Another Crazy Day in AI: Study Finds Smart Tools Fall Short on Basic Time Tasks
- Wowza Team
- May 19
- 4 min read
Hello, AI Enthusiasts.
How’s your Monday night going? Maybe you’re still catching up from the weekend—or recovering from the first workday of the week. Either way, if you thought your brain felt a little foggy today, don’t worry: even the most powerful AI models are tripping over tasks we mastered in kindergarten.
A new study found that even top models struggle with reading analog clocks and figuring out calendar math. Turns out, some human skills are still surprisingly hard to replicate.
But for content creation, AI excels at idea generation and repurposing—if you know where to draw the line. Meanwhile, clinical AI is quietly improving patient care by assisting with scans and vitals.
Let’s see what the rest of the week holds.
Here's another crazy day in AI:
New study shows AI can’t reliably tell time
How to balance AI and human input in content strategy
What is clinical AI—and how is it used in care?
Some AI tools to try out
TODAY'S FEATURED ITEM: Why Smart AIs Still Can’t Tell Time
Image Credit: Wowza (created with Ideogram)
Can you tell time better than AI?
What seems like a basic skill for humans—telling time or figuring out what weekday a date falls on—still stumps even the most advanced AI models. A new study, featured at the ICLR 2025 Workshop and written by science journalist Drew Turney for Live Science, explores these surprising weaknesses in large multimodal AI systems. Based on research by Rohit Saxena and colleagues at the University of Edinburgh, the study tested how well leading AI models handle tasks like reading analog clocks and understanding calendar dates—and the results are eye-opening.
Despite excelling at language, image generation, and even passing professional exams, these models consistently fail at interpreting clocks and calendars—tasks most kids master by elementary school. The research points to key gaps in spatial reasoning and logical arithmetic—skills crucial for AI to operate in real-world, time-sensitive applications like scheduling, automation, and assistive tools.
Where the models struggled:
Clock reading accuracy was low across the board—even simple, standard clocks were misread.
Designs with Roman numerals or missing second hands made the task even harder for models to parse.
Calendar questions involving lesser-known dates or those requiring date calculation led to frequent errors.
One model (GPT-o1) performed relatively well on calendar tasks, but others showed inconsistent or near-random results.
Rather than computing answers, models often relied on pattern prediction from training data—leading to logic gaps.
Tasks that blend visual interpretation with structured reasoning continue to expose AI’s current limitations.
These findings offer a useful window into how today’s AI models process—and sometimes misprocess—visually grounded tasks that require structured logic. It's not just a matter of recognizing what's in an image, but being able to apply a sequence of reasoning steps to interpret it meaningfully. Reading a clock, for instance, involves understanding the spatial relationship between hands, accounting for visual variations, and then translating that into a precise time. Similarly, calendar reasoning often demands date arithmetic, which doesn’t come naturally to models that rely on predicting likely answers over calculating exact ones.
As AI becomes more embedded in everyday tools and workflows, understanding these limitations matters. Not to cast doubt on its capabilities, but to develop better expectations around what these systems can and can’t do yet—and where human oversight remains essential. This study underscores the importance of testing AI on the kinds of small, specific tasks that reflect how we actually use technology day to day. In that context, even something as ordinary as telling time becomes a meaningful benchmark.
Read the full article here.
Read the full paper here.
OTHER INTERESTING AI HIGHLIGHTS:
How to Balance AI and Human Input in Content Strategy
/Adam Tanguay, Head of Growth at Jordan Digital Marketing, on Search Engine Land
AI can speed up content creation, but it can’t replace human creativity or strategic thinking. Adam Tanguay outlines a scalable, client-ready SEO content process that uses AI where it excels—ideation, structuring, and repurposing—while keeping humans in charge of originality, accuracy, and brand voice. His agency builds a custom GPT per client, feeding it brand-specific data to streamline workflows and scale content without sacrificing quality. The result: more efficient teams, better SEO, and content that still feels human.
Read more here.
What Is Clinical AI—And How Is It Used In Care?
/Madeleine Streets, Senior Content Manager, on TechTarget
Clinical AI refers to AI systems used directly in medical care—like helping doctors interpret scans, monitor vitals, and support treatment decisions. Unlike broader healthcare AI, which includes billing or marketing applications, clinical AI is focused on improving patient outcomes in settings like hospitals and urgent care. It augments the work of doctors and nurses by spotting early signs of illness, tracking patient data in real time, and automating documentation. While still evolving, clinical AI is becoming a crucial tool in hands-on patient care.
Read more here.
10 applications of AI in healthcare
SOME AI TOOLS TO TRY OUT:
That’s a wrap on today’s Almost Daily craziness.
Catch us almost every day—almost! 😉
EXCITING NEWS:
The Another Crazy Day in AI newsletter is on LinkedIn!!!
Leveraging AI for Enhanced Content: As part of our commitment to exploring new technologies, we used AI to help curate and refine our newsletters. This enriches our content and keeps us at the forefront of digital innovation, ensuring you stay informed with the latest trends and developments.
Comentarios