```dataview LIST WITHOUT ID file.link WHERE file.cday = date(this.file.name) SORT file.cday ASC ``` # Idea for RoundTable Okay, so I need to pair a few hundred words for the AI and teaching roundtable. Here is my idea. The slogan for my idea is the key to using LLMs for education is to see them not as learning machines, but as thinking machines. So, one of the primary concerns about using LLMs in education is that LLMs hallucinate, or, put it another way, lie. Here we want to suggest that a better way of thinking about these things is they are not machines for learning, they are machines for thinking. # text for the ai education roundtable We argue that large [[language models]] (LLMs) are not machines for learning but machines for thinking. When we stop seeing LLMs as sources of knowledge and start seeing them as tools that help students [[process knowledge]], we sidestep concerns about factual accuracy and hallucination. The focus moves from whether LLMs provide correct information to how they can help students engage with information they've already received from [[authoritative sources]]. First, they provide judgment-free spaces where students can test their understanding without fear. For example, a biology student struggling with complex genetic concepts can repeatedly ask basic questions to an LLM that they might hesitate to ask in class. Unlike asking questions in class or study groups, exploring ideas with an LLM carries no social risk. Students can work through confusion, test half-formed thoughts, and make mistakes as part of their learning process. Second, LLMs mirror student thinking back to them. When a student explains a concept to an LLM, they must first organize their own understanding. For instance, an engineering student might explain fluid dynamics principles to an LLM and realize through the response that they've conflated two key equations. The LLM's response reflects their explanation, helping students spot gaps or inconsistencies they might have missed. This is like explaining something to understand it better, but without social pressure. Third, LLMs connect ideas across different subjects quickly and flexibly. For instance, a student studying neuroscience who encounters Kantian ethics in a philosophy module could use an LLM to explore how these seemingly disparate fields might relate to each other. Students can investigate how concepts from one course apply to another, building the interdisciplinary connections that university education aims to develop. By reframing LLMs as thinking tools rather than learning tools, we address practical teaching challenges while avoiding concerns about AI replacing critical thinking. ## italian Sosteniamo che i modelli linguistici di grandi dimensioni (LLM) non sono macchine per l'apprendimento ma macchine per il pensiero. Quando consideriamo LLM come strumenti che aiutano gli studenti a elaborare la conoscenza anziché fonti di informazione, evitiamo le preoccupazioni relative all'accuratezza fattuale e alle allucinazioni. L'attenzione si sposta dal verificare se forniscono informazioni corrette al modo in cui facilitano l'interazione con contenuti già acquisiti da fonti autorevoli. In primo luogo, offrono spazi privi di giudizio dove gli studenti possono testare la loro comprensione liberamente. Uno studente di biologia che fatica con concetti genetici complessi può porre ripetutamente domande basilari a un LLM che esiterebbe a formulare in classe. Questa esplorazione non comporta rischi sociali, permettendo agli studenti di affrontare la confusione, testare pensieri incompleti e commettere errori come parte naturale dell'apprendimento. In secondo luogo, gli LLM fungono da specchio del pensiero. Quando uno studente spiega un concetto, deve prima organizzare la propria comprensione. Uno studente di ingegneria potrebbe illustrare principi di dinamica dei fluidi e scoprire, dalla risposta dell'LLM, di aver confuso due equazioni fondamentali. Questo processo rivela lacune o incongruenze nel proprio ragionamento, simile a spiegare qualcosa per comprenderlo meglio, ma senza pressioni sociali. In terzo luogo, questi sistemi collegano rapidamente idee tra discipline diverse. Uno studente di neuroscienze che incontra l'etica kantiana potrebbe utilizzare un LLM per esplorare le correlazioni tra questi campi apparentemente distanti, costruendo quelle connessioni interdisciplinari che l'educazione universitaria cerca di sviluppare. Riconcettualizzando gli LLM come strumenti di pensiero piuttosto che di apprendimento, affrontiamo sfide pratiche dell'insegnamento evitando preoccupazioni sulla sostituzione del pensiero critico da parte dell'intelligenza artificiale. In this paper we argue that [[psychedelic aesthetics]] can be modelled on pictorial aesthetics. Both pictures and psychedelics generate distinctive perception-like experiences that can be aesthetically appreciated in two ways: for the scenes they present (scene-oriented aesthetics) and for [[the relationship between]] their physical elements and what these elements depict (design-oriented aesthetics). Drawing on the [[predictive processing]] framework, we demonstrate how psychedelics relax the brain's priors, creating unique perceptual experiences that go beyond merely enhancing existing [[aesthetic properties]]. We show how psychedelics offer distinctive ways of seeing that parallel pictorial experience, and how they create a partial opacity of perceptual processes that reveals something about how perception itself is constructed. While psychedelic experiences vary across individuals, they instantiate recognisable patterns that can be shared and discussed, providing a basis for genuine [[aesthetic appreciation]].