Tickin' away

Good morning Staker! Here’s what’s cookin’ today: AI agents come to Wall Street; How your brain manages multiple languages at the same time; and, shocker, AI might be running out of material to learn from…👀

Let’s get into it!

Gomez Addams witnessed it every time Tish uttered something in French.

If you speak two languages fluently or know someone who does, you’ve likely experienced “mixing” before, in which even just one sentence is expressed while almost involuntarily going back and forth between each language.

Previous research has looked into what’s going on in the brain when a multilingual person is mixing. It found the brain is always operating both “engines” at the same time, and when speaking one language (let’s say Spanish), the prefrontal cortex is tasked with suppressing the other (let’s say English).

The activity in this region during speech for a multilingual person is similar to that of any other advanced cognitive task. Multitasking in general is a good way of looking at it, except in this case, the second task is a suppression rather than expression.

New research published in June, however, found an extra layer of brilliance in terms of what’s going on in the brain of a multilingual person while they’re speaking. Scientists used MEG scans—which observe brain activity at the millisecond level—to see what was happening inside the heads of Spanish-English speakers while they were mixing.

The scans revealed a single left-brain system was applying grammar rules, regardless of whether they were speaking English or Spanish.

In other words, while the languages themselves act as two separate engines, the capacity for the application of grammatical structure transcends the language itself and is seemingly an innate ability that works in tandem with the most efficient engine, explaining why multilingual speakers will go back and forth.

A novel virus like the one that causes COVID-19 takes years—if not decades—to be fully understood by scientists. This is no better exemplified than by the persistently elusive phenomenon known as “long COVID.”

While medical professionals continue to be stumped in terms of finding a concrete explanation for the post-COVID condition, they agree on its most persistent symptoms: fatigue and brain fog.

The problem with knowing symptoms and not their underlying cause is that it makes them difficult to treat. Scientists have had to think pretty far outside the box when it comes to treating long COVID, and that includes a recent study that prescribed fluvoxamine, metformin, or a placebo, to 399 Brazilian patients in the REVIVE-TOGETHER trail.

Fluvoxamine is an anti-depressant medication and metformin is used to treat diabetes.

Those who received fluvoxamine reported being less tired, along with positive results on multiple quality-of-life metrics. Metformin didn’t help remedy symptoms for those already experiencing them.

Still, it’s difficult to determine why fluvoxamine helped or if it was actually the cause of the symptom alleviation, given its typical use in treating depression.

It notably did not cure long COVID, but its use in treating it should be further studied, according to the researchers.

Large language models need almost unimaginable amounts of data to continue growing their intelligence, but Stanford University’s 2026 AI Index Report suggests they may be running out, having already trained on most of the existing data on the internet.

“AI researchers have publicly claimed that the available pool of high-quality human text and web data for training large models has been exhausted, a state often referred to as ‘peak data,’" the authors said.

The potential solution could lead to the mother of all diminishing returns: synthetic data, in which new data to train on is itself generated by the models that need to train on it.

Like an academic discipline that hampers its own validity by reducing the diversity of work cited in future studies, AI models could end up creating a homogenous echo chamber of large language code that becomes increasingly less sophisticated.

“…simply adding more data does not necessarily lead to performance gain. There is still no definitive evidence that synthetic data can fully offset real-data depletion in pre-training contexts,” the authors said.

In journalism, the “lede” is the opening section of a news story that’s designed to motivate the reader to continue reading. “Burying the lede” is an proclamation made when the most important or interesting part of the story falls into the main body and gets “buried” in favour of something less interesting or important.

It’s very much an industry term, though its usage has pierced the veil between newsrooms and living rooms over time.

Still, the better question is probably why it’s spelled that way rather than what it means.

Simply put, “lead” is a homograph, in that it’s actually multiple words spelled the same but with different pronunciations that mean different things. For example, the heavy metal “lead” is pronounced led, whereas being first in something is to “lead” or lede.

In newsrooms in the 1970s, Linotype machines were still in use, which used thin strips of metal to separate the lines of type on a page. The strips were made out of lead, and in a newsroom, a story’s lead feature was always being discussed. In order to minimize confusion, newsrooms started using phonetic spelling for the latter.

It was also right around this time that Linotype machines were being phased out, so the pervasiveness of the “lede” spelling has often been attributed to a sense of nostalgia and even elitism within major newsrooms.

A couple of bad eggs

There’s nothing like a great bad guy, but are the villains as memorable as the heroes? Find out with today’s classic movie villains trivia 😏