Scott Condron @_ScottCondron

inference providers have to choose to be either fast or cheap if you choose speed as a provider, you'll have idle gpus. but if you optimize to be cheap, customers are going to be waiting the pay-as-you-go pricing model means you pay per token, and the provider pays for gpu time. being fast means putting fewer people on each gpu, so each token costs more groq and cerebras chose to be fast. they keep the # of people they serve per gpu low coreweave, novita, and crusoe chose to be cheap. they charge much less this is just how gpus work. on a blackwell, if one person uses the gpu alone, they get 1,000 tokens a second, and the gpu produces 1,000 tokens a second in total because one person can't keep it busy if many people share the gpu, it produces 60,000 tokens a second in total, but each person only gets about 20 to 100 of those

Raising the security bar for agents helps everyone building them. If you want to evaluate SPACE on your own CKS cluster, our setup guide has the steps: crwv.co/utuZ7

Perplexity @perplexity_ai

3 days ago

We're open sourcing SPACE, the sandbox platform that runs every task in Perplexity Computer. SPACE gives each agent task its own Firecracker microVM and controls credentials and outbound traffic from outside the VM. Available now under Apache 2.0: github.com/perplexityai/S…

137 212 2K 291K 2K

🔦 Beam, Reflection's first model with open weights, is live today on CoreWeave Serverless Inference. Call it through an API and leave the GPUs to us. We are also proud to share that @reflection_ai used @wandb in CoreWeave Forge during Beam's training. Developers in the private preview have been calling it fast and accurate, and they like that it doesn't ramble. See what @TheCodingSoup on our team built with private preview access below. Get started with Beam here: crwv.co/utuxk

Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners. We're releasing: - Weights, tokenizer, and reference implementation - BF16, FP8 and NVFP4 weights and a DFlash2 drafter - The open source harness we built and trained Beam on: Mirror CLI Weights and model card on Hugging Face: huggingface.co/reflection

recommended reading. i like the term "coherence" over "taste". historically, simplified, a human would approach a software writing human, asking them to build a thing. that's an iterative process, until software writing human truely understands what the other human needs and wants. which is also why we moved away from waterfall. you often can't know what you want before you've build a first shitty solution you can feel out. and your wants might change. in this scenario, the outcome is what matters, not the program structure itself, unless that makes the outcome worse/impossible (bugs, perf, ease of adding features). the software writing human's skill is not so much programming itself, but understanding the other humans wants, pulling out domain knowledge from the other human, asking questions, and ultimately translating the other human's wants into programs, while ensuring the program structure is fit to be modified for future needs. now, can you replace the software writing human with an agent? for simple one-off stuff, yes. if the human with a want is technically minded that extends to more complex programs. most humans suuuuuck at this iterative colab with agents tho, as they are bad at structuring and communicating their wants. and agents currently still suck at the software writing human skill of colaborating with another human to understand their wants and translate them to a program. my gut tells me that this is because the training data and RL for coding agent tasks is largely composed of trajectories where a software writing human and an agent colaborate. that missed the other half of the problem: the human with a want and software writing human interaction. and until we can encode that agents for writing programs will likely only be effective in the hands of software writing humans, or very technically minded humans. another thing agents currently suck ar is translating software writing human shallow instructions into architecture that's not a house of cards. i think this too is a consequence of the lack of training data/RL-able trajectories that encode the process a team of software writing humans go through when building a thing. we do have digitized artifacts (design docs, commits/PRs, issue trackers) that capture some but not all of that. i think this will improve, as we all are now doing design sessions with our agents, so labs could in theory train on that. however, even if this data were available and becomes part of training, the pretraining data still has tons of shit code from around the internet in it, ehich posttraining would have to combat. welcome to my morning word salad.

Ben Davis @davis7

4 days ago

davis7.substack.com/p/the-labs-hav…

2 2 55 26K 71

Spillover mode Sandboxes on your own CW K8s clusters Scale to hundreds of thousands of CPU cores In any zone, globally Choose your compute vs serverless with each request Private networking with VPN or Direct Peering Sandboxes at CoreWeave Scale

Agent stack, consolidation status: Harnesses: ✅ Claude code, Codex, OpenCode, Pi Telemetry: ✅ OTel GenAI Tools: ✅ MCP Evals: 🫠

@just_cameron Hey Cameron, fyi we have free LLM judges on traces until 2027 on our new OTEL native observability tool coreweave.com/blog/automate-…

@ClaudeDevs @CoreWeave coreweave.com/blog/automate-…

@ClaudeDevs Free LLM judges and automated failure discovery on OTEL gen ai traces on @CoreWeave until Jan 1

Now in public beta, Claude Platform can send OpenTelemetry traces from the Messages API, Claude Managed Agents, and the Claude SDKs to your observability stack.

Good thing we want all in on OTEL gen ai so this works out of the box on a native OTEL backend, no proprietary telemetry conversion. Try now with Agent Lens on @CoreWeave - with free LLM judges and automatic failure discovery on traces until Jan 1 docs.coreweave.com/products/agent…

ClaudeDevs @ClaudeDevs

4 days ago

Now in public beta, Claude Platform can send OpenTelemetry traces from the Messages API, Claude Managed Agents, and the Claude SDKs to your observability stack.

270 351 6K 957K 3K

@charles_irl 👏

lol you think we just rent GPUs? @deok_filho would like a word. CoreWeave Forge runs Serverless Inference, Serverless RL, fine-tuning, Notebooks and Sandboxes, all available today. Try CoreWeave Forge Pro free for 30 days: crwv.co/utcVG

As AI agents write code, run experiments, and work together, complex evaluations get harder to interpret. A single eval score obscures the ways agents reached it. Today, we're releasing Transect: an open-source tool that turns agent transcripts into interactive timelines. 🧵

@AISecurityInst 👏👏

Submit this form if you'd like to use Serverless GPU Sandboxes! forms.gle/QQ1tF8NtaMPqa9…

@NathanFlurry Amazing! Excited to try this out! 👏

We’re a few years in and no one has cracked the “PMs building scorers” workflow. Annotation is noisy, defining good is hard, PMs are busy Excited with what @CoreWeave has cooking here! Lmk if you have a working agent and want to make it better/cheaper on open models

I think you need to constrain the space of what can be built, but keep it unhinged enough to make it fun / interesting. Also, I think you want to have a discussion board which stops short of actually being a feature request.

I'm going to reply here as I figure this idea out x.com/_ScottCondron/…

Scott Condron @_ScottCondron

2 weeks ago

Ok, this (autonomous product) but with Sign in with ChatGPT. Get some skin in the game from the users. Maybe they sign up to be in a raffle for a contribution of tokens for a given feature and then they get a number of feature request votes

0 0 2 282 0

Is anyone building an autonomous product, like a Wikipedia-style website where it uses AI to get consensus on feature requests and then builds them automatically? Would be a fun social experiment