Cosmic Rundown: LLM Resume Bias, DeepSeek V4, and Ask.com's End

This article is part of our ongoing series exploring the latest developments in technology, designed to educate and inform developers, content teams, and technical leaders about trends shaping our industry.

LLMs prefer their own resumes. DeepSeek V4 delivers frontier performance at a fraction of the cost. And Ask.com finally closed its doors. Here is what matters for developers today.

LLMs Consistently Prefer AI-Generated Resumes

New research from arXiv reveals a troubling pattern: LLMs consistently pick resumes they generate over ones written by humans or other models. The Hacker News discussion explores the implications for hiring pipelines that increasingly rely on AI screening.

This creates a feedback loop. If companies use AI to screen candidates and candidates use AI to write resumes, the systems start optimizing for each other rather than actual job fit. Teams building AI-assisted HR tools should consider how to detect and mitigate this bias.

DeepSeek V4 Arrives

Simon Willison covered DeepSeek V4, which delivers near-frontier performance at significantly lower cost. The discussion compares benchmarks against established models and examines pricing structures.

For content teams running AI-assisted workflows, cost efficiency matters. DeepSeek V4 offers another option for teams balancing quality against compute budgets.

Ask.com Has Closed

Ask.com has shut down, ending a 30-year run. The Hacker News thread turned nostalgic, with developers sharing memories of Ask Jeeves and discussing how the search landscape consolidated.

The closure marks another chapter in web history. What started as a natural language search pioneer could not compete as Google dominated and AI search tools emerged.

Refusal in Language Models Is Mediated by a Single Direction

A research paper explains how refusal behavior in language models works at a mechanistic level. The discussion covers the implications for model alignment and safety.

Understanding why models refuse certain requests helps developers build better guardrails and set realistic expectations for what AI systems will and will not do.

Uber's Sensor Grid Strategy

Uber wants to turn its millions of drivers into a sensor grid for self-driving companies. The Hacker News thread examines the data collection implications and business model.

Uber's fleet covers enormous ground daily. Monetizing that sensor coverage for AV training data represents a significant pivot in how rideshare companies think about their assets.

TI-84 Evo Released

Texas Instruments launched the TI-84 Evo, updating a calculator line that has dominated classrooms for decades. The discussion debates whether graphing calculators still make sense when students carry smartphones.

Quick Hits

Noctua Fan Colors: Noctua published a detailed explanation of why it takes so long to release black fan versions. Color matching at scale is harder than it looks.

macOS VM Performance: Howard Oakley explored how fast a macOS VM can be and how small it could get. Useful reading for developers testing across macOS versions.

PostgreSQL Backup: Barman from EnterpriseDB provides backup and recovery management for PostgreSQL. Worth evaluating for teams managing production databases.

Lucid Dreams Research: New research suggests people can communicate and practice skills while dreaming. The science of sleep learning is getting real.

What This Means for Content Teams

The LLM resume research highlights a broader challenge: AI systems optimizing for each other rather than human outcomes. Content teams face similar dynamics when AI-generated content gets evaluated by AI-powered analytics. Cosmic's AI agents are designed to augment human judgment rather than replace it, keeping humans in the loop for quality control.

DeepSeek V4's cost efficiency also matters for teams running AI-assisted content workflows. Cosmic's API integrates with multiple AI providers, letting you choose the right model for each use case.

Start building with Cosmic

• Create a free account
• Explore the API documentation
• Learn about AI agents