AI - AI Dynamics

→ Voir le post original sur X — @jessyseonoob

Growing Difficulty Verifying AI-Generated Video Authenticity

By

@jessyseonoob

–

31 March 2023 0h46

fait à partir de cette vidéo ? ça va devenir de plus en plus compliqué d'arriver à savoir ce qui est vrai.https://t.co/hgva2E3Jsd
— ⚡Jessy SEO formation #IA générative #chatGPT 🦊 (@jessyseonoob) 30 mars 2023

fait à partir de cette vidéo ? ça va devenir de plus en plus compliqué d'arriver à savoir ce qui est vrai.

Steering GPT behavior through weekly model retraining and evaluation

By

–

31 March 2023 0h00

With GPT-3, didn't have a way to steer the model's behavior. Now we pick a clusters of issues (e.g. model doubling down on wrong answers — something that used to be a huge problem), build evals, collect data, retrain the model weekly, and deploy. Anyone can test if we've improved

GPT-4 Alignment Progress: Scaling Beyond Human Judgment

By

–

31 March 2023 0h00

But aligning GPT-4 is a step, not the destination. Great progress (compare to early complaints about GPT-3), but will need to scale to problems where human can't even judge the outputs. GPT-4 itself already used a lot in our alignment, step towards this:

GPT-4 Deployment and Practical AI Alignment Progress

By

–

31 March 2023 0h00

Deploying GPT-4 subject to adversarial pressures of real world has been a great practice run for practical AI alignment. Just getting started, but encouraged by degree of alignment we've achieved so far (and the engineering process we've been maturing to improve issues).

Model Generalization of Helpful AI Assistant Behavior Across Topics

By

–

31 March 2023 0h00

One example observation — much of model's good behavior across such a broad range of tricky or adversarial topics comes from the model having generalized the human instructors' concept of being a helpful AI assistant. Not something I saw coming!