AI
Matt Ross and
AI
AI-powered document editors, then, among other tools, use all that training data to make it even easier to write outcome-oriented garbage that moves us further and further from the truth of ourselves and each other as beautiful, complex, experiencing beings.
Dan Hunt • Internet as Practice
Superhuman performance of a large language model on the reasoning tasks of a physician
arxiv.org
GitHub - huggingface/text-embeddings-inference: A blazing fast inference solution for text embeddings models
huggingfacegithub.com
What if you could instruct customers to migrate via prompt? First time I have seen this. Hard not to unsee it as a possibility.
White faces generated by AI are more convincing than photos, finds survey | Artificial intelligence (AI) | The Guardian
amp.theguardian.com
Max Kennerly (@maxkennerly.bsky.social)
bsky.app
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.
The web is no longer primarily human-generated text created for human readers; it is increasingly AI-generated text created for algorithmic amplification, mixed with human-generated text optimized for the same algorithmic reward functions.
As AI systems train on this degraded corpus, they learn to reproduce its characteristics—shallow reasoning,... See more
As AI systems train on this degraded corpus, they learn to reproduce its characteristics—shallow reasoning,... See more