Perhaps then inventors of promising ideas should make multiple attempts at popularizing their ideas if they care about association, multiple attempts at explaining why the idea is important and demonstrations of killer applications.
FWIW I think LangChain has evolved a lot and is a nice time saver once you figure out the patterns it uses. The LangSmith observability is frankly fantastic to quickly get a sense of how your expected LLM flow engineering ends up working out in practice. So much FUD here, unwarranted IMO. Don’t forget, reading code is harder than writing it, doesn’t warrant throwing out the baby with the bath water. Don’t fall for NIH :) Haven’t had issues running in prod recently either since they’ve matured their packaging with core/community/partner etc. For agentic use cases look at LangGraph for a cleaner set of primitives that give you the amount of control needed there.
It makes no sense to estimate the total cost of the proprietary equivalent of _all_ that is currently OSS at $177M. It would be spread over at minimum thousands of companies and each company would try to get their margin, needs to be rewarded for the risk they’re taking, etc.
The HBS method to get to 3.5X isn’t sensible (as the author points out, not everyone would build) but the truth is somewhere in-between.
The COGS of software would be significantly higher if there was no OSS. But everyone knows that already. I don’t think any new information has been created here.
I really enjoy tinkering with LLM outputs that generate code that can be executed directly. Especially the faster models like GPT-3.5 Turbo are a joy to play with.
Am I the only one surprised that the author of einops is looking for work? In an era of an AI arms race between many big labs? If you’re rolling your own networks, I’d definitely reach out to this guy!
Great analysis on the value of collecting and curating knowledge, even without synthesizing it into "best practices." As an AI engineer I collect a lot of resources here, hoping it helps someone https://codingwithintelligence.com/
The difference is packaging it as a consumable PyPI package that can easily be used in a project (they even call out for separating this out into a stand alone project but that they lack the time to do so: https://docs.sweep.dev/blogs/chunking-2m-files#future- )
In addition, we expand and fix the implementation, for example it now supports limiting on token count instead of character count, and we fix some white space inconsistencies in parsing/chunk reconstruction.