#studylog — Public Fediverse posts

So viele Neuigkeiten…

Lesedauer 2 Minuten

Ich habe noch keine offiziellen Nachrichten, aber einer Laune folgend besuchte ich dieser Tage die Webseite meiner Uni und fand dort die vorläufige Bewertung meiner Master Thesis. Ich müsse nur noch, so wurde ich dort informiert, auf die Bestätigung durch das Board of Examiners warten und dürfte mich ab dann „MSc Artificial Intelligence“ nennen. Yey!

Ich selbst hatte zuletzt beim Blick in meine Abschlussarbeit nur noch Müll gesehen, ganz ähnlich zu dem Effekt wenn man ein Wort wieder und wieder sagt bis es sich im Kopf bedeutungslos anfühlt. Ich hatte ernsthaft Zweifel ob ich eigentlich genug Erkenntnisgewinn aus meinem Thema (Der Versuch einer entkoppelten AI Architektur um mit Hilfe lokaler LLMs erklärbare Ergebnisse zu produzieren) wringen konnte. Scheint gereicht zu haben (zumindest bisher), denn im Normalfall ist diese finale Bestätigung nur noch Formsache. Aber wenn die nicht auch manchmal schiefgehen könnte, gäbe es sie nicht… Die metaphorische Kuh ist also noch nicht vom Eis.

Die andere Neuigkeit war eigentlich fast noch aufregender: Letzte Woche wurde ich informiert, dass ich eine sogenannte „Invitation to Apply“ bekommen hatte. Das ist einer der wichtigsten Schritte Richtung Residency hier in Kanada und damit wirklich gigantisch gute Nachrichten. Bisher sind wir ja „nur“ mit einer zeitlich limitierten und an meinen Arbeitgeber gekoppelten Arbeitserlaubnis hier in Kanada und diese ITA ermöglicht uns nun eventuell eine unbefristete und arbeitgeberunabhängige Aufenthalts- und Arbeitserlaubnis zu erwerben. Folgerichtig habe ich die Woche mit dem Zusammentragen von Unterlagen verbracht (Lebenslauf, Kontoauszüge, aktuelle Urkunden, Zeugnisse, etc. etc.) und werde ab jetzt nervös den Prozessverlauf verfolgen.

Es wäre ja wirklich großartig wenn sich alles dieses Jahr auflösen würde wegen dem ich mir immer wieder echte Sorgen gemacht habe. Noch ist es viel zu früh, Korken knallen zu lassen, aber es ist zumindest viel wahrscheinlicher geworden, dass wir unseren Weg so weitergehen können wie wir es uns erhofft hatten.

Läuft also unerwartet gut zur Zeit und ich hoffe mal, das geht so weiter…

#Residency #studylog

Neuer Blogpost: AI links, human notes #3

Lesedauer 2 Minuten

My bookmark list keeps growing. For this blog series, I spare you the tech books and tutorial videos I’m spending most of my time with and focusing on the broader picture. Let me know in the comments (or by answering this post on the fediverse) how you like my selection.

Text:

• Meta plans AI chatbots in their networks. Not much longer and we won’t know if the person we’re chatting with online is human or virtual.
• Researchers at Carnegie Mellon University use machine learning to „see“ people by analyzing wifi signals. Yes, this technology can turn your home wifi router into a camera system. I find this fascinating and terrifying at the same time.
• Is AI coming for our jobs? Hollywood actors went on strike because Hollywood studios consider using AI replicas of them – for free. Ouch.
• It looks like chatGPT can recognize faces but if you ignore that, then mulöti-modal LLMs are real game changers for blind people.
• Bill Gates wrote a blog post about the risks of AI and says he is confident that they are manageable.
• btw, OpenAI assumes they’re en route to AGI (Artificial General Intelligence) and outlined principles on how to approach it.
• Too much doom? No worries, the world economic forum thinks AI will give us with long term job growth.
• … unless… Goldman Sachs is right. They predict 300m jobs loss.
• How smart are AI systems actually? It may be surprising, but this is hard to say because we lack a solid definition we can all agree on. So, let’s read up on the question of how we would know how smart they are.
• Students (and lawyers) everywhere wonder „Can anyone tell that I asked chatGPT to write my thing?“. The short answer is „probably„.

