Gen AI links of the morning
2026-08-12 19:01:03.310253+02 by Dan Lyke 0 comments
Perplexity blocks Time’s ads served to AI agents, calling them ‘deceptive’. Apparently Time started serving pages as markdown for the consumption of AI agents, started serving ads in that markdown, and...
The ads work by having ad tech firm Mobian generate FAQ-formatted content based on a brand brief, then insert it into Time’s markdown pages and track how often AI search engines surface it, and how favorably. Ally Bank and the Project Management Institute had their brand-written, brand-approved messaging sit in the same markdown version of the page an agent reads as fact. If a page’s markdown version can be quietly seeded with paid content, users have no way of knowing whether an answer reflects a publisher’s reporting or a client’s brief, appears Perplexity’s logic.
One of the big problems with LLMs is that there is no control channel. Sure, the "system" prompt may get weighted more heavily, but everything is data.
Kinda like humans, come to think of it...🤔 [goes and updates ad blocker]
Via.
May used Meta AI to make the cover, which resurrects late frontman Freddie Mercury. He did so in order to make a “very serious” point about the environmental impact of AI to UK Prime Minister Andy Burnham.
But I'm now very serious. I believe the price we are about to pay for this 'inevitable' development is way too high. We must not allow these hideous Data Centres to be built in the UK.
Science Alert: 'Virtually Absent': AI Models Practically Erase Female Characters From Kids' Stories
In this paper, we use anthropomorphic fiction as a stress test for gender representation in contemporary LLMs. Across 23.8K story completions, we find that models frequently respond to gender ambiguity through avoidance, either by repeating the animal noun or using neutral pronouns. This surface-level neutrality does not translate into equitable representation. When the models do explicitly assign gender, they overwhelmingly default to masculine protagonists, while feminine protagonists are nearly absent. This experiment leads us to a broader claim about AI alignment with regard to gender and broader social representation: neutrality bites. In other words, we argue that prioritizing neutrality with regard to representations of gender, race, sexuality, or other identity categories, above all else, can function less as inclusion and more as erasure. <?blockquote>
Via.
David Culley has some thoughts on Transformers and human susceptibility.
Great Scott Gadgets: LLMs Generate Code; Communities Build Projects
Our communities are already noticing a shift away from collaborative stewardship and toward transactional interactions with projects, and we’re seeing open source project maintainers left to deal with increasing volumes of generated code. When people interact privately with systems trained on communities, it gradually weakens the shared understanding of how projects work and redistributes the burden more heavily onto (often underfunded) project owners.
Via.