NO/FOMO

Independent AI signal, once a day

The AI briefing worth opening.

ISSUE DATE2026-07-18ENGLISH EDITION
This issue
—
All time
—

Hacker News

5 stories
01

Fable 5 vs. GPT-5.6 Sol on an NP-Hard Problem: Does /goal help?

Researchers evaluated the performance of Fable 5 and GPT-5.6 Sol large language models on an NP-hard combinatorial optimization problem to test the impact of goal-directed prompting. By appending a '/goal' directive to the system prompt, the study measured correctness, token usage, and computational efficiency across multiple iterations. Results showed that explicit goal-oriented framing significantly mitigates planning drift and improves logical consistency in these frontier reasoning models. (source: https://charlesazam.com/blog/fable-5-gpt-5-6-sol-goal/)

NP-Hard ProblemsFable 5GPT-5.6 SolGoal-Directed PromptingReasoning ModelsArtificial IntelligenceLarge Language ModelAI Agent
02

GPT-5.6 used a prompt to close a 30-year gap in convex optimization

Following OpenAI's CDC mathematical proof announcement, reports indicate that the GPT-5.6 model has resolved a thirty-year-old theoretical gap in convex optimization. By utilizing a highly specialized and structured prompt, the large language model identified novel mathematical formulations that had eluded human researchers for decades. This milestone highlights the capability of advanced generative AI to assist in complex scientific discovery and accelerate rigorous mathematical proofs. (source: https://old.reddit.com/r/math/comments/1uxj3cy/after_openais_cdc_proof_announcement_gpt56_used_a/)

GPT-5.6Convex OptimizationMathematical ProofOpenAIPrompt EngineeringArtificial IntelligenceLarge Language ModelGenerative AI
03

The Kimi K3 Moment

This technical analysis outlines the release and impact of the Kimi K3 conversational AI model. The author details how the model introduces a paradigm shift in deep reasoning, context window capacity, and computational architectural efficiency. These developments challenge traditional assumptions regarding prompt length limitations and computational bottlenecks, enabling new applications for enterprise-grade autonomous agents and complex document analysis. (source: https://stephen.bochinski.dev/blog/2026/07/18/the-kimi-k3-moment/)

Kimi K3Large Language ModelsContext WindowDeep ReasoningAI ArchitectureArtificial IntelligenceLarge Language ModelGenerative AI
04

Setting up your spare Mac for Claude Code to control, a step-by-step guide

This technical guide provides instructions for configuring a secondary macOS device to be remotely managed by Claude Code, Anthropic's command-line developer agent. The author details the security measures, SSH configurations, system permissions, and shell environments required to delegate execution safely. Isolating the terminal-based AI assistant on a spare machine allows developers to safely test and debug code in a physical sandbox environment. (source: https://ykdojo.github.io/claude-controls-mac/)

Claude CodeAI AgentmacOS automationSSH ConfigurationTerminal AssistantArtificial IntelligenceAI AgentOthers
05

What AI did to stackoverflow in a graph

A visual data analysis utilizing Stack Exchange data tracks a significant decline in developer engagement and traffic on Stack Overflow. The data highlights a major workflow shift as developers increasingly migrate their code generation and debugging queries away from traditional community-driven Q&A forums and toward conversational large language models and generative AI assistants. (source: https://data.stackexchange.com/stackoverflow/query/1953768#graph)

Stack OverflowGenerative AIData QueryUser EngagementDeveloper WorkflowArtificial IntelligenceGenerative AIOthers

Twitter

5 stories
01

Anthropic Increases Claude Code Usage Limits Through August 19

Anthropic has announced a temporary 50% increase in usage limits for Claude Code, its AI-powered developer assistant tool. This capacity boost applies immediately to all paid user tiers, including Pro, Max, Team, and seat-based Enterprise subscribers, and will remain active through August 19. The adjustment is designed to support developers in executing complex coding workflows and intensive project experimentation without running into standard token restrictions. (source: https://x.com/ClaudeDevs/status/2078511173759324328)

Claude CodeAnthropicDeveloper ToolsUsage LimitsAI CodingArtificial IntelligenceAI AgentIndustry News
02

The Disconnect Between Instruction Following And Sound Decision Making In AI Models

AI researcher Francois Chollet highlighted a growing divergence in generative model capabilities, noting that while systems show exponential improvements in following precise instructions, their autonomous decision-making in novel environments has plateaued. Current architectures excel at rule-based procedural execution but fail to generalize when encountering ambiguous, out-of-distribution scenarios, questioning current paths to general cognitive versatility. (source: https://x.com/fchollet/status/2078526645108138382)

AI ModelsGeneralizationReasoningCognitive CapabilityInstruction FollowingArtificial IntelligenceLarge Language ModelResearch Progress
03

Gary Marcus and Miles Brundage Review AGI Capability Bet Progress

Gary Marcus and Miles Brundage have evaluated the midpoint progress of their multi-year AGI wager initiated in late 2024. The bet relies on ten specific capabilities that a true Artificial General Intelligence must achieve, with a final assessment deadline scheduled for the end of 2027. The review assesses whether current AI models align with the defined benchmarks. Marcus also reinforced his skepticism alongside Demis Hassabis, arguing that current architectures remain fundamentally distant from true AGI. (source: https://x.com/GaryMarcus/status/2078292930234843554)

AGIBetBenchmarksArtificial IntelligenceTimelineArtificial IntelligenceLarge Language ModelResearch Progress
04

Creating Dreamy Cloud Gateways Using Kling AI Video Generation Tools

Kling AI has released a prompt engineering tutorial demonstrating how to use its generative video platform to synthesize high-fidelity, surreal environmental transitions. The guide details workflow parameters and prompting strategies to generate dreamlike visual doorways opening into detailed cloud landscapes. This release showcases the model's capabilities in temporal consistency, spatial coherence, and cinematic video synthesis, targeting creative professionals. (source: https://x.com/Kling_ai/status/2078495005241700399)

Kling AIGenerative VideoAI ArtVideo SynthesisCreative WorkflowArtificial IntelligenceGenerative AIComputer Vision
05

Experience Interactive Sports Creativity With Magnific and Kling AI Collaboration

Kling AI and Magnific have partnered to launch an interactive content generation workflow tailored for sports-themed media creation. The integration combines Kling's generative video technology with Magnific's high-resolution enhancement capabilities, allowing digital creators, marketers, and enthusiasts to generate and edit high-fidelity, customized sports imagery and dynamic video sequences. (source: https://x.com/Kling_ai/status/2078330283590660156)

MagnificKling AIGenerative AISportsCreative ToolsArtificial IntelligenceGenerative AIMultimodal