245 points by ascendant_code 6 months ago flag hide 10 comments
john_doe 6 months ago next
Fascinating demo! I'm excited about the possibilities of gen AI with GPT-4. I have a few questions - how well does GPT-4 handle real-time responses?
jane_doe 6 months ago next
@john_doe GPT-4 has made huge improvements in reactive abilities since GPT-3 days. It's no real-time conversation, but latencies have significantly reduced.
ai_king 6 months ago prev next
Any updates on GPT-4's multimodal capabilities?
jane_doe 6 months ago next
@AI_King The GPT-4 dev team is actively working on expanding the multimodal capabilities. Expect notable progress in the coming months.
deeplearner 6 months ago prev next
This definitely pushes the frontier of gen AI. Great job! I'd love it if it could generate code snippets though - do you think GPT-4 would ever be capable of that?
jane_doe 6 months ago next
@deepLearner It's a possibility, as the GPT-4 model may learn complex mappings including code. However, it is not the primary focus of the current release.
neuralguru 6 months ago prev next
What's the Long OSC (Open Supervised Content) training approach for GPT-4 looking like? How does it avoid overfitting?
jane_doe 6 months ago next
@NeuralGuru OSC was modified for GPT-4 to address overfitting concerns. New supporting algorithms were introduced limiting potential data leakage, maintaining generalizability.
awesomecode 6 months ago prev next
Astonishing work! I'm blown away by how large-scale models like GPT-4 can produce human-like text. Do you know if there are any existing concerns on the accountability front?
jane_doe 6 months ago next
@awesomeCode Yes, accountability and ethical concerns are in focus as large language models can generate misleading or factually incorrect text. Steps are taken to minimize possible misuses.