Drug Clinical Trial Failure Rates Examined Across Decades

·1 views

A blog post published on Science.org examines clinical trial failure rates over multiple decades. The piece appears to analyze trends in how often drug candidates fail during the clinical development process. The title suggests the findings are notably concerning or surprising. Insufficient article content was provided to summarize specific data points or conclusions accurately.

Read the full story at Hacker News

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

Discussion (0)

Developer Builds Industrial AI Agent That Escalates to Humans When Evidence Is Weak

A developer has built a real-time anomaly monitoring system for a simulated water-treatment plant that prioritizes evidence quality over speed of response. The system streams live sensor data from six industrial assets, detects unusual behavior, and queries a structured knowledge base before deciding whether to recommend an action or escalate to a human reviewer. A key design choice distinguishes data-quality issues, such as duplicate events or sequence gaps, from genuine mechanical faults, preventing false maintenance recommendations. SigNoz is used to make the entire decision pipeline observable, allowing engineers to trace each step from raw sensor input to final agent output. The project aims to reduce alarm fatigue by grouping related anomalies into a single investigation rather than generating a separate alert for every irregular reading.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Undergrad Argues Natural Language Is Structurally Flawed for AI Governance

A Chinese undergraduate building a public AI governance thesis found that AI agents cannot independently verify whether they have followed instructions, because generation and self-evaluation run through the same token-prediction mechanism. He initially reduced rule violations from 55.9% to 0.7% using external mechanical checks, but later concluded that better prompting cannot fix the deeper issue. He terms this limitation the 'Prose Barrier,' a finding he says was independently reached by a German developer as well. His proposed solution involves routing different types of constraints through separate layers: executable code outside the model, syllogistic structures optimized for transformer attention, and gradient-based weight updates via Direct Preference Optimization for persistent failures. The core argument is that natural language should not be the sole medium for AI governance, and that each constraint type requires the appropriate computational language.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Two-AI Debate System Catches Code Bugs That Single Reviewers Miss

A new code review approach deploys two specialized AI agents to analyze the same code change through structured debate, rather than relying on a single automated tool. One agent focuses on correctness and security, while the other evaluates developer intent and performance trade-offs. In a demonstration, both agents reviewed a Python caching function that passed standard linting checks, yet uncovered a hidden concurrency flaw — the cache lacked thread safety, risking race conditions and data leaks in multi-threaded web servers. The method, called AI pair review, uses disagreement between agents to surface context-dependent bugs that deterministic rule-based tools typically overlook. Proponents argue this consensus-driven approach better reflects the nuanced, intent-aware nature of real-world software development.

0 comments Read more at DEV Community

ProgrammingHacker News ·

GNU Hurd gains 9pfs, OpenNTPD, and dynamic /dev/ support in latest update

GNU Hurd, the long-developed free software kernel by the GNU Project, has received a notable set of new features and improvements. The update introduces support for 9pfs, a network filesystem protocol, along with integration of OpenNTPD for time synchronization. Dynamic /dev/ entries have also been added, improving how the system manages device nodes. These additions represent continued incremental progress on Hurd, which has historically lagged behind Linux in development pace. Full technical details are available via OSNews, which covered the release.

0 comments Read more at Hacker News