Physicist Max Tegmark says competition too intense for tech executives to pause development to consider AI risks

Det er muligt, at blyanten har en fremtid. Men det bliver nok næppe blyanten eller andre analoge opfindelser, der kommer til at gøre den store forskel, når vi skal finde fremtidens arbejdskraft. Vi kan derimod med rette have store forventninger til den digitalisering af samfundet, der for øjeblikket – til gavn for både virksomheder og offentlig sektor – er i rivende udvikling.
Men hvis vi skal veksle digitalisering til den nye valuta i dansk politik – øget arbejdskraft – kræver det, at vi investerer og digitaliserer med omtanke.

De fejlagtige ejendomsvurderinger viser, hvor galt det kan gå, hvis vi bygger for store forventninger til algoritmer og AI ind i vores forvaltning. Lad det være det sandhedens øjeblik, der forhindrer, at vi konstruerer fremtidens systemer på en måde, så mennesker skrives ud af loopet.

EU legislation will herald greener devices, with greater longevity for software and access to user-replaceable parts

Today let’s step outside the news cycle and turn our attention toward a topic I’m deeply invested in but only rarely write about: productivity platforms. For decades now, software tools have promised to make working life easier. But on one critical dimension — their ability to improve our thinking — they don’t seem to be making much progress at all.

Meanwhile, the arrival of generative artificial intelligence could make the tools we use more powerful than ever — or they could turn out to be just another mirage.

To understand where things went wrong, I want to focus on the humble note-taking app: the place where, for so many of us, thinking begins.

Whisky is distilled to around 70% alcohol by volume (vol-%) then diluted to about 40 vol-%, and often drunk after further slight dilution to enhance its taste. The taste of whisky is primarily associated with amphipathic molecules, such as guaiacol, but why and how dilution enhances the taste is not well understood. We carried out computer simulations of water-ethanol mixtures in the presence of guaiacol, providing atomistic details on the structure of the liquid mixture. We found that guaiacol is preferentially associated with ethanol, and, therefore, primarily found at the liquid-air interface in mixtures that contain up to 45 vol-% of ethanol. At ethanol concentrations of 59 vol-% or higher, guaiacol is increasingly surrounded by ethanol molecules and is driven to the bulk. This indicates that the taste of guaiacol in the whisky would be enhanced upon dilution prior to bottling. Our findings may apply to other flavour-giving amphipathic molecules and could contribute to optimising the production of spirits for desired tastes. Furthermore, it sheds light on the molecular structure of water-alcohol mixtures that contain small solutes, and reveals that interactions with the water may be negligible already at 89 vol-% of ethanol.

Ugly Numbers from Microsoft and ChatGPT Reveal that AI Demand is Already Shrinking
The only areas where AI is flourishing are shamming, spamming & scamming

I’ve been asked, in various roles1, to give my opinion on the challenges posed by Large Language Models (LLMs)2, also known as “stochastic parrots” (Bender, Gebru, McMillan-Major, & Shmitchell, 2021), for assessing academic writing assignments. A concern seems to be that students legitimately can use these systems and that then we would be unable to assess their ability to write essays.

My opinion, in brief, is that LLMs cannot legitimately be used to write academic essays.