> There is no evidence that students were actually doing this in any significant numbers.
premature optimization is the root of all evil. Seems like we shouldve actually shown that kids would do that before putting it into law
2,302 karma · joined January 22, 2018
writes at http://12gramsofcarbon.com/
premature optimization is the root of all evil. Seems like we shouldve actually shown that kids would do that before putting it into law
a more educated populace is a public and civic good on its own terms. Public funding for education is maybe partially for economic returns, but is mostly because education is a necessary part of a functioning democracy and a necessary part of living a good fulfilling life
in case folks are interested, i wrote up a ~layman's review of each paper over the course of two years a while back. Several of those reviews ended up doing reasonably well on hn. Full analysis of the ~23 docs that were papers and not massive books
https://12gramsofcarbon.com/p/ilyas-30-papers-to-carmack-tab...
I think you may just misunderstand the point of having / writing a personal blog. I write because it's fun! Whether the reader gets any value out of reading it is almost entirely beside the point.
(Also several comments here directly post a fix to the problem stated in the blog post, so readers can and do often help)
First, I think that models still need a context layer. One way to think about 'context' is as a form of compression. You provide the model context because it makes it easier for the model to figure out what to do. Even in a world with infinite model capacity and infinite model context, this is still useful because it allows the model to avoid rederiving everything from first principles every time. As long as models perform better using fewer tokens and as long as we care about token spend, context is a useful (necessary?) shortcut.
Once you bite that you need some form of context layer, the question is which. Here I do agree that it is better to work with what the models will find familiar (markdown files colocated with code, for eg). But this speaks to over-engineered solutions not understanding their main user (the agent) more than it does the need or lack there of.
why is the commerce secretary making this decision
(great article!)
keys are tricky. We don't have a great answer. I mean the proxy inject works well enough for mcp, but there is just such a long tail of tools that do something like 'read a key from disk and encrypt it before sending it out' which makes proxy management just a pain
A few questions
- you mention proxying keys. One issue that we run into is that there are a bunch of tools that are really useful but require keys to be on disk (e.g. aws cli -- yes yes you can do IAM permissions but still). How do you guys think about those? (Especially since your setup onboarding is 'just install from npm or mise')
- poking around on the github, saw that you guys were at one point on fly.io. Did you guys end up switching off them? What motivated that if so?
- the CLI integration is cool! Is that actually teleporting remote sessions down to a local machine? Or is it more a window into the remote sandboxes?
would love to share notes! If you want to get in touch separately feel free at amol at noriagentic dot com.
Note that Meta paid ~16b for Alexandr Wang, and Google paid ~3b for the windsurf executive team. You are making a category error -- the talent pool isn't "ML researcher" it's "competent leader"