Jill Walker Rettberg
@jilltxt
Co-Director at Center for Digital Narrative, Univ of Bergen. Lead research project AI STORIES (ERC Advanced Grant). Australian and Norwegian. Blogging at jilltxt.net since 2000.
Scientists from all disciplines are welcome. That would be the discipline "other - other" for many of us here, but still important to sign. The signatory list will be presented to the UN Secretary-General. scientistswarning.forestry.oregonstate.edu/sites/defaul...
And finally, the article starts by introducing the idea of semi-aura, which is a term they coin. That's not usually the function of a systematic review.
So one major issue here is that the "rigorous methods" they apparently scored for are not generally appropriate for research on Benjamin's concept of aura in relation to AI art. Sample size? Response rates? These are methods for medicine and behavioural sciences, not art theory or humanities.
Each included article is given a CASP score. The problem is that CASP has many different checklists for different types of study and the paper doesn't explain which they used. They do share the scores they assigned to each included paper.
They include tables of the 48 included articles in their appendices, and this is very useful for other scholars. Here's two screenshots as a sample. Publication as a dataset on Dataverse or similar would be good, these need to be copy-and-pasted from the appendix. link.springer.com/article/10.1...
They excluded articles that lacked "methodological rigor" (that might be fine, but what rigour are they looking for given the topic is art theory - theoretical or empirical rigour?) and so on. But included some articles that weren't peer-reviewed, which seems odd.
This article tries to use the method but does it halfway. They search for articles using strings "such as 'Art + AI' and 'Aura + AI'" - nope, we need the exact search. The exclusion criteria are kind of vague (no code book is shared to show how decisions were made) and only one person screened.
Based on interviews with kids in Oslo, Emilie Owens argues that brainrot is "a decompression-driven genre of participation whereby young people actively resist the pressures of productivity and self-optimisation". journals.sagepub.com/doi/epub/10....
In another case we analyse, the user is positioned as the lone genius with ChatGPT as superhuman helper. But in all these narratives the evaluation phase is omitted. Unlike a movie or bedtime story, the fictionality is obscured, there is no retroactive telling, no reason for telling the story.
Masterplots are key to affirmative narration. When the user starts to lose interest in one masterplot, the chatbot switches to another. Suicidal thoughts are renarrated as stories leading to poetically tragic endings, or to heroic warrior deaths after a long-fought battle.
We analysed chatlogs from the tragic suicides where families sued OpenAI, meaning that excerpts from the victims' chats with ChatGPT have been made public. We argue that narrative strategies are repeatedly used, in particular what we call affirmative narration. Sycophancy is a narrative strategy.
- web search without AI summary: 0,3 Wh - short text query to LLM: 0,42 Wh - AI-generated image: starts at 2,9 Wh - web search with AI summary: 3,0 Wh - AI-generated video: starts at 415 Wh www.nrk.no/urix/fn-rapp...
Ja, det er ærlig talt litt rart at dette er åpningssskjermen til en jobbapp. Den er vel ikke lenger en jobbapp.
I dag brukte jeg tre ganger så lang tid på en enkel jobboppgave (scanne kvittering på reise til reiseregning) fordi Microsoft sin 365 app har erstattet den enkle scanne funksjonalitetenen med en praterobot som også gjerne vil at jeg skal lage korte videoer. Filene mine finner jeg heller ikke.
I'm in Berlin to give a talk this evening at the Peter Szondi Insitut at Freie Universität - if you're in Berlin, please come and join the discussion! www.geisteswissenschaften.fu-berlin.de/we03/institu...
Another paper also cites my AI-generated podcast paper, and correctly, so that's nice - and very fast. That paper is about using AI-generated podcasts in higher education - authors found presentation of their papers was 91% accurate but only 43% balanced. doi.org/10.1080/2997...
Doesn't this make it look as though I argue that AI can augment human creativity? That is THE OPPOSITE of what I write. Maybe you could interpret it as saying "Rettberg says that some say this, others that". I'm pretty sure the whole article was written by Grammarly/ChatGPT.
Pangram har bare 1 av 10,000 falske positiver, den er rimelig treffsikker. Og helt enig med det du lett så med menneskelig lesning.
Seriøst, @altinget.bsky.social? Robotgenererte svar på spørsmål som framstilles som om de er fra faktiske politikere og dere «tar forbehold om feil?» Hva er egentlig poenget med å betale for journalistikk om dette er hva dere leverer? Nei takk. www.altinget.no/digital/stat...
Remarkable and detailed description of a chat between a junior lawyer and a chatbot where the bot generated letters to the court with fake quotes of laws, told the lawyer to verify, lawyer still didn’t. There are so many stories like this now… caselaw.nationalarchives.gov.uk/ewhc/ch/2026...
Also, maybe others could use (or adapt) Teknologirådets methods here. Here are the forms they used to log their use of LLMs over two 6 week periods. These are in the appendix. teknologiradet.no/publication/...
The LLM did a terrible job at converting a project description to a presentation. There are quotes of some of the horribly superficial writing it produced (p 25). It also "took extensive freedoms" by inserting the estimated cost for a project group although this wasn't in the project description.
Lin Prøitz introduces the final speaker, Daniella Gáti. LLMs use statistics to generate “likeliness”, the likely answer. Which we queer people recognise as being normative - reasserting average language. You can ask for a story from a non-heteronormative view, you’ll get the average version.
Is it possible to use AI for queer uses in a way that is good for queer people?
Soon-to-be-published speculative fiction by Oliver Haimson: “AI Powder - just sprinkle some AI on it!”
CDN’s Pride event has started!! First up is Joakim Johansen Østbye about queer gaming experiences in Norway. If you’re in Bergen we’re at Kvarteret (Pride House!) until about 21:00.
I går fikk jeg tre KI-genererte e-poster fra folk som vil ha noe fra meg. Jeg svarte på den tredje med et «automatisk» svar. Jeg lurer på å bare fortsette å late som om jeg har KI-filter så jeg slipper å være kjip - jeg bruker lenger tid på å tolke disse e-postene, jeg er møkk lei.
På tide å minne om hvordan KI «oppskalerer» bilder. Obama-eksempelet er fra 2020 men prinsippet har ikke endret seg - et menneske ser lett at det lavoppløselige er Obama. KI legger derimot til en statistisk prediksjon ut fra helt andre bilder, og Obama blir hvit. www.theverge.com/21298762/fac...