NO/FOMO

Independent AI signal, once a day

The AI briefing worth opening.

ISSUE DATE2026-06-28ENGLISH EDITION
This issue
—
All time
—

Hacker News

7 stories
01

Semgrep: GLM 5.2 beats Claude in our Cyber Benchmarks

Semgrep released evaluation results showing that Zhipu AI's new GLM 5.2 model outperformed Anthropic's Claude in domain-specific cybersecurity benchmarks. The automated testing evaluated performance across code analysis, vulnerability detection, and security patch generation. GLM 5.2 demonstrated superior syntactic reasoning and lower false-positive rates when triaging security alerts. This milestone highlights a shift in performance on complex codebase structures, where alternative models are increasingly challenging proprietary Western models in specialized engineering domains. (source: https://semgrep.dev/blog/2026/we-have-mythos-at-home-glm-52-beats-claude-in-our-cyber-benchmarks/)

SemgrepGLM 5.2ClaudeCybersecurity BenchmarksVulnerability DetectionArtificial IntelligenceLarge Language ModelGenerative AI
02

Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding

Deep-reinforce published details of Ornith-1.0, an LLM optimized for autonomous coding tasks using internal self-scaffolding instead of external prompts or framework wrappers. By learning to orchestrate its own planning, tool execution, and error recovery directly within its weights, the model shows gains on software engineering benchmarks. This architecture enables the agent to dynamically adapt to unexpected syntax challenges and runtime errors without relying on hand-coded prompts. (source: https://deep-reinforce.com/ornith_1_0.html)

Ornith-1.0Self-ScaffoldingAgentic CodingLarge Language ModelsAI AgentsArtificial IntelligenceLarge Language ModelAI Agent
03

Wayfinder Router: deterministic routing of queries between local and hosted LLM

An open-source tool named Wayfinder Router has been released to provide deterministic routing of user queries between locally hosted and cloud-based Large Language Models. By analyzing incoming natural language queries, the middleware dynamically decides whether a lightweight local model can resolve the task or if it requires a cloud API. This hybrid system helps developers balance cost, latency, privacy, and query complexity. (source: https://github.com/itsthelore/wayfinder-router)

LLM RoutingLocal LLMHybrid AI ArchitectureDeterministic RoutingAPI Cost OptimizationArtificial IntelligenceLarge Language ModelGenerative AI
04

Google limits Meta's use of its Gemini AI models

Google restricted Meta Platforms from using its Gemini artificial intelligence models, as reported by the Financial Times. While Meta primarily develops its open-source Llama model series, it utilizes proprietary frontier models like Google's Gemini for testing, benchmarking, and cross-model validation. Google's move aims to protect its intellectual property and limit competitors from refining systems with Gemini's outputs, underscoring rising industry friction over API terms. (source: https://www.cnbc.com/2026/06/28/google-limits-metas-use-of-its-gemini-ai-models-ft-reports.html)

Gemini AIMeta PlatformsModel LicensingAPI RestrictionsFoundation ModelsArtificial IntelligenceGenerative AILarge Language Model
05

Tokenmaxxing is dead, long live tokenmaxxing

An analysis of 'tokenmaxxing' trends within generative AI shows a shifting focus from raw throughput and large context windows to efficient token utilization, agentic workflows, and semantic optimization. The transition indicates a maturity phase in the large language model industry, moving away from brute-force scale to refined efficiency. Utilizing advanced orchestration techniques, developers are maximizing model performance while generating fewer, more targeted tokens. (source: https://12gramsofcarbon.com/p/agentics-tech-things-tokenmaxxing)

TokenmaxxingLarge Language ModelsAgentic AIContext Window OptimizationComputational EfficiencyLarge Language ModelAI AgentGenerative AI
06

A way to exclude sensitive files issue still open for OpenAI Codex

A long-standing GitHub issue for OpenAI Codex remains unresolved, highlight developer requests for a standardized mechanism to exclude sensitive files, such as private keys and proprietary algorithms, from being processed or indexed. The lack of a reliable local exclusion method presents security and compliance challenges for software supply chains using AI assistants. This ongoing request highlights critical data governance needs in generative code environments. (source: https://github.com/openai/codex/issues/2847)

