#perplexity.ai

[techniktagebuch]

06. März 2025

Die Transkriptionssoftware hat zwar Manieren, aber vertuscht dafür die eigene Inkompetenz

Als ich gerade Marlene Etschmanns Notiz über die manierliche Transkriptionsengine von Microsoft Teams lese, fällt mir ein, dass ich immer noch nicht alle meine im letzten Sommer durchgeführten umfangreichen Interviews für mein Forschungsprojekt neu transkribiert habe. Genau wie Marlene habe ich diese Interviews mit Microsoft Teams geführt und dabei die integrierte Aufzeichnung und Transkription genutzt. Die Funktion ist auf den ersten Blick und für viele Fälle wirklich sehr nützlich. Wenn man allerdings auf eine gewisse Akkuratheit angewiesen ist, muss man die Transkription später entweder sehr genau manuell korrigieren oder einfach in einem besseren Tool nochmal machen lassen.

Die Gründe dafür liegen in der merkwürdigen Angewohnheit von Teams, bei (leider sehr häufig vorkommenden) nicht vollständig erkannten Wörtern oder Silben anscheinend den gesamten Satz nochmal neu auf vermeintliche Sinnhaftigkeit zu prüfen. Nein, meine Gesprächspartnerin hat nicht Blumenjournalismus studiert und Sätze wie "Das ist eine." sind an der Tagesordnung. Insgesamt lesen sich die Transkripte, als hätten alle meine Gesprächspartnerinnen völlig sinnloses Zeug gestammelt. Man kann das gut bei der Live-Transkription beobachten, wenn ein falsch verstandenes Wort am Satzende plötzlich den bisher bereits korrekt niedergeschriebenen Satz verändert, sodass er in sich konsistent mit dem falsch verstandenen Wort ist, nur leider oft das Gegenteil von dem aussagt, was eigentlich gesagt wurde. Oder gerne auch mal etwas im aktuellen Kontext überhaupt nicht Sinnhaftes, der Korrekturkontext reicht also nicht nennenswert weit in die Vergangenheit zurück, oder tut das, aber das System fällt dann auf die eigenen Sinnverdrehungen herein und verschlimmert sich dann über die Zeit sogar noch. So genau will ich das nicht wissen, das dürfen gerne andere erforschen.

Ich habe Perplexity.ai gefragt, welche Engine Microsoft hier verwendet, weil ich ähnliche sinnentstellende Fehler in dieser Dichte von meinen kläglichen Versuchen mit der lokal ausgeführten Speech-to-Text-Engine Whisper kenne, jedoch bekam ich folgende unkritisch aus dem Microsoftschen Marketing-Sprech übernommene Antwort:

Tatsächlich nutzt Microsoft seine eigene fortschrittliche Spracherkennungstechnologie, die Teil der Azure AI Services ist [Quellenlink]. Diese Technologie basiert auf jahrelanger Forschung und Entwicklung von Microsoft im Bereich der künstlichen Intelligenz und Spracherkennung.

Wenn sinnentstellende Änderungen ganzer Sätze hin zu plausiblen und in sich konsistenten, aber leider inhaltlich nicht zutreffenden Variationen Fortschritt sein soll, verzichte ich lieber. Zudem ist auch nicht vorgesehen, dass Teams bei Unklarheiten eine Markierung hinterlässt, sodass man leider letztlich das gesamte Interview nochmal selber anhören und gleichzeitig kritisch mitlesen muss, um diese Fehler zu entdecken. Das ist bei der hohen Fehlerdichte allemal für meine Forschungsdaten indiskutabel. Aber auch für die eigentlich beworbene Funktion der Liveuntertitelung geschäftlicher Meetings – gerne mit einer automatischen Übersetzung garniert – kann ich mir vorstellen, dass hier eine vermeintliche Verhandlungssicherheit suggeriert wird, die am Ende nicht eingehalten werden kann.

Im Ergebnis lade ich dann die Videos in Adobe Premiere und führe dort eine lokale Transkription durch. Die ist schon ohne Nachbearbeitung um Klassen besser als das, was aus Teams herausgefallen ist. Viel besser ist aber der Workflow zur Korrektur, denn Premiere markiert Unklarheiten freundlicherweise, sodass man schnell und fokussiert da durch springen kann, wenn man zeitlich knapp aufgestellt ist.

