Google says DiffusionGemma hits 1000+ tokens/sec on an H100 by treating text like an image. We spun up a RunPod container to build a dashboard and a game to find out if the speed is real.
00:00
Better Stack
1,489 posts
Better Stack
@BetterStackHQ
Tracing reimagined with eBPF and OpenTelemetry.
Instrument your clusters without any code changes.
Get an AI SRE and a robust MCP server.
Joined January 2021
- The US government ordered Anthropic to shut down Claude Fable 5 and Mythos 5 for everyone, citing national security concerns. Here's a breakdown of what happened.
00:00 - Better Stack reposted
just watched the entire podcast episode from @BetterStackHQ with @dreamsofcode_io I deeply resonate with much of what Elliot was saying. He did a fantastic job articulating my feelings about the industry rn especially in the era of coding agent. Its just so good to see people - Better Stack reposted
We have moved from incident.io to @BetterStackHQ for our incident reporting! • status.scholarxiv.com The status page also reports more monitors; platform, API, AI-Chat, Sandboxes and Subscriptions with status history and chart. We've also made our health check
- Connect Linear to Better Stack AI SRE and handle issues with simple prompts.
00:00 - Are you tired of explaining the same repo to Claude Code every morning? Check this out! PAI fixes that with memory, skills, and workflows so your AI actually remembers how your project works.
00:00 - Better Stack reposted
A bit late, but still thrilled to see @BetterStackHQ thoughtful review of AFFiNE. If you care about data ownership and want the freedom to switch seamlessly between docs and whiteboards, AFFiNE is worth checking out!This open-source tool merges Notion, Miro, and Markdown into one workspace. Docs, diagrams, and databases on an infinite canvas. It's called AFFiNE. Devs, this is one to watch.
00:00 - Anthropic just dropped Fable 5, the first Mythos class Claude model. State of the art on nearly every benchmark and 20% ahead of GPT 5.5 on SWE Bench Pro. The pricing/usage might be an issue though…
00:00 - NotebookLM is great… until you need private docs, code, or research. We tested Open Notebook, a self-hosted alternative for devs, against NotebookLM and AnythingLLM.
00:00 - Stop wasting tokens. Headroom compresses tool outputs, code and RAG chunks before your LLM reads them. Built by a senior Netflix engineer, it runs fully locally, has zero-cost compression, and is fully open-source.
00:00 - Mount S3 as a local drive with JuiceFS. Learn how multi-tiered caching delivers local NVMe speeds, run local AI from the cloud, and build a real-time tracking dashboard with Better Stack.
00:00 - Better Stack reposted
3rd week in a row, no downtime! @BetterStackHQ 99.95% global uptime from the last 30 days. @digitalocean tight, tight, tight! 🙂↕️
