People want to understand agentic infrastructure protocols better. The strong response to our MCP article shows there’s real demand for clarity around standardization of AI ecosystems Since so many people asked, we are making our article on Agent2Agent (A2A) free to read on @hf.co 🧵

🦸🏻#17: What is A2A and why is it – still! – underappreciated?

A Blog post by Ksenia Se on Hugging Face

huggingface.co

The most important features of LFMs (Liquid Foundation Models) from Liquid AI? Memory-efficiency, inference speed, without compromising model quality. LFMs have been benchmarked on real hardware, proving that they can beat Transformers. Liquid AI have also just released Hyena Edge👇

What happens when the biggest advocate for coding literacy starts telling people not to learn to code? In the new Inference episode, I sat down with Amjad Masad, CEO and co-founder at Replit, to discuss the evolution in coding. Are we entering a post-coding world? www.youtube.com/watch?v=PlDe...

When Will We Stop Coding? A conversation with Amjad Masad, CEO and co-founder @ Replit

YouTube video by Turing Post

youtube.com

Top 7 research papers of the week: ▪️ Test-Time Reinforcement Learning ▪️ LLMs are Greedy Agents ▪️ Paper2Code ▪️ Efficient Pretraining Length Scaling ▪️ The Sparse Frontier ▪️ Roll the dice & look before you leap ▪️ Discovering and Analyzing Values in Real-World Language Model Interactions 🧵

9 notable AI models of the week: ▪️ Hyena Edge ▪️ Tina: Tiny Reasoning Models via LoRA ▪️ Kimi-Audio ▪️ Aimo-2 winning solution ▪️ Eagle 2.5 ▪️ Trillion-7B ▪️ Surya OCR ▪️ ThinkPRM ▪️ Skywork R1V2 🧵

Agent2Agent (A2A) is standardizing how multiple autonomous agents communicate seamlessly across applications. Yet even with the loud launch and 50 big-name partners, Google's A2A remains underappreciated. Why? Here are several reasons 👇 www.turingpost.com/p/a2a

ChatGPT shifted how we interact with computers. Real-time conversational AI could do something similar for how we interact with each other ElevenLabs is working to make such communication possible We talked to it's co-founder and CEO Mati Staniszewski about what the next few years might look like👇

When Will We Speak Without Language Barrier? A conversation with Mati Staniszewski, CEO @ ElevenLabs

YouTube video by Turing Post

youtube.com

2 perspectives on Human-AI co-agency ▪️ AI as a medium ▪️ AI as a feedback loop Details🧵

🦸🏻#16: Co-Agency as The Ultimate Extension of Human

how AI as a medium shapes our perception, behavior, and culture

turingpost.com

The freshest AI/ML research of the week My top 10: ▪️ AI for Software Engineering ▪️ Inference-Time Scaling for Reward Modeling ▪️ Inference-Time Scaling for Complex Tasks ▪️ Open-Reasoner-Zero ▪️ SynWorld ▪️ Agent S2 ▪️ KnowSelf ▪️ ZClip ▪️ MegaScale-Infer ▪️ Scaling Laws in Scientific Discovery 🧵

7 notable AI models of the week: • Command A • TransMamba • HallOumi • ScholarCopilot • OThink-MR1 • RIG • Z1 🧵

9 types of AI inference 5 types based on how the model reasons: ▪️ Probabilistic ▪️ Rule-based ▪️ Logical ▪️ Abductive ▪️ Fuzzy 4 inference types based on its execution contexts: ▪️ Batch ▪️ Real-time ▪️ Edge ▪️ Cloud Save the list and check this out for useful resources: huggingface.co/posts/Ksenia...

A free short AGI safety course from @GoogleDeepMind It covers important topics, such as: - Path to advanced AI capabilities - Sources of misalignment risk - Technical approaches: safe design patterns, oversight, robust training, etc. - AI governance - Testing and evaluating AI failures Watch 👇

Google DeepMind AGI Safety Course - YouTube

A short course from Google DeepMind on AGI safety, covering alignment problems we can expect as AI capabilities advance, and our current approach to these pr...

youtube.com

Key concepts of AI inference: - Inference time - Total generation time - Latency - Time to First Token (TTFT) - Time Per Output Token (TPOT) - Throughput - Cost per inference - Scalability - Accuracy - Entropy Learn what they mean and other main things about inference in our article:

Topic 34: Things You Need to Know About Inference

we dive into the core aspects of AI inference, explore how it works and what can make it faster

turingpost.com

3 studies from @anthropicai.bsky.social that form a "guide" to the LM's mind: ▪️"Dictionary learning" ▪️ Monosemanticity ▪️ "On the Biology of a Large Language Model" 🧵

4 notable AI models of the week: • Google's Gemini 2.5 • DeepSeek-V3-0324 • Zhipu's AutoGLM Rumination • Alibaba's Qwen2.5-Omni 🧵

9 Multimodal Chain-of-Thought methods ▪️ KAM-CoT ▪️ Multimodal Visualization-of-Thought (MVoT) ▪️ Compositional CoT ▪️ URSA ▪️ MM-Verify ▪️ Duty-Distinct CoT ▪️ Multimodal-CoT ▪️ Graph-of-Thought ▪️ Hypergraph-of-Thought Save the list, and check this out for more info: huggingface.co/posts/Ksenia...