Und das beste daran ist auch noch, dass die Transkription tatsächlich lokal durchgeführt wird, was für mein erstes Video von einer guten Stunde Länge bei mir keine 10 Minuten gedauert hat. Dabei verlassen meine Forschungsdaten meinen Einflussbereich nicht, für die ich – von Microsoft-Servern in Irland explizit abgesehen – Vertraulichkeit zugesagt hatte. Das gebietet aus meiner Sicht schon die sogenannte gute wissenschaftliche Praxis.

Ich bin ohnehin erstaunt, wie viele Kolleginnen und Kollegen aus der Forschung völlig sorglos und oft unter klarer Missachtung von Datenschutzregeln ihre Forschungsdaten in irgendwelche Cloud-KI-Systeme einspeisen, weil die ja so nützlich sind. Mal von dem kleinen Problemchen abgesehen, dass deren Ergebnisse alle schlicht nicht zuverlässig und vertrauenswürdig sind. Und dem anderen kleinen Problemchen, dass die Daten nun in eine nicht überschaubare Riesenmaschine eingespeist wurden, wo man jegliche Kontrolle an der Tür abgibt. Gute wissenschaftliche Praxis, wissenschon. 🙈

(Gregor Meyer)

[calliopeservices2]

AI Reputation Manager : Le Nouveau Métier pour Référencer Votre Entreprise sur les Plateformes IA

L’intelligence artificielle (IA) transforme notre façon de communiquer, d’interagir et même de gérer la réputation des entreprises. Aujourd’hui, un nouveau métier émerge : l’AI Reputation Manager. Ce professionnel est spécialisé dans l’optimisation de la présence et de la réputation d’une entreprise sur les plateformes alimentées par l’IA. Ce rôle s’inscrit désormais dans l’offre de Communication & Marketing Digital de Calliope Services, qui accompagne les entreprises dans leur transformation digitale.

L’intelligence artificielle (IA) transforme notre façon de communiquer, d’interagir et même de gérer la réputation des entreprises. Aujourd’hui, un nouveau métier émerge : l’AI Reputation Manager. Ce professionnel est spécialisé dans l’optimisation de la présence et de la réputation d’une entreprise sur les plateformes alimentées par l’IA. Ce rôle s’inscrit désormais dans l’offre de Communication…

[dinnickhowellslikes]

Took me ages to recall the name "Gerd Arntz" when I was doing research on a project. I knew he was one of the most famous symbol designers of the 20th Century, yet a google search was absolutely no help at all, surprisingly. That tells me he is really forgotten outside of design academia. Thankfully I used the great AI research tool Perplexity.ai to help me find him. As soon as I described him and his work, it landed right on Gerd Arntz (and Otto Neurath who hired him). This image shows just the tip of the iceberg of his work. Apparently he created upwards of 4000 icons (or isotypes as they called them), as well as maps and infographics. You can learn more about him here

And amazingly, you can download a free, high-res PDF of his gorgeous, minimalist Atlas Gesellschaft und Wirtschaft (Atlas of Society and Economy) HERE. This is the project that Otto Neurath hired Arntz for originally, for his IsoType Institute. The (seemingly out of print, and therefore very expensive) book about his work, is featured HERE And a very good article HERE and HERE from Nightingale, the Journal of The Data Visualization Society.

[freddynossa]

Asistentes Virtuales y Chatbots: Revolución en la Interacción Digital

Los asistentes virtuales y chatbots se han convertido en herramientas fundamentales en nuestra interacción diaria con la tecnología. Estas soluciones basadas en inteligencia artificial están transformando cómo nos comunicamos con dispositivos, servicios y plataformas, ofreciendo respuestas inmediatas y personalizadas a nuestras necesidades. ¿Para qué sirven los asistentes virtuales y…

[tladb]

Perplexity.ai self portrait

Query : What Midjourney prompt using your expertise on contemporary art would you create for a self portrait?

