Skip to main content
Benny Chan

Benny Chan

Hong Kong filmmaker

Aug 19, 2026

EliasHOST
11:43
How would you go and do that? How would you make an open source model as competitive as a frontier
11:50
model? So when our customer provide the data and evaluation, we have a reinforcement learning fine tuning platform where you can use the data and then you can get the model to run multiple responses on the same data.
12:08
You evaluate those responses and then nudge the model towards the direction where the responses are correct.
12:14
Now, there are a lot of nuances to that.
12:17
For example, crafting a high-quality eval is very, very difficult, making sure all the issues in training is gone so that given a high-quality eval, even if it's a very hard eval, the model can converge.
12:33
That's something Fireworks provide.
12:36
And then all the monitoring and observability goes into the system because RL is a very complicated process.

10 MINS LATER

EliasHOST
22:54
I think a lot of these models are being used for automation, but to what depth? i'm interested in understanding and do you have maybe an example that you could give where you think this automation will be so deeply embedded in a business i it maybe operates completely by itself a task as an example

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.