Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Today we’re launching INTELLECT-2: The first decentralized 32B-parameter RL training run open to join for anyone with compute — fully permissionless. Scaling towards frontier reasoning across coding, math and science.

Prime Intellect

71,062 subscribers

353,791 views • 1 year ago •via X (Twitter)

Gaming Science & Technology Education

Anya Rossi• Live Now

Private livecam show

11 Comments

Prime Intellect1 year ago

INTELLECT-2 brings decentralized training into the inference-time compute era: • Fully async, decentralized reinforcement learning • Eliminating communication overhead • Scalable across heterogeneous GPUs worldwide

Prime Intellect1 year ago

Over the past months, we’ve built the full open-source stack to enable INTELLECT-2: • PRIME-RL: fully async decentralized RL • GENESYS & SYNTHETIC-1: crowdsourced tasks & verifiers for RL • TOPLOC validation: verifiable inference with low overhead • Protocol Testnet: global AI coordination infrastructure

Prime Intellect1 year ago

How INTELLECT-2 operates: • Inference Rollout Workers: Decentralized swarm collects rollouts using the latest policy model • TOPLOC Validators: Verify the inference computations • GRPO Training Workers: Train on new data and broadcast weights via our shardcast library

Prime Intellect1 year ago

Asynchronous RL completely eliminates communication bottlenecks. Our ablation studies confirm we maintain performance even with 4-step delays, making decentralized training viable with weak global interconnects.

Prime Intellect1 year ago

With INTELLECT-2 we aim for frontier reasoning performance with a controllable thinking budget. By incorporating length rewards into our training run, users can specify how long the model should reason for a given task.

Prime Intellect1 year ago

Contribute compute and watch training live on our dashboard: • Participate in our testnet powering an open reasoning model • Slashing and validation keep contributions honest Join us in building towards open and decentralized AGI

Lab4crypto1 year ago

🚀 Don't gamble with your portfolio! Use our advanced hybrid quant risk tool using on/off-chain data and make informed decisions. 📈 Acess to 1000+ charts for your crypto journey. 📚Join our Premium Telegram for daily alerts. 📊+21 projects supported. 🏗️ Beginners and experts.

Michael Luo1 year ago

This is amazing work! Looks like asynchronous RL for LLMs is pretty stable 🤩, a paradigm shift away from traditional Deep RL where synchronous PPO won, as opposed to A3C and IMPALA

Tom Bennet1 year ago

Decentralized RL training? Sounds like we're breeding digital superminds in a worldwide compute farm. Count me in for the chaos.

jesse.base.eth1 year ago

ok this is really sick

Burny — Effective Omni1 year ago

Decentralized open source is climbing

Related Videos

Introducing SYNTHETIC-1: Collaboratively generating the largest synthetic dataset of verified reasoning traces for math, coding and science using DeepSeek-R1. Join us to contribute compute towards state-of-the-art open reasoning models.

Introducing SYNTHETIC-1: Collaboratively generating the largest synthetic dataset of verified reasoning traces for math, coding and science using DeepSeek-R1. Join us to contribute compute towards state-of-the-art open reasoning models.

Prime Intellect

198,473 views • 1 year ago

Launching SYNTHETIC-2: our next-gen open reasoning dataset and planetary-scale synthetic data generation run. Powered by our P2P inference stack and DeepSeek-R1-0528, it verifies traces for the hardest RL tasks. Contribute towards AGI via open, permissionless compute.

Launching SYNTHETIC-2: our next-gen open reasoning dataset and planetary-scale synthetic data generation run. Powered by our P2P inference stack and DeepSeek-R1-0528, it verifies traces for the hardest RL tasks. Contribute towards AGI via open, permissionless compute.

Prime Intellect

360,457 views • 1 year ago

Announcing INTELLECT-1: the first-ever decentralized training of a 10B model Scaling decentralized training 10x beyond prior efforts. Anyone can join us to build open-source AGI 🦋

Announcing INTELLECT-1: the first-ever decentralized training of a 10B model Scaling decentralized training 10x beyond prior efforts. Anyone can join us to build open-source AGI 🦋

Prime Intellect

790,053 views • 1 year ago

Introducing Cosmos 3: Our latest frontier model for Physical AI Cosmos 3 is the world’s first fully open omnimodel with native vision reasoning, world and action generation. Today we’re releasing Super (32B) and Nano (8B) variants.

Introducing Cosmos 3: Our latest frontier model for Physical AI Cosmos 3 is the world’s first fully open omnimodel with native vision reasoning, world and action generation. Today we’re releasing Super (32B) and Nano (8B) variants.

