cuppa

livebrewed 03:06 AM
← Jun 29

Tuesday

jun30

2026

Jul 01 →

the brief

Open weights and open interfaces led the day: DeepReinforce’s Ornith-1.0 targets agentic coding, LongCat-2.0 scales MoE to 1.6T, and an Open Memory Protocol proposes portable memories across assistants. On tooling, Claude Code added org-wide defaults and tightened MCP security, while AllenAI’s DiScoFormer explores unified generative modeling. Macroscopically, South Korea’s $1T bet signals the chip race accelerating.

the poursit · sip · 7 items

pulse

(05)
  • anthropics/claude-code· First-partyJun 29, 11:27 PM

    Claude Code adds org defaults, safer MCP

    Admins can set organization default models, sessions get clearer names, chats gain clickable attachments, and a security change prevents claude mcp list/get from auto-spawning self-approved .mcp.json servers.

    v2.1.196 — What's changed Added support for organization default models — admins set it in the org console; it shows as "Org default" (or "Role default") in /model when you haven't picked one yourself Added readable default names for sessions at start, making them easier to identify and message Added clickable file attachments in chat — Cmd/Ctrl-click reveals the file in Finder/Explorer Security: claude mcp list/get no longer spawn .mcp.json servers that a repo self-approved via a committed ....

    signal 9hype 1release_notesclaude_codemcplaunchsource ↗
  • simonw/blog· AnalysisJun 29, 04:17 PM

    Ornith-1.0 open-weight agentic coder

    DeepReinforce releases MIT-licensed coding models—9B/31B dense and 35B/397B MoE variants—built on Gemma 4 and Qwen 3.5, claiming state-of-the-art among open-source agentic coding systems.

    Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding — <p><strong><a href="https://deep-reinforce.com/ornith_1_0.html">Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding</a></strong></p> This is an interesting new open weights (MIT licensed) model, the first model release from DeepReinforce.</p> <blockquote> <p>[...] with variants including 9B Dense, 31B Dense, 35B MoE, and 397B MoE. Built on top of pretrained Gemma 4 and Qwen 3.5, it achieves state-of-the-art performance among open-source ...

    signal 8hype 1model_releaseopen_weightsagentic_codinglaunchsource ↗
  • hn/frontpage· AggregatorJun 30, 12:30 AM

    LongCat-2.0 MoE hits 1.6T params

    LongCat-2.0 is a large-scale MoE with 1.6T total parameters and 48B active, signaling continued open efforts to scale chat models beyond dense baselines.

    LongCat-2.0, a large-scale MoE model with 1.6T total and 48B Active — Article URL: https://longcat.chat/blog/longcat-2.0/ Comments URL: https://news.ycombinator.com/item?id=48727116 Points: 34 # Comments: 8

    signal 6hype 2model_releasemoeresearch_bloglaunchsource ↗
  • hn/frontpage· AggregatorJun 30, 12:05 AM

    Open Memory Protocol for AI assistants

    A proposed open standard and reference implementation aims to share a single user memory store across Claude, ChatGPT, and Curso, making agent memories portable between tools and sessions.

    Open Memory Protocol – One Memory Store for Claude, ChatGPT, Curso — Article URL: https://github.com/SMJAI/open-memory-protocol Comments URL: https://news.ycombinator.com/item?id=48726966 Points: 11 # Comments: 5

    signal 6hype 1protocol_specagent_memorygithub_repotechnicalsource ↗
  • hn/frontpage· AggregatorJun 29, 10:21 PM

    South Korea earmarks $1T for chips, robots

    A massive national plan to boost memory-chip output and humanoid robotics underscores intensifying global competition to control AI hardware and embodied AI supply chains.

    South Korea to spend $1T on more memory chip production and humanoid robots — Article URL: https://arstechnica.com/ai/2026/06/south-korea-to-spend-1t-on-more-memory-chip-production-and-humanoid-robots/ Comments URL: https://news.ycombinator.com/item?id=48726102 Points: 135 # Comments: 81

    signal 5hype 1semiconductorsgovernment_policyroboticslaunchsource ↗

findings

(02)
  • huggingface/blog· First-partyJun 29, 06:02 PM

    DiScoFormer unifies density and score modeling

    AllenAI’s DiScoFormer trains a single transformer to learn both density and score across data distributions, simplifying generative modeling while demonstrating strong cross-domain results in reported experiments.

    DiScoFormer: One transformer for density and score, across distributions

    signal 7hype 1paperresearchtransformertechnicalsource ↗
  • hn/frontpage· AggregatorJun 30, 12:38 AM

    Memory-safe setjmp/longjmp in Fil-C

    A detailed design note shows how Fil-C implements safe context switching using setjmp/longjmp patterns, with code-level techniques for stack discipline and memory safety in coroutine-style runtimes.

    Memory Safe Context Switching (longjmp, setjmp) in Fil-C — Article URL: https://fil-c.org/context_switches Comments URL: https://news.ycombinator.com/item?id=48727177 Points: 44 # Comments: 16

    signal 6hype 1memory_safetyc_languagecontext_switchingtechnicalsource ↗