Tinius Digest March

Large language models (LMM) and omniscient digital assistants will revolutionise how people discover and consume content, challenging existing platforms and requiring new thinking about journalism’s production, distribution, and commercialization.

LLMs are developing fast, changing the workflow of journalists and requiring entirely new technical skills.

Smaller media companies can now use new technology to provide automated hyperlocal journalism, even at the neighbourhood level.

New technology risks, revenue pressures, and stagnant subscription rates will likely result in more and stricter paywalls. As a result, people’s access to reliable journalism will be further limited, which could have socioeconomic and political ramifications.

Free or cheap access to new tools and technology creates opportunities for growth for new and innovative media companies. Large media companies have more significant opportunities to use and develop new tools but are often more sceptical of substantial changes.

Headlines containing negative language were significantly more likely to be clicked on than headlines containing positive language. For a headline of average length, each additional negative word increased the click-through rate by 2.3 percent.

Basic emotions such as ‘sadness’ increased the likelihood that a news headline was clicked on, while ‘joy’ decreased the possibility that a news headline was clicked on.

Even though Upworthy branded itself as a ‘positive news outlet’, the study uncovered a negativity bias in the data, which suggests that negative language increased news consumption while positive language decreased it.

Negative words have a stronger impact on consumption rates for news about “Government & Economy,” according to the study, suggesting that people are more inclined to read negative news in this area.

GPT-4 is found to be part of a new cohort of LLMs that ‘exhibit remarkable capabilities across a variety of domains and tasks’. The researchers believe GPT-4 and Google’s PaLM are ‘challenging our understanding of learning and cognition.’

GPT-4 can solve novel and complex tasks across various domains, including mathematics, coding, vision, medicine, law, psychology, and more, without special prompting. Its performance is strikingly close to the human-level performance.

Even though GPT-4 is seen as an early version of an artificial general intelligence (AGI) system, the researchers underline that there are considerable challenges ahead in advancing towards more profound and comprehensive versions of AGI. 

The report reflects on potential societal influences of the technology but finds that positive and negative impacts can’t be known in advance given the uncertainties about the use cases and applications created and the practices established within and across sectors.

Hate speech is most often directed towards Muslims (29.7%) and women (18.8%). These are the percentages of 40,894 comments that were defined as hate speech. 

Only 0.4 percent of Norwegian Facebook comments include hate speech. These comments include attacks directed towards, e.g. race/ethnicity, religion, sexual orientation, social class, and disability.

177,077 comments (1.7 percent) were identified as linguistic attacks. These comments contain stigmatising, derogatory, offensive, stereotypical, exclusionary, harassing, or threatening expressions. Discussions about the criminal justice system (4.9 percent) and crime (4.2 percent) most often trigger linguistic attacks in Facebook comments.

Of the 20 politicians with the highest percentage of linguistic attacks in comments on their Facebook pages, 12 are from the Norwegian Progress Party, three are from the Norwegian Green Party, two are from the Norwegian Liberal Party, two are from the Norwegian Democrats, and one is from the Norwegian Conservative Party.

In the study, participants correctly identified 65 percent of fake news when asked about both their likelihood to share the story on social media and their perception of its trustworthiness. However, when participants were asked about trustworthiness before the likelihood of sharing, they correctly identified 82 percent of the fake news.

Simply considering sharing news on social media reduces the extent to which people discriminate truth from falsehood when judging accuracy.

The study results suggest that people may be particularly vulnerable to believing false claims on social media, given that sharing is a core element of what makes social media ‘social.’

Norwegian political Twitter is not an isolated echo chamber. It consists of interlinked communities with different ideologies. The more polarised communities aren’t more uniform or partisan, suggesting that those factors don’t drive polarisation on Norwegian Twitter.

The degree of ideological polarisation differs across communities and topics, with some topics, such as political hate and far-right politics and economy and taxes, being more polarised than others.

The research challenges the idea that social media inevitably fuels polarisation. Instead, it shows that polarisation on social media is connected to specific contexts and topics. Norway’s political system and culture may encourage cross-ideological engagement and reduce polarisation on social media. However, there’s still a divide between mainstream and radical-right politics on Norwegian Twitter that can lead to polarisation.

Newspapers’ reduced revenue and scope have consequences: In 2022, some 35 of Sweden’s 290 municipalities lacked regular local news coverage, raising concerns about citizens’ access to local news.

Younger generations gravitate towards streaming services, podcasts, and online social networks. Older generations (45+ years) still prefer newspapers, with digital newspapers being the only online medium more popular among this demographic.

Advertising market becoming increasingly digital: In 2021, 69 percent of Swedish advertising investments went to online advertising, totalling SEK 32 billion out of SEK 46 billion.

72 percent of online advertising investments in Sweden went to global search engines and social media platforms in 2021. Decreased advertising revenue and shrinking readership lead to cost-cutting measures and challenges for local and regional newspapers.