When reading responses from Claude, you naturally come up with lots of inline questions. But asking them as follow-up messages creates a new branch of conversation and a new wall of text.
Now you're forced to context-switch between the original response you were reading and the answers, losing your place and your reading flow.
I wish there was some kind of "annotations" or "side conversations" attached to the responses that can be expanded inline without disrupting the original text.
@claudeai
Continuing on the post-training series— this time into RLHF.
Tried breaking down reward models, PPO, and the overall intuition in a simple way.
Give it a read, see if it clicks?
RLHF: https://t.co/GZHneBjnXU
Been digging into SFT and DPO to better understand LLM post-training a bit.
Tried writing simple breakdowns to make the intuition clearer and the math less scary.
Open to feedback, pushback, or corrections.
SFT: https://t.co/nR6p2sHP5M
DPO: https://t.co/SzkgTPzIep
@itsAdityaGoel Doesn’t work for most use cases like trip expenses, grocery shopping, or paying for family sharing plans, house utilities.
Feels like it’s only really viable at restaurants, no?
Wish I could split credit card payments directly with friends—have their share show up on *their* statements instead of adding it to Splitwise. A lot of us have the same cards; seems simple enough? Or am I missing something obvious? Would make expense tracking so much cleaner.
This thread is about me and my team's(@vntshh, @manteksinghh, @CodHeK_) approach towards winning a hackathon!
We attended Brembo's Gen AI hackathon (bay area) and bagged the grand prize of $10K
I will go into the nitty gritties of what we did during the 2 days and why we did it