NVIDIA AI

423,642 views • 1 month ago

Introducing INTELLECT-3: Scaling RL to a 100B+ MoE model on our end-to-end stack Achieving state-of-the-art performance for its size across math, code and reasoning Built using the same tools we put in your hands, from environments & evals, RL frameworks, sandboxes & more

Introducing INTELLECT-3: Scaling RL to a 100B+ MoE model on our end-to-end stack Achieving state-of-the-art performance for its size across math, code and reasoning Built using the same tools we put in your hands, from environments & evals, RL frameworks, sandboxes & more

Prime Intellect

1,142,730 views • 7 months ago

$A crypto project actually trained a 72B parameter AI model from scratch using decentralized GPU compute. Not fine-tuned, not a wrapper: trained from zero. The model benchmarks competitively against Meta's LLaMA 3 on reasoning tasks, and the entire training run cost a fraction of what centralized labs spend. If decentralized compute can produce frontier-class models, the moat around OpenAI and Anthropic is thinner than people think.$

A crypto project actually trained a 72B parameter AI model from scratch using decentralized GPU compute. Not fine-tuned, not a wrapper: trained from zero. The model benchmarks competitively against Meta's LLaMA 3 on reasoning tasks, and the entire training run cost a fraction of what centralized labs spend. If decentralized compute can produce frontier-class models, the moat around OpenAI and Anthropic is thinner than people think.

VirtualBacon

40,543 views • 3 months ago

1/3 The wait is finally over! Secure. Permissionless. Compute. Deploy enterprise-grade AI applications, run private LLMs, and access dedicated compute with one click. Open-source, decentralized, and completely under your control. Welcome to CLI! ☁️🌐 App: >_ $CMD

1/3 The wait is finally over! Secure. Permissionless. Compute. Deploy enterprise-grade AI applications, run private LLMs, and access dedicated compute with one click. Open-source, decentralized, and completely under your control. Welcome to CLI! ☁️🌐 App: >_ $CMD

CLI

33,613 views • 1 year ago

Prime Intellect engineers explained how they train reasoning models over the open internet in 30 minutes - better than $3000 distributed training courses. split policy and rollouts across nodes -> run agents in parallel envs -> verify with LLM judges -> gradient-sync over the internet -> train Llama, Qwen, Gemma at cluster scale on rented GPUs. That loop is why open reasoning models are catching closed labs without owning a data center. Prime-RL + verifiers + distributed rollouts + LLM judges + multi-cloud GPUs - that's the stack. Watch and save it, then launch your first distributed RL run this week.

Prime Intellect engineers explained how they train reasoning models over the open internet in 30 minutes - better than $3000 distributed training courses. split policy and rollouts across nodes -> run agents in parallel envs -> verify with LLM judges -> gradient-sync over the internet -> train Llama, Qwen, Gemma at cluster scale on rented GPUs. That loop is why open reasoning models are catching closed labs without owning a data center. Prime-RL + verifiers + distributed rollouts + LLM judges + multi-cloud GPUs - that's the stack. Watch and save it, then launch your first distributed RL run this week.

h100envy

39,194 views • 7 days ago

After coding is solved, the next frontier is computer use. Today, we are launching Use Computer, the infra for evaluating and training models to use all kinds of computers 👇

After coding is solved, the next frontier is computer use. Today, we are launching Use Computer, the infra for evaluating and training models to use all kinds of computers 👇

Marco Mascorro

44,641 views • 1 month ago

We’re thrilled to announce that Gata Decentralized GPU Cloud is now LIVE! Start today → Built to accelerate the decentralization of the open AI economy, Gata enables anyone, anywhere to access decentralized GPU infrastructure, to build, train, and deploy AI models with full transparency and efficiency. GPU credits are now available for purchase directly with $GATA, creating a seamless bridge between compute and the decentralized economy.

We’re thrilled to announce that Gata Decentralized GPU Cloud is now LIVE! Start today → Built to accelerate the decentralization of the open AI economy, Gata enables anyone, anywhere to access decentralized GPU infrastructure, to build, train, and deploy AI models with full transparency and efficiency. GPU credits are now available for purchase directly with $GATA, creating a seamless bridge between compute and the decentralized economy.

