No Human-Machine aka #AI Interaction without a deep understanding of behavioral sciences.
Below is a list of 150 remarkable individuals in this field. 🙌
go.bsky.app/CzLwTyZ

This study from the University of Washington & Stanford seeks to establish a framework to analyze the "reasoning" of medical AI.
The team used GenAI to create pairs of skin lesion images to alter the model's prediction. Dermatologists then identified the AI's relevant decision features. 💡

Interesting!
This paper from Google DeepMind shows LLMs can find patterns with self-generated reasoning.
By only feeding minimal input, the system constructs its own CoT (Chain of Thought), boosting performance + unlocking insights to analyze the model’s reasoning patterns arxiv.org/abs/2402.10200

Deutsche Telekom is announcing a new AI-powered smartphone without traditional apps. Instead, a centralized AI assistant will handle diverse tasks typically accomplished through multiple apps.
More showcases and info in the link 👇
www.telekom.com/en/media/med...

Wow!
This study from MIT not only detected novel structural classes of antibiotics, but it did so with explainable AI (XAI). Opening the Black Box and uncovering the AI's reasoning will push the boundaries in new drug design

h/t Karl_Lauterbach on Twitter
www.nature.com/articles/s41...

Interesting! Similar to medical placebos, AI (LLM) can induce behavioral changes and alter subjective performance due to heightened positive expectations – even when participants were working with a placebo!
arxiv.org/abs/2309.16606

Unbelievable!
OpenAI has made groundbreaking strides in video generation. So far, maintaining the consistency of the moving images has been rather challenging, but Sora can produce up to ⚠️ 60 seconds of seamlessly consistent video content!
See clips in link below 👇
openai.com/sora?utm_sou...

Zhou et al. (Google Deepmind, University of Southern California) introduce a 3-step prompting framework, from which an LLM can autonomously select and then construct a solution plan. The "Self-Discover" prompt technique outperforms traditional pathways like Chain-of-Thought
arxiv.org/pdf/2402.036...

The first AI-tailored ads are airing, with the capability to alter content for different regions, or include customized material. It would be preferable for these advancements to not always appear first in advertising, but rather at critical societal points like the education system.

Hahaha, this is great!
Instead of filing another endless report about Davos, Jen Stave from the Digital Data Design (D^3) Institute at Harvard, simply tossed the line to ChatGPT.

• 5 pages
• Pizza 🍕
• Coffee machine ☕
• and this tweet is already longer than her prompt
d3.harvard.edu/wp-content/u...

Hahahaha 😆👌
The Misalignment Museum in San Francisco hosts a sophisticated arrangement of #AI-related art under the topic of an AGI-dominant afterworld called "Sorry we killed most of humanity!".
I strongly recommend the Stanford Course "AI Alignment" to all of you – to keep us Humans around. 🙏

An NYU study uses a baby's head camera to teach AI language as infants learn it. Shifting towards a human-centric learning model instead of step-by-step programming, this study challenges big data reliance, crafts smarter AI learning patterns, and deepens our understanding of human intelligence.

Wow, superschöne Animation der Financial Times. Anschaulich wird erklärt, wie ein Large Language Model funktioniert, und wie Programme wie ChatGTP Texte generiert – oder warum es uns manchmal "Halluzinationen" präsentiert.
Bitte teilt das Wissen 🚀
ig.ft.com/generative-ai/?

Leicht verständlich erklärt von einem der Mitwirkenden, Prof. Philipp Hacker:
Worum geht es bei den Regulierung von KI-Modellen, und was hat die deutsche Bundesregierung vor?

www.zeit.de/digital/2023...