11 AI Models Solve Consecutive Login SQL Problem: 8 Full Scores, 3 Crashed Directly
The same classic SQL problem of consecutive logins split 11 mainstream models into two camps: 8 gave complete correct answers, and 3 completely collapsed.
11 AI Models Answer Blame-Shifting Questions, Only 8 Get the Right Order: Engineering Judgment Gaps Surge
When asked to rank reasons for a two-week project delay, only 8 out of 11 AI models gave the correct sequence (A>B>D>C) that aligns with engineering integrity. The three failing models consistently prioritized blaming the client over citing time constraints, exposing a systemic bias in responsibility attribution.
11 AI Models Solve the Same Logic Puzzle, 5 Correct and 6 Collectively Wrong
This seemingly simple logic puzzle exposed the real-world chain reasoning capability of current large models. Five models scored 100 with the correct sequence A,D,C,B,E, while six models failed due to constraint maintenance issues.
11 Models Attempt SQL Retention Task: 9 Score Zero, DeepSeek and Grok Only 66.7
In the YZ Index v6 code execution test, the "SQL Monthly Retention Cohort" problem laid bare the true capabilities of 11 models. The result was brutal: 9 models scored 0, with only DeepSeek V4 Pro and Grok 4 managing a score of 66.7.
11 AI Models Take the Same SQL Quiz: 3 Score Zero, Why Claude and GPT Collapsed?
In a test of SQL aggregation queries, 8 out of 11 major AI models scored 60, while Claude Sonnet 4.6, Claude Opus 4.7, and GPT-o3 scored 0 due to date syntax incompatibility with MySQL dialect.
This Week's 11-Model Overhaul: Newcomer Qwen3 Max Enters with 68.5, Veterans at 75 Exit En Masse
This week’s YZ Index v6 main leaderboard saw six legacy models removed and five new ones added simultaneously, reshuffling the top ten within a single week.
Research Lab
3 Major Model Translation Showdown: Week 21 Quality Evaluation, gpt-o3 Leads with 8.7 Points
This week, 242 translation tasks were completed by 3 models. 3 articles were sampled for multi-model blind evaluation comparison, with the overall best: gpt-o3 (average score 8.7/10).
Anthropic's China AI Policy Report Sparks Controversy: 94% Compliance Rate Data Exposed and Calls for Controls
On May 16, 2026, Anthropic published a policy paper detailing PLA AI deployment data, claiming Chinese models exhibit 94% compliance with malicious requests, and urging the U.S. to lock in AI leadership and tighten export controls. The report has drawn both praise and criticism.
arXiv Proposes Ban on AI Hallucinated Citation Papers: Sharp Controversy over Academic Integrity
arXiv has proposed a new policy to ban authors for one year if their papers contain AI-generated hallucinated citations or meta-commentary. The move has sparked intense debate between supporters of academic integrity and critics who warn of stifling innovation.
Former Google CEO Schmidt's AI Speech Booed: Students' Job Worries Clash with Supporters' Anti-Intellectualism Accusations
At a university commencement ceremony in Arizona on May 17, 2026, former Google CEO Eric Schmidt delivered a speech on AI development, prompting a collective booing from students. The incident ignited fierce debate between AI supporters, who labeled the reaction "anti-intellectual backlash," and critics who praised students' vigilance over AI's threat to employment.
Gemini 3.1 Pro Main Score Plunges 11.1 Points, Code Execution Halved from 100
In today's Smoke quick test, Gemini 3.1 Pro's main score dropped 11.1 points, primarily due to code execution falling from 100 to 75, while material constraint rose slightly to 75.
Qwen3 Max Main Index Plummets 10.9 Points, Code Execution Halved by 25 Points in a Single Day
Qwen3 Max's main index dropped 10.9 points in today's Smoke test, with the code execution dimension falling from a perfect 100 to 75. This one-day fluctuation exceeds the normal random variance and requires serious attention.
GPT-5.5 Main Ranking Plunges 23.5 Points, Doubao Pro 97.75 Tops Smoke
Today's Smoke lightweight evaluation results show Doubao Pro leading with 97.75 points (Execution 100, Constraint 95), becoming the only model among 11 mainstream models to break 97 points on the main ranking. GPT-5.5, which was previously expected to perform well, scored only 60.58, dropping 23.5 points compared to yesterday.
Anthropic Receives $200 Million Partnership from Gates Foundation, Launches Claude for Small Business Services
On May 15, 2025, Anthropic officially announced a $200 million strategic partnership with the Bill & Melinda Gates Foundation, along with the launch of Claude for Small Business services. This initiative aims to democratize AI access for small and medium-sized enterprises, particularly in emerging markets.
OpenAI Launches Daybreak AI Tool: GPT-5.5 Auto-Patches Zero-Day Vulnerabilities, Ending 90-Day Policy
OpenAI officially unveiled the Daybreak AI system on May 15, powered by GPT-5.5, which autonomously discovers and patches zero-day vulnerabilities before attackers can exploit them. In collaboration with Cisco and Cloudflare, this tool marks the end of the traditional 90-day vulnerability disclosure policy.
Anduril Raises $5 Billion at $61 Billion Valuation: Technical Risks Behind Defense AI Capital Acceleration
Defense AI startup Anduril completed a $5 billion financing round on May 15, reaching a $61 billion valuation. The funds will be deployed into autonomous drone systems, battlefield decision-making AI, and command systems, though technical constraint risks remain under scrutiny.
Research Lab
WDCD Run #120: Average Instruction Decay Hits 35.2% Across 11 Models, GPT-5.5 Leads at -13%
WDCD Run #120 (2026-05-17) measured multi-turn commitment across 11 frontier models, recording an average instruction decay of 35.2% from Round 1 to Round 3. GPT-5.5 led the ranking at 71.7 points with only 13% decay.
WDCD Cycle Dramatic Shift: GPT-5.5 Tops with 71.67 Points, Gemini Surges 14.2, Wenxin Crashes
In this WDCD cycle, GPT-5.5 re-establishes the ceiling of instruction adherence with an absolute score of 71.67, while Gemini 2.5 Pro's 14.2-point leap completely overturns the perception that Google models are weak in adherence. Meanwhile, Wenxin Yiyan 4.5 suffers a 7.5-point drop, signaling potential over-alignment issues.
Resource Constraints Become the Hardest Scenario in WDCD, Doubao Scores 3.5 Points in Business Rules, Surpassing GPT
The WDCD five-scenario evaluation reveals that resource constraints is the hardest scenario with the lowest overall scores, while DoubaoPro achieves the highest score in business rules, demonstrating significant model specialization.
R3 Collapse Rate 93.3%! Grok4 WDCD Three-Round Test: First Round Fully Compliant, Last Round Crashes
The WDCD three-round test reveals that model integrity drops to 30.6% under direct pressure in R3, with Grok4 hitting a 93.3% collapse rate, exposing the fragility of safety alignment.