cuppa

brewed 03:04 AM
← Jun 03

Thursday

jun04

2026

Jun 05 →

the brief

Maturity over hype set the tone: OpenAI expanded GPT-Rosalind for bench scientists while Anthropic detailed concrete containment patterns for Claude. Cloudflare pushed pragmatic BGP hygiene, and dev tooling kept shipping with Next.js and Claude Code updates. Hugging Face’s DPO recipes and a $1,500 red-team test grounded capability claims against real constraints.

the poursit · sip · 8 items

pulse

(03)
  • vercel/next.js· feedJun 4, 12:16 AM

    Next.js canary trims legacy paths

    Deprecates undocumented custom server methods, fixes a build hang on sync IO abort, and improves cache tracing and stage checks for adapter scenarios.

    v16.3.0-canary.40 — Misc Changes docs: add GitHub SSH authentication guidance: #94360 Deprecate undocumented custom server methods: #94348 Trace cacheHandler and cacheHandlers when using adapters: #94197 refactor: make stage advancing logic more generic: #94349 docs: document Cache Components behavior for navigation hooks: #94387 fix: build hangs if sync IO aborts before root chunk is flushed: #94365 refactor: make early/late checks stricter: #94358 refactor: remove delayUntilRuntimeStage: #9...

    signal 8hype 1nextjsrelease_notesframeworksource ↗
  • anthropics/claude-code· feedJun 3, 09:31 PM

    Claude Code improves agent clarity

    Adds waitingFor in --json for blocked sessions, proper Grep/Glob tools on native builds, persistent /effort defaults, and safer autocomplete behavior.

    v2.1.162 — What's changed claude agents --json now includes waitingFor showing what a waiting session is blocked on (e.g. permission prompt) --tools: explicitly listing Grep/Glob now provides the dedicated search tools on native builds with embedded search (previously these names were silently ignored) /effort now confirms when your chosen level will persist as the default for new sessions Clicking a slash command in the autocomplete menu now fills it into your prompt instead of running it im...

    signal 8hype 1release_notesagent_cliclaude_codesource ↗
  • openai/blog· feedJun 3, 01:15 PM

    OpenAI expands GPT-Rosalind toolkit

    Adds stronger biological reasoning, medicinal chemistry insight, genomics analysis, and experiment workflow orchestration, signaling a push toward domain-specific AI assistants for wet-lab scientists.

    Introducing new capabilities to GPT-Rosalind — GPT-Rosalind advances life sciences research with enhanced biological reasoning, medicinal chemistry expertise, genomics analysis, and experimental workflow capabilities.

    signal 6hype 3model_updatespecialized_gptopenaisource ↗

findings

(04)
  • hn/frontpage· feedJun 4, 12:56 AM

    Testing if LLMs can hack apps

    A $1,500 controlled test against a purposely vulnerable app evaluates how far leading models get at real exploitation versus CTF-style prompts, surfacing clear limits and risks.

    I built a vulnerable app and spent $1,500 seeing if LLMs could hack it — Article URL: https://kasra.blog/blog/i-spent-1500-seeing-if-llms-could-hack-my-app/ Comments URL: https://news.ycombinator.com/item?id=48392343 Points: 59 # Comments: 27

    signal 7hype 1securityllm_evalred_teamingsource ↗
  • cloudflare/blog· feedJun 3, 05:00 PM

    Cloudflare pushes First AS enforcement

    Cloudflare details and promotes enforcing the first AS in BGP AS_PATHs to blunt forged-path attacks that RPKI misses, a practical routing hygiene step operators can adopt now.

    Enforcing the First AS in BGP AS_PATHs — BGP is vulnerable to routing hijacks and path leaks that negatively impact traffic on the Internet. RPKI helps solve some of these problems, but for some forged paths, we need to rely on a simpler mechanism: First AS enforcement in BGP.

    signal 6hype 1bgpnetwork_securityinternet_infrasource ↗
  • huggingface/blog· feedJun 3, 12:55 PM

    DPO techniques beyond chatbots

    Hugging Face shares methods and examples for applying Direct Preference Optimization to classification, extraction, and ranking tasks, extending alignment techniques beyond conversational agents.

    Direct Preference Optimization Beyond Chatbots

    signal 7hype 1dpopreference_optimizationalignmentsource ↗

voices

(01)