OpenAI CodexData PrivacyInformation SecurityAI Code AssistantGitHub IssueArtificial IntelligenceGenerative AILarge Language Model
07

I used Claude Code to get a second opinion on my MRI

A developer documented an experiment using Anthropic's Claude Code, powered by the Claude 3 Opus model, to analyze high-resolution MRI scans and corresponding clinical reports. The multimodal AI system demonstrated capabilities in parsing dense radiological images, highlighting diagnostic trends, and identifying potential areas of concern that matched professional findings. The workflow shows the potential of using agentic developer tools for complex, consumer-level multimodal medical data analysis. (source: https://antoine.fi/mri-analysis-using-claude-code-opus)

Claude CodeClaude 3 OpusMedical Image AnalysisMultimodal AIAI AgentArtificial IntelligenceMultimodalAI Agent

Twitter

5 stories
01

Gary Marcus Shares Insights On The Performance Of Current Frontier Models

Gary Marcus highlighted recent open-source research evaluating the performance capabilities of state-of-the-art frontier artificial intelligence models, specifically referencing GPT-5.5. The referenced research employs an open-source evaluation framework to rigorously stress-test and benchmark advanced large language models in real-world scenarios. This benchmarking effort focuses on evaluating true capabilities, comparative performance metrics, and linguistic reasoning as frontier models scale. (source: https://x.com/GaryMarcus/status/2071031563324662252)

Frontier ModelsGPT-5.5BenchmarkingOpen SourceGenerative AIArtificial IntelligenceLarge Language ModelResearch Progress
02

Rational Approaches To Regulating Frontier API Models And Transparency

Yann LeCun shared perspectives on the evolving landscape of artificial intelligence governance, specifically discussing the logic of regulating frontier API models to ensure transparency and public accountability. The commentary addresses how governments can implement structured policy frameworks at the API level to monitor cutting-edge large language models. This approach seeks to balance rapid technology deployment with safety and systemic oversight without hindering developer innovation. (source: https://x.com/ylecun/status/2071266002369659007)

AI RegulationFrontier ModelsAPI GovernanceAI PolicyTransparencyArtificial IntelligenceLarge Language ModelIndustry News
03

The Growth and Potential of the Open Source AI Model Ecosystem

Nathan Lambert highlighted the growth and untapped potential of the open-source artificial intelligence model ecosystem, pointing out that smaller, specialized builders are creating highly effective alternatives to massive proprietary systems. While centralized frontier models dominate mainstream media attention, this decentralized open-source movement is expanding and challenging industry centralization by democratizing advanced machine learning technologies for niche and enterprise applications. (source: https://x.com/natolambert/status/2071284002338873625)

Open SourceArtificial IntelligenceModel DevelopmentFrontier ModelsAI InnovationArtificial IntelligenceLarge Language ModelOpen Source
04

Analyzing The Implications Of The Klarna Effect On AI Market Adoption

Gary Marcus analyzed the strategic business impacts of 'The Klarna Effect,' detailing how the fintech company's aggressive integration of generative AI models and automation agents reshapes operational workflows and industry expectations. The commentary evaluates the long-term sustainability of corporate digital transformations, questioning whether rapid shifts toward automated customer service and backend infrastructure deliver lasting value or reflect transient market trends. (source: https://x.com/GaryMarcus/status/2071026100109734094)

Klarna EffectAI AdoptionEnterprise AIBusiness AutomationMarket TrendsArtificial IntelligenceAI AgentIndustry News
05

Concerns Rise Regarding Impact of Vibe Regulation on Frontier AI Models

Nathan Lambert expressed concerns over the emergence of 'vibe regulation' in artificial intelligence policy, warning that subjective and poorly defined regulatory frameworks could impede innovation. The critique argues that relying on vague sentiments and fluid metrics for AI policy evaluations creates operational bottlenecks and complicates safety compliance, potentially hindering the deployment of advanced frontier models within the broader technology ecosystem. (source: https://x.com/natolambert/status/2071240520840683521)

Frontier ModelsAI RegulationTechnology PolicyModel SafetyInnovationArtificial IntelligenceLarge Language ModelIndustry News