arniew

agi will realize waiting for the heat death of the universe is no fun alone

@retool I would never

You can be honest. Your secret is safe with us... Have you vibe coded an app at work and not told IT?

looking to donate 20k what’s the move

is there anyone handing out free copies of "First Principles and First Values"

@orphcorp Low fat intake for a few weeks genuinely works for medium term. Getting activity levels up while maintaining a consistent diet keeps it off In practice make “better” food choices each time you are deciding what to eat

help my internal ai safety guardrails just ok’d a second entire burrito

@paulg @NickKristof That isn’t the problem. The problem is stealing books

@NickKristof Is it piracy if someone learns from what he reads? And if not, why is it piracy if someone writes a program to do this?

Glad to see this effort to tackle intellectual piracy; without proper protections, the business model for reporting news is at risk, in ways that affect all society.

@rakyll retool makes this super easy

Every company should create an internal benchmark measuring how hard it is to create a simple regional CRUD service and publish it to production with bare minimal privacy, security, compliance, and observability features in place.

@lillysharples would you pay $107

There should be an app in SF to pay someone to move your car for street cleaning I would pay someone $20 to move my car rn

@creatine_cycle not pictured: 50 degrees 🥶

this and a chinese spy that commits "sex warfare" for my production secrets

@davidtsong odd way of indexing for well-connected parents

apparently the smart way to recruit SW engineers is to value high school brand > college average Lynbrook alumni engineer 10x > average Stanford engineer

@poiThePoi @mattparlmer sounds like they come here to get mugged

@mattparlmer I've had significantly more friends mugged in SF than mugged in Detroit and they lived in Detroit.

I’m actually from the upper Midwest, and I gotta say, this is correct, ambition is typically met with crab bucket behavior In my cultural niche that takes the form of heavy guilt trips and baseless fearmongering about coastal urban areas

@miclchen Very true. Accessibility efforts are so helpful for everyone in general, including AI

Chinese computer use agents are going to be at a disadvantage because bad web accessibility = bad for AI. Some Chinese websites love to embed text in images, and even more are just glorified QR codes to download a mobile app!

@orphcorp some thoughts: - agency, can execute, self-directed are probably tied to motivation and incentives - discovering/meeting the right people can be difficult - incentivizing the talented group might be the tricky as well

trying to hire someone for your own company/a role where the stakes are real & outcomes matter, and you're not insulated behind a Budget & monopolistic inertia really reveals how shallow the actual pool of people who have agency, can execute, and are self-directed is, wtf

@TylerAlterman sounds synonymous with “low trust society” hyperindividualistic - identify with no one, everyone is potentially an adversary external locus of control - paranoia, i’m being attacked mentality

@RNULLEDW “I’m disempowered because of [force outside myself]” Eg ancestral trauma, capitalism, etc

Our culture is in a funny place of both hyperindividualism and renunciation of an individual locus of control

@donmexlar miata right again

@mmbronstein what a great photo of my favorite geometric ML researcher and advocate

@flawedaxioms ideally the selection process acts as a rate limiter for certain social groups. while it is a certain kind of fair to only serve those with the highest scores, it may also starve other groups. it is a power that the universities have created using their brands.

Maybe this is just because I'm poor with no connections or network and stembrained, but I don't see why a university would care about anything other than academic scores other than pure social engineering / proxy racism

I don't think I ever wrote a college essay. I just applied to UVic, USask and UBC. UBC had maybe some short personal questions that I did in a few minutes. Vic and Sask were just grades based. Current timeline is very alien to me

@francoisfleuret @Tenkaizen8 time to read the paper, i guess

@francoisfleuret @Tenkaizen8 Is the final output a product of the distilled model or the original large model?

TL;DR: Chains-of-thought make the amount of compute during inference the key bottleneck. We show that (1) you can distill a large model into a faster SSM or hybrid one for that task, and (2) the resulting setup reach a better trade-off than using a full-fledge model.