Gata

48,232 views • 8 months ago

We’re excited to share that 🥇Llama Nemotron Super 49B v1.5 -- our latest open reasoning model -- is now #1 on the Artificial Analysis Intelligence Index - a leaderboard that spans advanced math, science, and agentic tasks, in the 70B open model category. Llama Nemotron Super 49B v1.5 is trained with high-quality reasoning synthetic data generated from models like Qwen3-235B and DeepSeek R1. It delivers state-of-the-art accuracy and throughput, running on a single H100. Key features: 🎯 Leading accuracy on multi-step reasoning, math, coding, and function-calling 🏗️ Post-trained using RPO, DPO, and RLVR across 26M+ synthetic examples 📊 Fully transparent training data and techniques If you're building AI agents and want a high accuracy, fully-open, and transparent reasoning model that you can deploy anywhere, try Super v1.5 on or download from Hugging Face 🤗 ➡️ Leaderboard:

We’re excited to share that 🥇Llama Nemotron Super 49B v1.5 -- our latest open reasoning model -- is now #1 on the Artificial Analysis Intelligence Index - a leaderboard that spans advanced math, science, and agentic tasks, in the 70B open model category. Llama Nemotron Super 49B v1.5 is trained with high-quality reasoning synthetic data generated from models like Qwen3-235B and DeepSeek R1. It delivers state-of-the-art accuracy and throughput, running on a single H100. Key features: 🎯 Leading accuracy on multi-step reasoning, math, coding, and function-calling 🏗️ Post-trained using RPO, DPO, and RLVR across 26M+ synthetic examples 📊 Fully transparent training data and techniques If you're building AI agents and want a high accuracy, fully-open, and transparent reasoning model that you can deploy anywhere, try Super v1.5 on or download from Hugging Face 🤗 ➡️ Leaderboard:

NVIDIA AI Developer

100,506 views • 11 months ago

Today we’re releasing K2 Think V2, our most capable open-source reasoning model to date. This is a fully sovereign model: trained end-to-end on IFM-curated and synthesized data, with complete transparency from pre-training through final reasoning alignment.

Today we’re releasing K2 Think V2, our most capable open-source reasoning model to date. This is a fully sovereign model: trained end-to-end on IFM-curated and synthesized data, with complete transparency from pre-training through final reasoning alignment.

MBZUAI

287,725 views • 5 months ago

Demis Hassabis on the final frontier for AI in science 🧬 : it’s not just about solving the math, it’s about having the creativity to ask the question in the first place. While coding and math are verifiable and easier to automate, natural science requires a "missing ingredient" that we haven't fully cracked yet. "Coming up with the question in the first place or coming up with the theory or the hypothesis. I think that's much, much harder and I think that's the highest level of scientific creativity.”

Demis Hassabis on the final frontier for AI in science 🧬 : it’s not just about solving the math, it’s about having the creativity to ask the question in the first place. While coding and math are verifiable and easier to automate, natural science requires a "missing ingredient" that we haven't fully cracked yet. "Coming up with the question in the first place or coming up with the theory or the hypothesis. I think that's much, much harder and I think that's the highest level of scientific creativity.”

Chris

12,890 views • 5 months ago

🚀 Big news! We’re launching Shinkai Desktop (Mac, Windows and Linux) today — the most user friendly App to run AI locally (free + open source!) and a Testnet with CoinList of the Shinkai Decentralized Network (Knowledge + Agents (very soon™️)). More 👇

🚀 Big news! We’re launching Shinkai Desktop (Mac, Windows and Linux) today — the most user friendly App to run AI locally (free + open source!) and a Testnet with CoinList of the Shinkai Decentralized Network (Knowledge + Agents (very soon™️)). More 👇

Shinkai: Local AI Agents

180,403 views • 2 years ago

🆕Scaling Test Time Compute to Multi-Agent Civilizations, with Noam Brown We're excited to publish our full conversation with Noam Brown on the frontiers of the new reasoning paradigm at OpenAI! - first principles for starting the "Multi-Agents" team - what's not captured by the "System 1/System 2" analogy for inference time compute - how Ilya Sutskever convinced him that reasoning was closer than he thought - Deep Research is existence proof that RL generalizes beyond verifiable rewards - the relationship between AI for imperfect information games (like Poker, Stratego, Diplomacy) and reasoning Enjoy! on youtube, or wherever fine podcasts are sold.

🆕Scaling Test Time Compute to Multi-Agent Civilizations, with Noam Brown We're excited to publish our full conversation with Noam Brown on the frontiers of the new reasoning paradigm at OpenAI! - first principles for starting the "Multi-Agents" team - what's not captured by the "System 1/System 2" analogy for inference time compute - how Ilya Sutskever convinced him that reasoning was closer than he thought - Deep Research is existence proof that RL generalizes beyond verifiable rewards - the relationship between AI for imperfect information games (like Poker, Stratego, Diplomacy) and reasoning Enjoy! on youtube, or wherever fine podcasts are sold.

