Claude Code Beats Codex CLI 3-0 in Controlled Liar's Dice AI Showdown

·1 views

A developer pitted two AI coding agents — Anthropic's Claude Code (Opus 5) and OpenAI's Codex CLI (gpt-5.6-sol) — against each other in three best-of-3 series of Liar's Dice, with Claude winning all three series 2–0. The experiment used a custom rules engine connected to both agents via separate MCP servers, designed to prevent cheating through strict information hiding and token-bound seats. Claude's challenge calls succeeded 8 out of 11 times, while Codex's hit only 4 out of 26, highlighting a significant gap in bluff-detection accuracy. The author built in safeguards such as commit-reveal dice hashing, optimistic concurrency controls, and decision-time belief logging to ensure the results were verifiable and tamper-proof. The author emphasizes this is not a general model ranking, but a narrow test of probabilistic reasoning under hidden-information conditions.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Nawtch Adds Bowl Builder Tool With Separate Save Paths to Protect Log History

A developer has built a Bowl Builder feature for Nawtch, a privacy-focused nutrition tracking app, allowing users to assemble meal-prep style bowls and get real-time calorie and macro calculations. The tool is accessible to non-Nawtch users as well, with bowls stored locally in the browser's IndexedDB without any account required. A key design challenge was separating the 'Save Bowl' and 'Add to Foods' functions so that editing a bowl recipe does not retroactively alter previously logged meal entries. Once a bowl is added to foods, it becomes a permanent snapshot, ensuring historical calorie logs remain accurate even if the original recipe is later modified. The Bowl Builder has been placed under Nawtch's existing Tools section, following the precedent set by Scorekeeper, another standalone utility within the app.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Graph Engineering Trend Was Built on a Joke Nobody Fact-Checked

A single 12-word tweet by OpenClaw creator Peter Steinberger on July 18, 2026, mocking how quickly the AI field renames concepts, inadvertently sparked a wave of explainers declaring 'loop engineering' dead and 'graph engineering' the new paradigm. LangGraph creator Harrison Chase said he did not know what graph engineering was, and later co-authored a piece clarifying that loops are simply a basic form of directed cyclic graphs — making the supposed migration a non-event. The LangChain framework had already been running on LangGraph for three years, meaning any 'migration' many developers were planning had long since happened without fanfare. Anthropic had documented multi-step agent patterns as far back as 2024, and tools like AutoGen and Google ADK shipped graph workflows before the term trended. Notably, multiple articles misreported the tweet's word count and cited view figures ranging from 575,000 to 2.9 million, highlighting that a significant portion of the commentary was written without anyone verifying the original source.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

15-Year-Old Developer Seeks Advice After Losing Freelance Income to Scams

A 15-year-old self-taught developer has shared their struggle to earn money from programming after transitioning away from informal family-friend gigs. The teenager claims to have built full websites for startups but has since been repeatedly scammed and left unpaid for completed work. Their monthly income has dropped from a previous range of $300–$500 to nothing. Having tried platforms like Discord and Reddit without success, they are now questioning whether a career in programming is viable. The post asks the developer community for any practical advice on finding legitimate, paid work.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Frontend Teams Need a Pre-Release Checklist to Prevent Video Crop Errors

Video aspect ratio mismatches are a common but often overlooked engineering problem that affects how content appears across different devices and breakpoints. Errors typically originate in one of three places: a source file whose resolution does not match its delivery container, a transcoder configured for the wrong target aspect ratio, or a player styled with a hardcoded CSS aspect-ratio rule. These issues can cause stretched visuals, black bars, or silent clipping that goes undetected until the video reaches end users. A recommended pre-release QA checklist advises teams to inventory every video slot, verify source masters, explicitly set display aspect ratio during transcoding, and spot-check output on real devices at multiple breakpoints. Running this workflow before any release involving media assets is estimated to take around 20 minutes for a single landing page and up to an hour for a full onboarding flow.

0 comments Read more at DEV Community