security actually breaks: algorithms + human factors • 3 pillars of trustworthy AI ops • How trust infrastructure works in practice (Rubrik Agent Cloud) • A real financial services case where static rules fail Register now: bit.ly/4cZcbI3

AI is our chance to finally understand how students think, where they get stuck and how to help them learn Neeru Khosla, co-founder of @CK12Foundation, in our interview Watch the full video to find out how AI can change education ->

@ToxinHex @deepseek_ai here you are :) alphaxiv.org/overview/visua…

@TheTuringPost @deepseek_ai Ohh Christ were the fuck is the link man~ thx for the information but at least drop the link to the paperwork ☹️😔

There’s a serious gap in multimodal models – they work with images, but still reason in language, which isn’t that precise for visual stuff. @deepseek_ai just dropped an idea to solve this: let the model literally point to exact locations in the image while it thinks. They call it "Thinking with Visual Primitives." These visual primitives are: - points (specific locations) - bounding boxes (areas in the image) Using them, the model knows what exactly it’s referring to and achieves ~77% better accuracy on average (vs. Gemini 3 Flash's 76.5% and 71.1% for GPT-5.4) Plus, only ~80–90 visual tokens are kept in memory after compression thanks to the efficient architecture Here is how it works:

Follow @TheTuringPost for more. Get deep analysis, guides & breakdowns of what AI is about now. Join 100,000+ readers from top AI labs, VC funds & universities.: turingpost.com/subscribe

5. By combining pointing + reasoning, the model improves both quality and efficiency: ~77% accuracy on average vs. Gemini 3 Flash's 76.5% and 71.1% for GPT-5.4 ~289 tokens per image (~2–4× fewer than other models) fed into the model ~80–90 tokens kept in memory after compression thanks to a more efficient architecture. So this is all about better reasoning with less visual data, because the model knows what exactly it’s referring to.

4. Training was also an important part They trained two separate specialists (boxes + points) → refined them with reinforcement learning → merged both experts into one model + distilled their behavior into it.