Mastodon FeedReblogged by fromjason ("fromjason.xyz ❤️ 💻"):
lety@doesstuff.social ("Lety Does Stuff") wrote:
A “model“ is not its “dataset“ and “open-weight“ is not “open-source“.
How DeepThink trained its models (read: software's ability to statistically quantify the relationship of information in their datasets) is unique and performant. That DeepThink-R1 has reasoning comparable to ChatGPT-4o when other competitors haven't come close is no small feat.
Further, DeepThink allows you to inspect and modify its models' biases and the weights they use to decide what information to present. They chose to be open-weight, which is broadly more commendable than OpenAI's decision to keep everything proprietary.