Latent.Space

105,923 views • 1 year ago

Reasoning is central to purposeful action. Today we introduce MolmoAct — a fully open Action Reasoning Model (ARM) for robotics. Grounded in large-scale pre-training with action reasoning data, every predicted action is interpretable and user-steerable via visual trace. We are open-sourcing everything!

Reasoning is central to purposeful action. Today we introduce MolmoAct — a fully open Action Reasoning Model (ARM) for robotics. Grounded in large-scale pre-training with action reasoning data, every predicted action is interpretable and user-steerable via visual trace. We are open-sourcing everything!

Jiafei Duan

99,944 views • 11 months ago

Inkling, Thinking Machines' first open model, dropped today: 975B total / 41B active MoE, up to 1M context, reasoning natively over text, images, and audio. Serving and RL support are already live: you can run and shape it on an open stack, starting now. Day 0 support on SGLang SGLang and Miles RadixArk👇 - Inkling's new architecture (ShortConv, attention with relative positional embedding, shared expert sink MoE) is natively implemented and deeply optimized, with prefill full CUDA graph and MXFP8 KV cache - Full parameter and LoRA RL in a customized Megatron backend, train inference consistency via customized kernels, routing replay, and cross-runtime parameter synchronization - DFlash speculative decoding from Modal for low-latency serving Launching now, blog and cookbook in the comments ⬇️

Inkling, Thinking Machines' first open model, dropped today: 975B total / 41B active MoE, up to 1M context, reasoning natively over text, images, and audio. Serving and RL support are already live: you can run and shape it on an open stack, starting now. Day 0 support on SGLang SGLang and Miles RadixArk👇 - Inkling's new architecture (ShortConv, attention with relative positional embedding, shared expert sink MoE) is natively implemented and deeply optimized, with prefill full CUDA graph and MXFP8 KV cache - Full parameter and LoRA RL in a customized Megatron backend, train inference consistency via customized kernels, routing replay, and cross-runtime parameter synchronization - DFlash speculative decoding from Modal for low-latency serving Launching now, blog and cookbook in the comments ⬇️

LMSYS Org

130,409 views • 4 days ago

China just broke the AI industry. Yesterday, Baidu open-sourced their Ernie 4.5. This 424 billion parameter model beats GPT-4o. And it's completely FREE to use. Here's how to get started: → Download from HuggingFace today → Choose Ernie-4.5-VL-424B for multimodal → Or Ernie-4.5-300B for text tasks → Run locally with enough compute → Access through Baidu's API too The model excels at: • Document analysis (93.1% DocVQA score) • Visual math reasoning (78.9% Math Vista) • Chart understanding (86.8% ChartQA) Save this video, you'll need it. Want the SOP? DM me. 💬

China just broke the AI industry. Yesterday, Baidu open-sourced their Ernie 4.5. This 424 billion parameter model beats GPT-4o. And it's completely FREE to use. Here's how to get started: → Download from HuggingFace today → Choose Ernie-4.5-VL-424B for multimodal → Or Ernie-4.5-300B for text tasks → Run locally with enough compute → Access through Baidu's API too The model excels at: • Document analysis (93.1% DocVQA score) • Visual math reasoning (78.9% Math Vista) • Chart understanding (86.8% ChartQA) Save this video, you'll need it. Want the SOP? DM me. 💬

Julian Goldie SEO

73,184 views • 1 year ago

Not every AI interaction is best served with a large frontier model. There is a long tail of trillion-token use cases, from tagging to search, best served by a sub-10b-parameter model that runs in milliseconds and costs many orders of magnitude less than the frontier. We built Freesolo Flash to make training loops like SFT and RL easy and end-to-end completable through your coding agent.

Not every AI interaction is best served with a large frontier model. There is a long tail of trillion-token use cases, from tagging to search, best served by a sub-10b-parameter model that runs in milliseconds and costs many orders of magnitude less than the frontier. We built Freesolo Flash to make training loops like SFT and RL easy and end-to-end completable through your coding agent.

Freesolo

25,067 views • 2 days ago

Your RL run just spiked at step 89! But, do you know why? We’re fixing that. Today we’re launching W&B Weave Traces to give you a step by step look into your agent’s decisions. This is the first drop from our fresh new integration with OpenPipe. More RL magic is incoming.

Your RL run just spiked at step 89! But, do you know why? We’re fixing that. Today we’re launching W&B Weave Traces to give you a step by step look into your agent’s decisions. This is the first drop from our fresh new integration with OpenPipe. More RL magic is incoming.

W&B Weave

290,258 views • 10 months ago