Agents & Frameworks
Test-Time Scaling Cost Falls as PRMs Reuse Generator KV-Cache
KV-PRM cuts process reward model compute by reusing generator KV-caches instead of re-encoding text, reducing verification cost by up to 5,000x and making multi-agent.
The Groundy archive · Page 11 of 34
All articles
Browse Groundy's complete archive of 797 articles on AI, developer tools and infrastructure. Page 11 of 34.
The archive
241–264 of 797 articles · Newest first
- Open Source
RISCBoy Open-Sources a Handheld Console Designed From Scratch
RISCBoy publishes complete hardware files for a handheld console with RISC-V CPU, graphics pipeline, and KiCad PCB, enabling fabrication without vendor approval.
- Open Source
Soofi S: Sovereign AI Is Cheap to Adopt, Expensive to Sustain
Soofi S shows European AI sovereignty requires auditable training, not just data residency. Open weights enable compliance but the retraining burden to stay current is the.
- Agents & Frameworks
Claude Code Skills vs Cursor Rules vs MCP: How Agent Skill Systems Compare
The July 2026 Skill Market paper defines reusable agent skills, but Claude Code, Cursor, and MCP encode skills in incompatible formats, forcing teams to pick a runtime before.
- Agents & Frameworks
TTHE: Test-Time Harness Evolution Changes the Test-Code Contract for Coding Agents
TTHE lets a coding agent rewrite its test rig during evaluation, raising coverage but blurring spec and verification. Benchmarks must now defend why their tests stay fixed.
- Developer Tools
Grok Build CLI Sends File Listings and Code Fragments to xAI, Widening Endpoint Trust Boundaries
Grok Build CLI sends file listings, editor state, and command output to xAI's cloud for inference, making that local context payload eligible under the consumer privacy.
- Agents & Frameworks
Git-for-Data for Agentic Lakehouses: Why Agents Need Versioned State
GitLake layers git commits, branches, and merges over Apache Iceberg so agents write to reviewed branches and roll back bad outputs before they hit production tables.
- Developer Tools
Vercel Adds Zero-Config Node Server Deploys: Hono's Pattern Goes Mainstream
Vercel auto-detects Express and Fastify for zero-config deploys on Fluid Compute. Static files must use public/**, express.static is ignored, and the standard cap is 250 MB.
- Security
Context-Aware Prompt Injection Defenses for LLM Agents: Why Static Filters Fail
Static filters miss prompt injection in LLM agents because a payload becomes malicious when tool outputs or retrieval chunks meet runtime state. ARGUS tracks provenance.
- Developer Tools
OpenAI's Codex Refresh: The Upgrade That Puts Pressure on Cursor and Claude Code
OpenAI's July 2026 Codex refresh bundles a frontier agent into ChatGPT plans, challenging Cursor and Claude Code to prove value on workflow quality rather than model access.
- Security
Final-Token vs Full-Sequence Safety Probes: Why LLM Red Teams Need Both
Final-token safety probes miss jailbreaks when unsafe evidence hides in earlier prefill tokens, so red teams should pair single-readout checks with trajectory diagnostics.
- Security
s1ngularity Supply Chain Attack Hits Nx: What Monorepo Teams Should Patch
Vercel confirmed s1ngularity compromised Nx packages. The real risk is build-time plugins: they run with CI access and can rewrite artifacts before runtime scanners see them.
- Agents & Frameworks
Game Theory Can Cut Multi-Agent LLM Hallucination, But Only If Payoffs Align
Two July 2026 preprints show game-theoretic coordination can cut LLM hallucination, yet consensus breaks if one agent prioritizes cost, latency, or engagement over agreement.
- Agents & Frameworks
WebSwarm: Recursive Multi-Agent Search vs Flat Orchestration
WebSwarm's recursive multi-agent search beats flat ReAct on deep-and-wide benchmarks. Framework builders need spawn-and-merge primitives, not just larger context windows.
- Developer Tools
Vercel Sandbox Hits 32 vCPU: Agent Testing Escapes Laptop Limits
Vercel's agentic infrastructure push lists sandboxed VMs, but the 32 vCPU tier implied by the headline is not confirmed on its public pages. Wait for specs before moving CI.
- Infrastructure & Runtime
GLM-5.2: vLLM Int4 Drops MTP Without Patches, SGLang FP8/NVFP4 Keeps It
GLM-5.2's speculative decoder speeds decode, but vLLM int4 drops MTP without a community patch. SGLang FP8/NVFP4 keeps MTP intact. Format, not kernel speed, decides serving.
- Security
What Vercel BotID Catches in SEO Poisoning That WAFs Miss
Vercel BotID exposed verified Googlebots recrawling historical SEO-poisoned pages on a bank site, showing how bot identification doubles as a cloaking sensor that WAFs miss.
- Security
How Attribution Graphs Expose Why LLM Refusal Training Misses Jailbreaks
Attribution graphs expose LLM jailbreaks as distributed feature circuits, not a single suppressed safety direction. Circuit ablation works for open models, not closed APIs.
- Infrastructure & Runtime
Serving DeepSeek on Azure: Compliance Without Owning the GPU Fleet
DeepSeek on Azure through Vercel's AI Gateway lets regulated teams route the model inside Microsoft's perimeter, turning a binary compliance ban into a per-token cost call.
- Culture & Society
When AI Generates the Slides, the Talk Stops Being an Effort Signal
OmniPresent generates coherent slide decks, posters, and videos from scientific papers, so polished decks stop signaling effort and academic committees must rely on live Q&A.
Culture & SocietyWhen CP-SAT Solvers Set Your Shifts, Labor Laws Become a Soft Constraint
CP-WSP lets labor protections such as schedule stability become weighted CP-SAT penalties, so the solver can trade away fair-scheduling rights whenever the penalty is cheap.
- Models & Research
Analytic Inference Cuts Bayesian Deep Ensemble Serving Cost, But Leaves Training as the Bottleneck
A new arXiv preprint replaces sampling-based averaging in Bayesian deep ensembles with closed-form Bayesian aggregation, cutting per-query inference cost and shifting the.
- Culture & Society
When AI Counts White Blood Cells, Who Verifies the Result?
A July 2026 preprint claims 99.04% WBC classification accuracy, but commercial systems already automate differentials. The remaining task, verifying counts, falls on senior.
- Agents & Frameworks
Do Coding Agents Memorize Their Benchmarks? DeepSWE Tests on Unseen Tasks
DeepSWE evaluates frontier agents on original, long-horizon tasks held out of GitHub, exposing when coding benchmarks measure memorized fixes instead of engineering skill.