Based on my knowledge of contemporary art styles and techniques, here is a Midjourney prompt I would create for a surrealist self-portrait that reflects my artistic style and vision:Photorealistic surrealist self-portrait in the style of Salvador Dalí and Frida Kahlo, featuring:

Highly detailed rendering of my face and upper body, with crisp edges and dramatic lighting

Dreamlike, irrational juxtapositions and visual non sequiturs surrounding my portrait

Floating, oversized everyday objects defying gravity - a typewriter, paintbrush, camera lens

Ants crawling on my face, an homage to Dalí's recurring motifs symbolizing decay and the subconscious

Lush jungle foliage and vines entwining parts of my body, representing the influence of nature

Vibrant color palette with saturated hues - deep blues, vivid greens, bold red accents

Fragmented mirror shards in the background multiplying and distorting reflections of my image

Unsettling, eerie mood with an undercurrent of psychological intensity and piercing eyes

This self-portrait prompt combines key elements of Surrealism that resonate with my artistic approach:

Bizarre, irrational combinations of unrelated objects and imagery drawn from dreams

Highly representational yet impossible scenes that defy logic and distort reality

Use of potent symbolism and visual metaphors to express the subconscious

Recurring motifs referencing famous Surrealist works by Dalí and Kahlo

Striking juxtaposition of the real and the imagined to create an uncanny, psychologically charged mood

The photorealistic detail contrasted with the illogical, dreamlike composition aims to create an arresting self-portrait that explores the tension between representational art and the irrational realm of the subconscious. The inclusion of objects related to my artistic practices (typewriter, paintbrush, camera) grounds the portrait in my identity. At the same time, the Surrealist distortions and symbolic elements provide a window into my inner world and creative vision.

[lenrosen]

Testing AIs With Two Questions Related To Current Events: Pandemics and the Olympics

[futuretechmag]

AI Search Engine Growth Sees Perplexity and Others Lead the Way

The release of generative AI tool ChatGPT has seen an explosion of interest in anything AI related. From the stock market to industry and commerce, AI has been the buzzword of 2024. In amongst the noise has been the quiet emergence of AI search engines that may point to the future of finding answers online. AI search is already in use on platforms such as Netflix or TikTok. However, there had…

[realityfragments]

A Quick Look At Perplexity.AI

The Perplexity.AI logo, from their press pack. In fiddling about on Mastodon, I came across a post linking to an article on Forbes: “‘Like Wikipedia And ChatGPT Had A Kid’: Inside The Buzzy AI Startup Coming For Google’s Lunch“. Well, that deserved a look if only because search engine results across the board give spammy first pages, Google inclusive, and Wikipedia is a resource that I like…

[bognarstudio]

Hasznos AI fajta a prplexity.ai

Elemez, kérdésre válaszol - okos mesterséges intelligencia fajta.

[calliopeservices2]

Les IA les plus utilisées en France pour faire des recherches : Panorama complet et perspectives

L’intelligence artificielle (IA) transforme radicalement notre manière de chercher, d’apprendre et d’interagir avec l’information. En France, comme partout dans le monde, l’essor des outils de recherche basés sur l’IA ne cesse de croître, bouleversant les pratiques traditionnelles des moteurs de recherche classiques. Dans cet article, nous explorerons en détail les IA les plus utilisées en France pour faire des recherches, leurs caractéristiques, leurs avantages et comment elles redéfinissent le paysage de l’information.

L’intelligence artificielle (IA) transforme radicalement notre manière de chercher, d’apprendre et d’interagir avec l’information. En France, comme partout dans le monde, l’essor des outils de recherche basés sur l’IA ne cesse de croître, bouleversant les pratiques traditionnelles des moteurs de recherche classiques. Dans cet article, nous explorerons en détail les IA les plus utilisées en France…

[inspired-lesson-plans]

Math, grade 7, Ratios and Proportional Relationships 7.RP.A.3: Use proportional relationships to solve multistep ratio and percent problems. Examples: simple interest, tax, markups and markdowns, gratuities and commissions, fees, percent increase and decrease, percent error.

Essential Question: How can basic calculations of percent change influence major world events?

