AI
Matt Ross and
AI
The share of human-agent and agent-agent interactions in the economy will increase over time, with agents engaging in price negotiation, placing orders from one another, coordinating supply and demand, and even rating each other to assign trustworthiness scores.
Faster Horses
The writing is getting better. The ideas are getting worse. There’s a new genre of essay that other academics reading this will instantly recognize, a clumsy collaboration between students and Silicon Valley. I call it glittering sludge .
Brian Klaas • The Death of the Student Essay—and the Future of Cognition
But the pleasant surprise: It has been enormously helpful in my attempts to resurrect and improve my rusty engineering knowledge, and level up my middle-aged forays into maker-ing things. The killer app for me is the ability to go deep in a domain for remedial learning. The most immediately helpful use-cases for me have been:
- Getting back into
Venkatesh Rao • You've Been Matrix-Multiplied
Embodiment, AI - IDFA Report-1
Exploring the intersection of embodiment, immersive media, and artificial intelligence in documentary storytelling, highlighting audience experiences, challenges, and the evolving strategies used by creators to convey complex realities.
LinkAt IDFA DocLab in November 2023, MIT researchers explored how the
different forms of possible embodiment were being used by artists, and how
they had impact on audience experience and understanding. We wanted to
understand what strategies and tools artists are using, how these projects
are being received by audiences, and whether the knowledge communicated
or experience delivered is distinct from that shared by more traditional forms
of flat screen media.
A chatbot helped more people access mental-health services
A chatbot helped more people access mental-health services
GitHub's latest AI tool can automatically fix code vulnerabilities | TechCrunch
Frederic Lardinoistechcrunch.com
Observing these patterns of influence gives clues about how our models generalize from their training data. For instance, if the models responded to user prompts by splicing together sequences from the training set, then we’d expect the influential sequences for a given model response to include expressions of near-identical thoughts. Conversely,... See more
Anthropic \ Tracing Model Outputs to the Training Data
Memorization vs learning concepts and forming models of the world.
Subsequent research by Gerstgrasser and colleagues at Stanford and MIT (arXiv:2404.01413) provides important qualification: model collapse is not inevitable if synthetic data accumulates alongside real data rather than replacing it. The pathological scenario assumes complete substitution—thoughtful data curation can prevent the worst outcomes.