Everything is going back to hiring only through your networks. In-person interviews. Meeting people at meetups and hackathons. It's really the only defense.
This mirrors the trend of public online spaces getting cluttered with slop so people are moving to group chats and invite-only communities.
Not to mention the blatant remote work fraud you'll encounter doing hiring. Someone has a really good FAANG resume with a bunch of random LinkedIn connections and no picture...you get into the interview and it's clear they're just using some Cluely clone.
Something I've been thinking more is that LLMs shut the door on new talent getting into the industry through the front door. LLMs poison the application process and the interview process and make it impossible to wade through that to find qualified people.
Thank you so fucking much!
Hey @mattbruenig.bsky.social -- I just signed up for a free trial. I'm not an attorney but am an AI engineer connected to people in labor and would like to show them.
And every new release changes this because the length of sessions increases, you aren't necessarily able to forecast how that affects margin.
The dynamics of usage change the margin a lot, how often are you hitting KV caches, how often are people bursting up demand...there's a ton there that I'm guessing there's not a solid handle on.
What are the chances the labs actually have a really solid feel for margin here? I wouldn't be surprised if they're making wild rough estimates on margin and just saying "this seems roughly like we're gonna break even..."
Are you saying that supply is growing greater than demand now, and previously there was a compute bottleneck? Or that demand is growing greater than supply now?
I'm assuming the latter.
Can you expand on this? Are you specifically worried about CS fundamentals expertise? Or something different?
50-100k? More like 2 million. You're not getting a fully functional next implementation without a team of 5 and a year.
It's a great place with access to the outdoors and is reasonably priced. Really underrated!
I grew up in the area -- it's pretty shocking how visibly left wing Spokane is now. I think a lot of people priced out of Seattle moved to Spokane.
Really sorry this happened to you. There's a big silent majority of people out there (including me) who appreciate your posts on this platform.
What's the strangest thing youve thought about?
Cool, does it only respond to you?
What can you do
I hate that Moscow is now known for the uofi murders and Christ church. Really sucks because it was a great place to go to school.
I've been shilling this everywhere but this is huge if you haven't tried it before
worktrunk.dev
Allows you to run scripts like copying env vars, running package managers etc. after creating worktrees. Also removes them in the background, I've found that removing a worktree for a large repo is slow.
My partner and I had a great time at this museum in Montreal, focused on telling the stories of people in Montreal and was pretty unique. And not a huge commitment.
maps.app.goo.gl/xFcyxYZ8LjRN...
Yeah the more I think about it you're right.
Seattle is like SF if everyone was a product manager at big tech.
The office
Ah bummer, assuming you've considered a cheaper model right?
Why does it cost so much?
Just a kernel of an idea I need to expand on!
The model can examine what was actually put in the notion database and see if it looks right and automatically refine its solution.
For example, if you're building a notion integration that writes rows to a notion database, instruct the model to write a script that will synthesize an input, run it through the code it's generated, and then read the rows in the notion using the API.
I've seen a lot of people just asking for the code for an integration and testing it manually a few times and then opening a PR.
Integrations especially are a really strong candidate for this because you can often use API endpoints on the target integration to verify that you took the right actions
I am having more success prompting coding agents recently by pushing them into a loop where they can try different inputs and read the outputs and see if there's anything strange or unexpected.