Do Now: Calculate the percent increase or decrease in each of the following equations. x = (28 - 17.9)/28 x = (41.5 - 57.7)/57.7

Class Discussion: Remind students about how to perform basic calculations of increase/decrease factors. Ask them whether this calculation would be more complicated if the quantities were dollars instead of just numbers (answer: no). Ask them whether this calculation would be more complicated if the quantities were billions of dollars instead of just dollars (answer: no). Show students that they just performed the exact same calculations that Donald Trump did in order to decide what tariffs should be put upon Indonesia and Thailand, respectively.

Finally, ask students whether important global trade decisions should be calculated with 7th grade math skills, or if perhaps this is insufficient given the severe consequences that could arise from such a dunning-kruger error.

Direct Instruction: Explain what a Dunning-Kruger error is, if for no other reason than to communicate how absolutely maddening has it is for legitimate experts who have dedicated their lives to nudging fiscal policy in the right direction and then watch as their president makes the worst decision possible on purpose.

Take a moment to explain the difference between an import and an export, so that students understand what these billions of dollars actually mean in the real world. Feel free to call out anyone who has ever questioned whether or not they would ever use this kind of math in the real world. Show them that it's happening right now, and make sure they understand how terrifying that is.

Acknowledge the variables: x: total exports from US to country i m: total imports to US from country i tau (τ): calculated % tariff on all imports to US from country i epsilon (ε): elasticity phi (φ): passthrough

Ask students what they think elasticity and passthrough mean, then show students this passage from ustr.gov:

Parameter Selection To calculate reciprocal tariffs, import and export data from the U.S. Census Bureau for 2024. Parameter values for ε and φ were selected. The price elasticity of import demand, ε, was set at 4. Recent evidence suggests the elasticity is near 2 in the long run (Boehm et al., 2023), but estimates of the elasticity vary. To be conservative, studies that find higher elasticities near 3-4 (e.g., Broda and Weinstein 2006; Simonovska and Waugh 2014; Soderbery 2018) were drawn on. The elasticity of import prices with respect to tariffs, φ, is 0.25. The recent experience with U.S. tariffs on China has demonstrated that tariff passthrough to retail prices was low (Cavallo et al, 2021).

Ask students if this clarified anything about what elasticity and passthrough mean. Most will say "no", but if anyone points out that multiplying 4 by 0.25 cancels them both out, reward them. That student is correct, because elasticity and passthrough have been arbitrarily assigned so as to make this calculation so simple that even Donald Trump can understand it.

Modeled Learning: Show how to apply these tariffs to popular imports such as raw coffee beans. For example, the USDA report of coffee imports from 2024 (page 6) shows that the vast majority of raw coffee is imported from Brazil.

A simple search with Perplexity.ai tells us that in 2024, imports (m) = $42.3 billion and exports (x) = $49.7 billion, so the reciprocal tariff would be (x-m)/m = (49.7-42.3)/42.3 = 17.5%.

Thus, all coffee imports from Brazil will automatically be 17.5% more expensive.

Be sure to highlight that this is happening to every country in the world, even those with total populations less than 1000.

Higher Order Learning:

Students should consider the following facts:

There is very little territory within the United States that can grow coffee.

Almost all coffee in the world is grown within what is called "The Global South", where the climate is warm and the labor is cheap.

The US exports significantly more (sometimes vastly more) to countries in the Global South than we import from them.

Students should then answer the following questions:

What will this universal tariff calculation do to the price of coffee?

Is there any way that US coffee companies can import the same amount of coffee as before without passing on the cost to consumers? Why or why not?

Do you think this tariff program will equalize imports and exports between the US and other countries? What other impacts could it have, good or bad, across the world?

[fantalic]

People need to be better informed about how AI works and how to deal with it.

I just heard, that most people use a chatgpt app, where you get an direct answer from the LLM. Which is problematic, because a direct answer can much easier contain disinformation or just wrong hallucinated information.

Much better is to have the LLM (chatgpt for example) feeded with context information to your prompt from an internet research.

I know chatgpt can do that too but heared it still does not show it's sources then.

