Fable/Mythos just got banned so we curated the distributed AI infra stack end to end from data curation to training and serving built upon the open source eco-system so frontier intelligence is not monopolized. AI has grown more close-source, not open and the safety and honesty of the models have become a black box and is impacting everyone who relies on them for their research and work. Our dependency could suddenly become unreliable. While I believe the SOTA model has a reason to keep close-sourced to reward the effort behind training the best model, the general public has the right to access good open-sourced ones for the public safety and I will keep working together with the community to make this happen along with @sgl_project @vllm_project @NovaSkyAI @NVIDIAAIInfra @LeRobotHF @huggingface @raydistributed etc.

Impressive work from the Cosmos team. Check out the Cosmos 3 technical report: arxiv.org/pdf/2606.02800

Nemotron 3 Ultra is an impressive 550B parameter (55B active) MoE model. It was trained with Ray / Megatron / vLLM (via NeMo RL).

Ray strikes again MAI-Thinking-1 uses @raydistributed for both pre-training fault tolerance and RL orchestration

It's interesting to see @MicrosoftAI uses ray actors not just for controller and rollout workers but problem workers for the posting training of the MAI-Thinking-1 model. Instead of introducing third party dependency like @modal for sandboxing, Ray actors could provide finer granularity and control for heterogeneous compute which could translate to better utilization of the unused CPU resources in the GPU cluster and easier communication of the agent execution results. Also the part of work I did was to support @sgl_project with @raydistributed backend to better support RL infra especially in weight syncing.

Amount paid out by $IDLE has increased signifantly. I can see this being adopted world wide as so many are trying to earn via crypto but are usually on the losing end. Meanwhile setting up your resources on earnidle.com is easy and a guaranteed income.

MILESTONE UPDATE: Total $IDLE transactions just surpassed 100,000. Adoption becoming real now. $9,000 USDC paid out directly to users. $IDLE also just became part of the @raydistributed ecosystem, which gives them access to 9 billion devices. So what’s not to like about this. $IDLE also has a real flywheel in full effect. The flywheel is not from charging a tax or AI perp/bot trading profits or scamming/rugging or using creator rewards. It simply comes from selling one of your resources (eg bandwith, wallet, GPU, API) that might be sitting idle each day/night and letting it earn some passive income - all paid in stable USDC. From that total, 10% is automatically swapped into $IDLE token via Jupiter swap and burned forever. Currently 105m tokens have already been burnt under the flywheel. Imagine when the mobile resource goes live!!! For live adoption stats see: dune.com/idle_protocol/idle

Is not the original main X got hacked this is a temporary x they are using before they recover the main X handle. It was a sim swap hack..

Wonderful! $Idle is the new goldmine..

1/ IDLE nodes are now part of the @raydistributed ecosystem, powered by @AnyscaleCompute - the framework powering AI workloads at OpenAI, Anthropic, and major AI labs globally. Every IDLE node is now a Ray worker. Every Ray job can route to IDLE. Distributed compute, USDC settlement on Solana. 🧵

Out here at @raydistributed Day NYC by @anyscalecompute. Ray has quietly become the backbone for distributed AI in production - community keeps shipping and scaling. @robertnishihara, @pcmoritz, @KeertiMelkote, @istoica05, & team are paving the way.

just integrated $IDLE nodes into the @raydistributed ecosystem (powered by @AnyscaleCompute) Every IDLE node is now a native Ray worker. That means any developer running Ray jobs can tap into 5,400 global nodes for inference, data processing & more and pay in USDC on Solana. Your device just became: • Ray worker • NVIDIA NIM endpoint • Mistral AI node • x402 payment endpoint Every job burns $IDLE Decentralized compute meets production AI infrastructure

this is actually huge for distributed computing

Read the latest on #AzureLinux 4.0 upcoming public preview and how @microsoft is working alongside the open source community on the building blocks of an open agentic stack with projects including #A2A and @raydistributed : opensource.microsoft.com/blo… @brendandburns