Podcast & Videos:

• Maybe the most inspiring use of AI I have heard so far: The Earth Species Project is a non-profit dedicated to using AI to decode non-human communication. Yes, that means using AI to talk with animals. Here’s a podcast explaining more.
• Writers are worried about the use of AI as well, specifically, the fact that their texts are being used to train AIs. The podcast „The Daily“ has more context.
• I don’t like the hyperbole of calling this video podcast an emergency episode. However, this conversation with Mo Gawdat about the direction AI is taking is still interesting and thought-provoking.
• Yuval Harari is probably one of the smartest thinkers we have and gave a keynote lecture called „AI and the future of humanity“ that is well worth your time.

#AI #studylog #linklog

#ai #english #linklog #studylog

https://taeglichkeiten.kopfstim.me/?p=562

Achtung:Direkt auf den Post gegebene Antworten tauchen evtl. auch als Kommentar im verlinkten Blogpost auf.

Neuer Blogpost: AI links, human notes #2

Lesedauer 3 Minuten

Well, it’s been a while since my last AI centric link list. Let’s add some recent reads and watches to your todo list.

Text:

• Brave just released an API for their SearchEngine with an emphasis on it’s use in training AI. It is a bit surprising given the privacy and security centric ethos of the Brave ecosystem.
• [Paper] On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
• [Paper] Language Models are Few-Shot Learners
• [Paper] Smaller Language Models are Better Black-box Machine-Generated Text Detectors
• Well, this is very old for everyone in the field but maybe new to you… The paper that gave birth to modern large language models is already 8 years old and still a read worthy your time: Attention Is All You Need
• This paper is a bit much for you? No worries: Here is an article walking you through all key ideas behind chatGPT, Bard & Co.
• Eliezer Yudkowsky is convinced that AI will kill us all. Normally we would ignore him but it turns out he’s a very accomplished scientist who kinda knows what he’s talking about.
• There is quite the number of scientists who warn about the risks of AI. Some may make you feel like you’re living in the end times (looking at you, Eliezer Yudkowsky). If you’re looking for a balanced view on some of their argument then Paul Christianos post „Where I agree and disagree with Eliezer“ is a highly recommended read.
• The case for how and why AI might kill us all – I think the title says it all… And, yes, it is Eliezer again.
• A little less panicky but not any less concerning is this piece in MIT Technology Review explaining why Geoffrey Hinton, one of the fathers of modern AI, decided to leave Google so he can tell us why he’s now scared of his creation.
• Even if you don’t believe that it is quite as dangerous there are still plenty of issues. Governments for instance realize these days that AI is far far ahead of current laws and regulations. So politics is playing catch up, and Europe announces to spin up an AI research hub to apply accountability rules.
• Not depressed already? Good. ars technica tells you all about „The mounting human and environmental costs of generative AI„. That might do the trick.
• Math is your cocaine? You’ll be thrilled to browse through this blog post explaining the math related to computation and memory usage for transformers. You’re welcome.

Videos:

• Yes, 67 minutes is not exactly short. But I watched it anyhow and then I shared it with everyone I knew. But be warned: you’ll be worried afterwards. It talks you through parallels between the current AI revolution and the events when we introduced social media. Then it walks you through AI in general and how the situation today relates to society, politics and the moment we find ourselves in. Scary stuff. The A.I. Dilemma
• So you want to know how to build your own GPT? Here are 2 hours that take you through the entire process from start to finish.
• Eliezer Yudkowsky is a very respected scientist who truly beliefs that AI will kill us all. His belief comes from years of study and understanding that it is a very hard problem. This 90 minute lecture he gave 6 years ago at Standford University explains why.
• Another lecture, this time from Robert Miles. In about 18 minutes he walks you though the subject of AI safety.
• Lex Friedman talks for his podcast with high caliber thinkers such as the philosopher Sam Harris. Here you find the 20 minute cut-out where they speak about AI specifically.

#AI #studylog #linklog

Image: Midjourney

#ai #english #linklog #studylog

https://taeglichkeiten.kopfstim.me/?p=535

Achtung:Direkt auf den Post gegebene Antworten tauchen evtl. auch als Kommentar im verlinkten Blogpost auf.