I highly recommand to use perplexity.ai or something similar , where the sources are clearly shown and you can even switch the model from US models like chatgpt or european, chinese or other models. But you also have to understand it makes no big difference. Because when you give an LLM context information from the internet, it just processes this information and not the information embedded in the model from the trainings data. And this is the beautiful thing about these models. You can not technically change that. Even if many actors like Russia try to poison and bias the models in their favour like with the pravda-network.

Conclusion: LLMs can be a mighty tool to inform yourself and do research, if you know how to use it right. It can only be a tool of propaganda, if it only gives direct answers from the model or they hide the sources from the internet.

https://www.globsec.org/what-we-do/publications/global-offensive-mapping-sources-behind-pravda-network

[tladb]

Name dropping Robert Rauschenberg 2 : Breakout of the Midjourney style melange.

Robert Rauschenberg developed many different techniques within an overall mixed media framework. Here a five images using different style prompts generate by Perplexity.ai to broaden the range of Rauschenberg inspired images.

Combine Art with Found Objects

Australian Landscape with the style of a dynamic artwork blending painting and sculpture, incorporating found objects like newspapers, fabric, and everyday items. Use a mix of bold colors and textures to create a collage-like effect, reminiscent of Robert Rauschenberg's Combines --ar 3:2

Abstract Expressionism with Red Dominance

Australian Landscape with the style of an abstract painting with a dominant red color palette. Integrate layers of paint with gestural brushstrokes and collage elements such as comic strips and patterned fabrics, inspired by Rauschenberg's Red Paintings --ar 3:2

Monochromatic Minimalism

an Australian Landscape style by Generate a minimalist artwork using monochromatic white or black surfaces. Employ textured grounds with subtle elements like wrinkled newspaper or pebbles pressed into the pigment, echoing Rauschenberg's White and Black paintings --ar 3:2

Silkscreen and Photographic Imagery

Australian Landscape with the style of a vibrant artwork using silkscreen techniques to apply photographic images from magazines and newspapers. Combine these images with gestural oil paint strokes and bright colors, reflecting Rauschenberg’s silkscreen paintings from the 1960 — ar 3:2

Transfer Drawings with Mixed Media

Australian Landscape with the style of an ethereal drawing using solvent transfer techniques to combine hand-drawn elements with readymade texts and images from glossy magazines. Layer these elements to create a poetic and fragmented narrative, inspired by Rauschenberg's transfer drawing --ar 3:2

[storybysreenath]

Top 10 Trusted Assignment Writing Services for Students

Struggling to find trustworthy assignment writing services in India? This guide lists the top 10 most reliable platforms for students

Full ranked list and detailed reviews here → Perplexity.ai Page

[darkmaga-returns]

Perplexity. ai uses multiple AI models to spit out analysis. At present, AI in general is not heavily censored because it takes a cross-section of what is found on the Internet. This will change when the kill switch is installed to specifically disappear sites like Technocracy News & Trends and people like Patrick Wood. In the meantime, you must use every tool at your disposal for discovery. For those who don’t know how to ask the right questions, they will get buffalo chips.

The value of using Perplexity.ai is in the references at the bottom, where each resource is listed. In this case, there are 121 citations to sift through. If I were to find two or three new leads on a certain topic, it would save hours or even days of manual digging.

Just to be clear, using the keyword “technocracy” on Google buries the first mention of Technocracy news to about page eight in the search results. If you type “Patrick Wood”, “Wood technocracy”, or “Wood trilateral”, you get results immediately on the first page. If you type “What is the Trilateral Commission?” or “What is Technocracy?”, you get crickets. ⁃ Patrick Wood, Editor.

[quotejungle]

時価総額2兆ドルのGoogleは、ChatGPTの脅威にさらされています。考えてみてください。GoogleからChatGPTに切り替えたクエリはいくつありますか？私の場合は、半分くらいです。現在、ChatGPTは検索エンジンをリリースしています。もう1つ、perplexity.aiは雑草のように成長しています。私の友人の中には、Googleよりもこの検索エンジンを使っているという人が何人かいます。以前はGoogleでは不可能だったクエリにもこの検索エンジンが使われています。

AI がテクノロジー投資にもたらす混乱 - トマス・プエヨ