BBC news about India has been so negative in the past few years, I have stopped trusting them. Of course there are other news about them spoofing videos.
They've developed a sparse attention mechanism (which they document and release source code for) to increase model efficiency with long context, as needed for fast & cost-effective extensive RL training for reasoning and agentic use
They've built a "stable & scalable" RL protocol - more capable RL training infrastructure
They've built a pipeline/process to generate synthetic data for reasoning and agentic training
These all combine to build an efficient model with extensive RL post-training for reasoning and agentic use, although they note work is still needed on both the base model (more knowledge) and post-training to match frontier performance.
Another report. Sadly nothing will come out of it. 5 years down the line there will be another smoking gun, accusing Meta of selling DNA data to advertisers illegally. Nada. Nothing will happen.
Few reasons: keep access to the frontend features of each providers, have access to my chats I have in the individual frontend apps, to not have to trust a 3rd party provider, to not have to update the app each time a new model comes out
reply