Tinius Digest April

The industrial sector now leads in AI innovation, surpassing academia. Pre-2014, academia produced most major machine learning models, but by 2022, the industry had created 32 versus academia’s three. Developing advanced AI systems requires extensive data, computational power, and funding—resources more readily available to the industry than academia and nonprofits.

AI consistently achieves cutting-edge results, yet annual progress on numerous benchmarks is minimal. The rate of reaching benchmark saturation accelerates, but comprehensive testing suites like BIG-bench and HELM emerge to address these limitations.

The database reveals a 26-fold increase in ethical AI misuse incidents since 2012. This surge reflects the growing adoption of AI and heightened awareness of potential abuse.

AI adoption among companies, having more than doubled since 2017, now stagnates between 50 per cent and 60 per cent, according to McKinsey’s annual survey. Despite the plateau, companies that have adopted AI continue to pull ahead: utilizing AI experience leads to significant cost reductions and revenue growth.

An AI Index analysis reveals growing policymaker interest in AI, with the number of laws mentioning ‘artificial intelligence’ increasing from one in 2016 to 37 in 2022 across 127 countries. AI mentions in legislative proceedings of 81 countries have also surged nearly 6.5 times since 2016.

25 per cent of Norwegians aged 20 to 24 find news on TikTok, and among those aged 16 to 19, the proportion is as high as 40 per cent. Snapchat has the largest user share of news among teenagers and young adults.

8 out of 10 Norwegians used social media daily in 2022, and just over half of the population uses social media to keep up with the news.

The media consumption of older individuals is becoming more digital, aligning with trends observed in the general population. Of those aged 67 to 79, 84 per cent use the internet daily, marking a notable rise from 2021—meanwhile, half of those aged 80 and above access the internet on a typical day.

Among 16-24-year-olds, half have actively chosen to follow traditional, editor-controlled news sources on social media by following newspapers, radio, and TV channels’ pages and profiles.

Three out of four people over 67 years old watch TV news broadcasts. Seven out of ten people over 80 years old read print newspapers daily, while one in four reads online newspapers. 43 per cent of those over 80 years old get daily news on the radio.

Time spent on the internet has increased every year since 2000. On average, the population spends four hours on the internet daily, an increase of 21 minutes from 2021. Internet usage time varies significantly between different age groups. Teens between 16 and 19 spend the most time online, averaging 7 hours and 14 minutes daily. In comparison, those over 80 spend only half an hour on the internet daily.

Children increasingly gravitate towards short and dramatic videos on social media platforms such as TikTok and Snapchat. Designed to maximize stimulation with minimal effort and focus, these videos captivate young viewers.

The growing trend of ‘split-screening’ allows children to watch two short-form videos simultaneously on a single screen. This behaviour seems to evolve from ‘multi-screening’, which refers to children’s difficulty focusing on one screen-based activity at a time.

Users aged 16-24 are more likely to take deliberate breaks from social media or delete apps to manage their well-being. This age group acknowledges the potential negative impact of excessive screen time and takes steps to set boundaries for themselves.

Children are less likely to post their videos, with only 32 per cent doing so compared to 96 per cent who watch them. This trend corresponds with the decrease in content created by friends and a decline in interaction with such content.

45 per cent of the content from the most popular influencers among Norwegian children and adolescents contains advertising. 17 per cent of the 812 analyzed posts had hidden advertising.

There is significantly more advertising in posts on YouTube (69%) than on TikTok (31%). 32 per cent of the posts contained exposure to violence, body, alcohol, sexual content, or tobacco.

Girls between 13 and 15 years old are more likely to be exposed to influencers who drink alcohol and to sexual or scantily clad content. Posts targeting boys more often expose animated violence, alcohol, or scantily clad bodies.

Boys between nine and 18 years old exclusively follow male influencers, while girls follow influencers of both genders. Everyday vlogging is the most popular content among girls, while gaming videos are most popular among boys.

ChatGPT determined if about 50,000 headlines were good, bad, or irrelevant news for companies’ stock prices. The headlines got a numerical score and found a positive correlation between these scores and subsequent daily stock market returns.

ChatGPT outperforms basic models like GPT-1, GPT-2, and BERT. More importantly, incorporating advanced language models into investment decision-making can yield more accurate results than traditional methods.

The study has far-reaching implications beyond stock market predictions and shows some of the controversial usages of AI—shifting market prediction and investment decision-making methods.

The research includes data from over 760,000 inventors—showing a trend where inventors increasingly choose to work for larger, established companies instead of smaller, newer ones. The trend also weakens inventors’ drive to create their businesses.

When inventors work for big companies, they make more money, but their ideas decrease. The study found that inventions produced by larger companies may not be as good as those produced by smaller companies.

The findings in the study demonstrate how incentive measures to increase the pace of innovation—especially those targeted at larger companies—can impact new ventures and decrease overall innovation power. 

Expectations can shape people’s perception and memory, leading to perceptual illusions and false memories. This tendency to create false memories works if people expect to see or read something in a certain way—and if people expect to remember something in a certain way.

Creating false memories can occur within a timeframe of just one or two seconds after perception, making it difficult to trace the source of the false memories.

Understanding these illusions in short-term memory is essential for understanding how people’s expectations and biases influence how they perceive what they read and see. The findings also have practical applications in eyewitness testimony—and implicit the trustworthiness of journalistic sources.