AI DETECTORS Researchers led by Assistant Professor of Biomedical Data Science at Stanford University Dr. James Zou tested seven popular GPT detectors using 91 English essays written by non-native English speakers who took the Test of English as a Foreign Language (TOEFL).

Alarmingly, more than half the essays were inaccurately identified as AI generated – with one detector flagging nearly 98 percent of them as machine written. In comparison, the detectors correctly classified over 90 percent of essays written by Grade 8 students in the US as human generated.

The detectors rely on evaluating text perplexity, which measures surprising word choices in an essay. As a result, non-native English writers, who tend to use simpler vocabulary, are more susceptible to being wrongly labelled as AI generated.

To demonstrate this vulnerability, researchers edited the human generated TOEFL essays by employing more sophisticated language, which led the detectors to classify them as human written.

Given these findings, researchers urge extreme caution in the use of GPT detectors, particularly for reviewing job applications, college entrance essays or high school assignments.

Biases and susceptibility to minimal prompt manipulation highlight the unreliability of the detectors. The impact extends beyond the education sector and potentially silences non-native English writers in search engine rankings too.

STRESS DETECTION In an experiment, researchers observed that stressed individuals exhibit distinct patterns in their typing and mouse movements compared to their relaxed counterparts.

Stressed folk tend to move the mouse pointer more frequently and less precisely, covering longer distances on the screen. They also make more typing mistakes with fragmented and pause filled writing. Conversely, relaxed individuals follow shorter and more direct routes with the mouse, and take fewer but longer pauses when typing.

The study reveals that typing and mouse behaviour proved to be more reliable indicators of workplace stress than heart rates. This challenges conventional stress monitoring methods and highlights the potential of leveraging technology to provide early stress detection during work.

Researchers are currently testing data collected from Swiss employees using an app that records mouse and keyboard behaviour along with self-reported stress levels. Data protection and privacy are of paramount concern since the aim is to develop an anonymous and secure tool, to assist workers in recognising and managing stress, rather than creating a monitoring system for employers.

BACTERIA BATTLE According to a study conducted by Professor of Pathology Dr. James Kirby and his team at Harvard Medical School, nourseothricin – which is an 80-year-old antibiotic – could offer a new approach to combatting drug resistant bacterial infections.

Nourseothricin, which is derived from a type of soil fungus, consists of different forms of the complex molecule streptothricin. Initially discovered in the 1940s for its efficacy against gram-negative bacteria, the antibiotic was abandoned due to its toxic effect on the kidneys.

However, the rise of antibiotic resistant infections has prompted a reevaluation of nourseothricin.

The study focussed on highly purified forms of streptothricin – particularly Streptothricin F, which showed strong activity against drug resistant gram-negative bacteria. By binding to a bacterial ribosome subunit and inducing translation errors, this antibiotic offers a unique mechanism of action against such infections.

Using cryo-electron microscopy, researchers revealed that Streptothricin F extensively binds to the bacterial ribosome and causes translation errors, which ultimately lead to bacterial cell death. The antibiotic’s interaction differs from other known translation inhibitors, making it potentially valuable when other agents are ineffective.

LOLLIPOP SWABS Throat swabs have long been the go-to method for collecting samples to diagnose various illnesses including strep throat. However, scientists are exploring alternative and more pleasant methods such as saliva sampling, particularly for home testing.

In a bid to make saliva collection less stressful, researchers have developed a lollipop-based device called CandyCollect to combine the act of sampling with the delight of savouring a sweet treat.

CandyCollect features a spoon-like stick with a spiral shaped groove on top, covered with ‘isomalt’ candy. As the lollipop is consumed, saliva effortlessly flows into the groove, facilitating easy sample collection.

In previous lab tests, the device successfully captured strep throat causing bacteria. To expand its applications, researchers sought to compare CandyCollect with other commercially available at home saliva sampling methods and evaluated its efficacy with real participants.

Remarkably, CandyCollect consistently detected the target bacteria whenever the conventional methods did, achieving a 100 percent detection rate.

While ongoing studies continue to explore its potential, the adaptability and positive reception of the CandyCollect system are promising and could revolutionise the manner in which throat samples are collected for analysis.