People are model training more often.
@harvey actually trained its own model a few years and now they've started to train models again.
This is mainly due to the industry shifting from "how do we get intelligence into the market?" to "how do we deliver intelligence as cheaply as possible?"
As inference costs become more important, relying only on third-party models becomes less nice.
Training your own model means you get more control, cheaper price, and better performance.
Expect more companies to follow the same path.
@aaliya_va This is a real pain! Agents spend a lot of time & tokens on rereading the same files and repeating the same mistakes.
Another open-source project that helps me:
Codaph captures prompts, reasoning, and diffs across sessions and improves agents over time
https://t.co/375iNfDO6M
London builders, weβre running it back π¬π§
July 1st is the 3rd Codex Community Meetup London
Expect:
π OpenAI + Codex updates
π¨βπ» community demos
π» hacking time
π swag + prizes
A room full of agentic engineers
The first two events had 1,000+ registrations and packed rooms (oops... working on bigger venues π) This one ere gonna get as many people in as possible
If youβre using coding agents, building with them, or just curious where this is all going, come hang.
Waitlist usually moves fast, so grab a spot early π
cc @OpenAIDevs
P.S. Codex made this video with @Remotion and amazing pics by @WillHoldswxpna + others
I can't take all the credit! The crew that make these community events happen π«Ά
I appreciate each one of you, so much
thanks for being crazy with meπ€