AI

Turkle referenced the issue of behavioral metrics dominating AI research, and her concern that the interior life was being overlooked, and concluded by saying that the human cost of talking to machines isn’t immediate, it’s cumulative. 'What happens to you in the first three weeks may not be...the truest indicator of how that’s going to limit you,... See more

In 2021, researchers made a striking discovery while training a series of tiny models on toy tasks [1]. They found a set of models that suddenly flipped from memorizing their training data to correctly generalizing on unseen inputs after training for much longer. This phenomenon – where generalization seems to happen abruptly and long after fitting... See more

We, as humans, sometimes receive streams of tokens and produce tokens in response, forming words, sentences, lines of code ... but always with the ability to peek outside the stream and check in with ground-floor reality. We pause and consider: does this word really stand for the thing I intend it to stand for? Does this sentence capture the real... See more

Cut the bullshit: why GenAI systems are neither collaborators nor tutors

For example, if you ask a model to “return all active users in the last 7 days” it might hallucinate a `is_active` column, join to an `activity` table that doesn’t exist, or potentially get the wrong date (especially in leap years!).

We previously talked to Shreya Rajpal at Guardrails AI, which also supports Text2SQL enforcement. Their approach was... See more