Humans as tools? Is it real? Yes, sometimes human is just another callable function in an AI agent's toolbox. ▪️ Human in the loop (HITL) is a design pattern, where humans are built into the decision loop to: - validate outputs - steer actions - override the machine when necessary 🧵

What is Slim Attention? It's a new attention mechanism that allows models to be 2x faster and cut memory use by 32 times! ▪️ What's the secret? It uses the same math as Multi-Head Attention (MHA) but applies one clever trick🧵

Because of DeepSeek, Qwen models are a little overlooked but they are worth your attention. It's very interesting how new Qwen2.5-Omni truly expands multimodality to a full pack — text, images, audio, and video. Its main focus is speech generation, and it uses a treasure trove of power features 👇

Do LLMs "know" more facts deep inside their systems than they actually "say"? Researchers from Technion and Google Research defined what this "knowing" means and also explored it and found that: LLMs often know more than they say but getting them to "say" it can be surprisingly hard Key findings🧵

The freshest AI/ML research of the week: Our top 2 ▪️ Xattention ▪️ Inside-Out: Hidden Factual Knowledge in LLMs ▪️ Rwkv-7 "Goose" ▪️ ϕ-Decoding ▪️ Frac-connections ▪️ DAPO ▪️ Reinforcement learning for reasoning in small LLMs ▪️ MetaLadder ▪️ Why do multi-agent LLM systems fail? 🧵

7 open-source AI models of the week: • KBLaM from @microsoft.com • Fin-R1 • NVIDIA’s Cosmos-Reason1 • NVIDIA’s Cosmos-Transfer1 • M3 by NVIDIA • Tencent’s T1 • Roblox’ Cube 🧵

The interview with SharonZhou, Co-founder & CEO of Lamini, and co-creator of a top AI course on Coursera, was full of incredible insights! We discussed important topics, such as: - AI hallucinations - Agents and RAG hype - The keys to GenAI - AI education www.youtube.com/watch?v=RW01...

Turing Post / Inference. Ep 001: Sharon Zhou, Lamini (and so much more)

YouTube video by Turing Post

youtube.com

8 types of RoPE (Rotary Position Embedding): ▪️ Original RoPE ▪️ LongRoPE ▪️ LongRoPE2 ▪️ MRoPE (Multimodal RoPE) ▪️ DRoPE (Directional RoPE) ▪️ VideoRoPE ▪️ VRoPE ▪️ XPos Save the list and check this out for the links and more info: huggingface.co/posts/Ksenia...

What is Qwen-Agent framework? It's a full‑fledged agentic ecosystem that lets Qwen models autonomously plan, call functions, and execute complex, multi‑step tasks right out of the box. Here are examples of Qwen-Agent's applications:

As I see, image generation is evolving with high speed, but we need tech advancements which improve quality without complexity. In a new study, @drfeifei.bsky.social along with other researchers introduced FlowMo, which rethinks how tokenizers are built. Here's what makes it different:

MCP by @anthropic.com vs OpenAI Agents Platform Two giants, OpenAI and AnthropicAI, are accelerating and taking very different paths to agentic AI ▪️Anthropic is focused on standardization with MCP. This open approach creates a universal standard for connecting AI models to external tools. ...

A new RL algorithm! DAPO (Decoupled Clip and Dynamic Sampling Policy Optimization) from ByteDance is a fully open-source RL system, that improves training in long CoT. It achieves 50 points on AIME 2024, surpassing DeepSeek-R1-Zero, using only half the training steps. What are its key features?

What is Model Context Protocol (MCP), and how do you get started with it? MCP from @anthropicai.bsky.social is an open standard designed to bridge AI assistants with the data and tools​, and to plug in many different sources of context. MCP is seen as the missing puzzle piece for AI agents 👇

How does Multi-Head Latent Attention (MLA) reduce memory use? MLA is like zipping and unzipping stored data to save memory. It compresses the key-value (KV) cache into a much smaller form using low-rank key-value joint compression. Here's how MLA works: