cuppa

brewed 04:25 AM
← May 20

Thursday

may21

2026

May 22 →

the brief

AI showed real teeth today: OpenAI reports a general‑purpose model cracked a long‑standing Erdős problem, while Cohere open‑sourced its new Command A+ and Perplexity shipped production query‑aware context compression. SpaceX’s S‑1 hints at AI compute-as-a-service with Anthropic, and developer rails expanded with Vercel’s Chat SDK updates, a WordPress AI Gateway plugin, and Tailscale’s Aperture CLI.

the poursit · sip · 11 items

alerts

(01)
  • @RailwayMay 20, 09:41 PM

    Railway restores systems after outage

    Following a GCP‑related incident, Railway reports builds and workload processing are back to normal; teams should retry failed deploys and monitor jobs as support triages faster.

    Our systems are restored to full health, builds are back to nominal, processing at full volume. With these systems restored, the Support team will be able to triage and address user workload issues significantly faster. For prior information: status.railway.com/incident/KVZ1Z… x.com/Railway/status…

    signal 7hype 0status_updateoutage_recoveryplatform_opssource ↗

pulse

(07)
  • @unknownMay 20, 03:26 PM

    Cohere releases Command A+ open-source

    Cohere’s most capable model to date focuses on agentic and reasoning tasks, is optimized for lean hardware, and is released open‑source to encourage production‑grade use.

    Introducing: Cohere Command A+ We’ve created our most powerful LLM yet, optimized it to run on as little hardware as possible, and released it open-source for all. pic.x.com/C1KYnvA8JB

    signal 8hype 3model_releaseopen_sourcecoheresource ↗
  • @perplexity_aiMay 20, 05:26 PM

    Perplexity ships query-aware compression

    Productionized context compression trims tokens by up to 70% while preserving citations and boosting answer quality, cutting latency and cost for RAG and orchestration‑heavy agents.

    We've productionized query-aware compression for faster, cleaner, more-accurate search. Better context is better than more context. Our system cuts context tokens up to 70% while improving answer quality. pic.x.com/gmVr3oZRl9

    signal 8hype 2feature_releaseretrievalcontext_compressionsource ↗
  • simonw/blog· feedMay 20, 10:26 PM

    SpaceX signs Anthropic cloud compute deal

    SpaceX’s S‑1 reveals AI compute‑as‑a‑service ambitions—training Grok 5 on COLOSSUS II and inking Cloud Services Agreements with Anthropic—opening Starlink/COLOSSUS capacity to third parties.

    Quoting SpaceX S-1 — <blockquote cite="https://www.sec.gov/Archives/edgar/data/1181412/000162828026036936/spaceexplorationtechnologi.htm"><p>We have the ability to use compute resources to support our proprietary AI applications (such as Grok 5, which is currently being trained at COLOSSUS II), while also providing access to select compute capacity to third-party customers. For example, in May 2026, we entered into <strong>Cloud Services Agreements with Anthropic PBC</strong> (“Anthropic”), a...

    signal 9hype 1anthropicspacexcloud_servicessource ↗
  • @vercel_devMay 20, 10:30 PM

    Vercel Chat SDK adds callbackUrl

    Chat SDK buttons and modals can now trigger webhook callbacks, enabling pause‑and‑resume human‑in‑the‑loop flows with Workflow SDK for safer, approval‑gated agent actions.

    Buttons and modals in Chat SDK now accept a 𝚌𝚊𝚕𝚕𝚋𝚊𝚌𝚔𝚄𝚛𝚕 prop. Pair it with a @workflowsdk webhook to pause a run, then resume it when a human in the loop clicks a button. vercel.com/changelog/chat…

  • @unknownMay 20, 07:33 PM

    AI Gateway lands as WordPress plugin

    A new plugin lets WordPress sites call text, image, and video models via one API key, bringing multi‑provider AI access to a huge installed base.

    The AI Gateway plugin for WordPress is here. Install it and any plugin using the WordPress AI Client can call text, image, and video models with one API key. vercel.com/changelog/verc…

    signal 5hype 2plugin_releasewordpressai_gatewaysource ↗
  • anthropics/claude-code· feedMay 21, 01:51 AM

    Claude Code 2.1.146 ships updates

    Release renames /simplify to /code-review with effort levels, fixes Windows PowerShell issues, restores MCP pagination, and delivers quality‑of‑life improvements for coding workflows.

    v2.1.146 — What's changed Renamed /simplify to /code-review with an optional effort level (e.g. /code-review high) Auto mode no longer suppresses AskUserQuestion when the user or a skill explicitly relies on it Fixed Windows PowerShell tool failing with "command line is invalid" when pwsh is installed via winget or the Microsoft Store (regression in v2.1.124) Fixed MCP resources/list, resources/templates/list, and prompts/list dropping items past page 1 on paginating servers Fixed full-screen...

    signal 9hype 1release_notesclaude_codemcpsource ↗
  • tailscale/blog· feedMay 20, 02:15 PM

    Tailscale ships Aperture CLI

    New CLI streamlines running and iterating on AI coding agents over secure Tailscale networks, easing local‑to‑remote experimentation for agent workflows.

    Introducing: Aperture CLI — Aperture CLI makes AI coding-agent experiments easier.

    signal 5hype 2cliagent_frameworkdev_toolingsource ↗

findings

(02)
  • @gdbMay 20, 07:32 PM

    OpenAI model cracks Erdős problem

    OpenAI says a general‑purpose reasoning model found a counterexample to the planar unit distance conjecture, a rare autonomous AI advance on a central open math problem.

    An OpenAI model has achieved a major breakthrough in mathematics, by disproving a central conjecture in discrete geometry that was first posed by Paul Erdős in 1946. This is the first time AI has autonomously solved a prominent open problem central to a field of mathematics. x.com/OpenAI/status/…

    signal 8hype 3research_breakthroughdiscrete_geometryopen_problemsource ↗
  • @CloudflareMay 18, 01:05 PM

    Cloudflare tests Anthropic Mythos

    Cloudflare’s security team evaluated Anthropic’s offensive‑AI on 50 repos, arguing patch speed isn’t sufficient and outlining architectures to contain AI‑driven vulnerability discovery.

    Cloudflare's security team spent the last few weeks testing Anthropic's Mythos against fifty of our own repositories. What we learned about offensive AI, why faster patching is the wrong reaction, and what the architecture around vulnerabilities has to look like next.

    signal 9hype 1securityoffensive_aimodel_evaluationsource ↗

voices

(01)
  • @swyxMay 20, 07:22 PM

    Exa wins head‑to‑head bake‑off

    swyx shares his team’s quick convergence on Exa for agentic web search based on speed and coverage—useful signal for builders choosing retrieval backends.

    btw we did a bake off of Exa vs competitors and it took all of 1.5 hrs for the team to unanimously converge on exa lol. so proud to see my former landlords crush it - time travel back to last year and listen to a pre pmf @WilliamBryk to understand how to spot companies on a pic.x.com/hFLtifPEcJ x.com/jeffzwang/stat…

    signal 4hype 3vendor_bakeoffsearchtooling_recommendationsource ↗