suriya-ganesh

I've been working in the privacy space for ~4 years now and have worked across, homomorphic encryption, Federated Learning and finetuning.

I've also run multiple user studies on privacy during my time doing a master's in privacy at CMU. almost 10/10 times users will choose, easy to use and accessible over complex and more control over their data.

In fact, for one of the Mag7 we proved that, users only want to have the feeling of safety and don't really care what gets done with the data.

very unfortunate, but this is the world we live in.

I see this too often. But, realistically users do not care about the harvesting as it is unseen and behind the scenes. Most people just want get stuff done in a competent, fast and easy-to-use operating system.

>It also locks you into the cloud services of an advertising company

this is pretty much any company these days. microsoft is guilty of the same.

incredibly impressive demos. I wonder how the training data for these models look like?

is it separate batches of special "skills" that are added post training? how can they guarantee the models won't eventually lose a skill?

it's been a things since 2002. but it's a news aggregator not directly competing with newyork times

there has been a product called Google News since 2002. It was only aggregating information from news channels

Will the big labs leave anything for external competition?

This probably killed a thousand startups in this space.

in the early internet you wouldn't see google creating their own news site or facebook building their own animal farm. what happened to platformication of everything?

i suppose a quick test would be getting the model to write down Micken's essay end to end.

if the original essay was stuffed within the prompt window. the result will be word accurate.

unless this is a model trained specifically on Micken's essay (which claude is not).

This is unlikely. The way model distribution works is that the model retains a lossy representation of James Micken's writing. Very likely, it cannot repeat Micken's writing verbatim. Neither can it reason about the training cutoff in this manner.

It's a lossy representation

this is definitely not true.

with AI agents, you're obtaining a mildly lossy perspective of the code itself. whereas if you wrote it by hand, you'd have a more concrete understanding.

This is not too different from an engineering manager directing junior developers.

The stereotype of the engineering manager who forgot to write a line of code is not wrong.

since tbpn is known for their quite oblique satire. i wonder if this is some long April 1st thing.

also it should be noted that LinkedIn had a 5 year plan of migrating everything to azure but abandoned it after a year.

I've gone through this process and is this not a failure from the institute that are giving away these certifications for a fee without any due diligence?

intermediaries like delve have only amplified this failure.

it was obvious to anyone who was involved in this industry that, all of this is just security theatre with nothing really to back it up.

parking is another $200.

not to mention the tolls.

my point is, ~$1500 is grocery for a month for us. why would i spend that much on just a car?

I recently moved out of new york to new jersey. don't have a car.

my options now are either buy a car, pay insurance, maintenance etc. north $1,500/mo for the convenience of going to the grocery store or, get a subscription for all these services ~ $200/mo. and save myself the time spent back and forth doing grocery shopping. I cook 90% of my food.

not all solutions involve owning a car.

I wrote that PR and can guarantee you that it was not vibe coded.

the reason why I thought this PR was interesting is that, it only needed ~ 4 lines of real change, everything else was cascades due to that change. and the performance did improve with that "pointless formatting and whitespace changes"

ah! sorry. I meant to say, "the interesting part for me was that we removed code to improve performance."

I don't know what to say.

I did my graduate in Privacy Engineering and it was just layers and layers of threat modeling and risk mitigation. When the mother of all risk comes. People just give the key to their personal lives without even thinking about it.

At the end of the day, users just want "simple" and security, for obvious reasons is not simple. So nobody is going to respect it

It is always a config problem. somewhere somplace in the mess of permissioning issues.

I ended up at LAX 5 am for an 8 am flight. nobody in the whole airport, no TSA nothing until 6 am.

i took a very uncomfortable nap on the floor that day.

Interesting take, but this feels like one of those tarpit ideas that YC discourages their portco to start attacking.

Guaranteed this is going to attract a ton of abusers who are looking to use this for signing up to services, spamming or other nefarious purposes, which then blacklists the doman. This is an infinite whack-a-mole.

do you guys have some ways of handling it?

but this is not a technical attack that returns the metadata.

much more closer to the $5 wrench attack

https://xkcd.com/538/

This is verifiably false.

You still need to do most of the grunt work, verifying and organizing the code. it's just you're not editing the code directly. Speed of typing out code is hardly the bottle neck.

The bottleneck is visualizing it and then coming up with a way to figure out bugs or add features.

I've tried a bunch of agents, none of them can reasonably conduct a good architectural change in a medium size codebase.

call me a cynic. but I am pretty sure this is a mostly AI written repo.

> Learning is not supposed to be fun. It doesn't have to be actively not fun either, but the primary feeling should be that of effort... - karpathy[1]

And this honestly is low effort. all commits only have hours in between them

[1] https://x.com/karpathy/status/1756380066580455557?lang=en

Google is in a much better spot to filter out all AI generated content than others.

It's not like chatgpt is not going to cite AI videos/articles.

layers and layers of security practices over the past decade are just going out the window so fast.

It's quite wild to give root access to a process that has access to the internet without any guardrails. and then connecting all your personal stuff on top of it.

I'm sure AI has been a boon for security threats.

> the code is going to be relatively small and I do have doubts I can easily check

Continues to make an app with 150K lines.

>Yegge is leaning into the true definition of vibecoding with this project: “It is 100% vibecoded. I’ve never seen the code, and I never care to.”

I don't get it. Even with a very good understanding of what type of work I am doing and a prebuilt knowledge of the code, even for very well specced problem. Claude code etc. just plain fail or use sloppy code. How do these industry figures claim they see no part of a 225K+ line of code and promise that it works?

It feels like we're getting into an era where oceans of code which nobody understands is going to be produced, which we hope AGI swoops in and cleans?

This is an unfair comparison.

using a transport serialization and deserialization protocol for IPC. It is obvious why there was an overhead because it was architectural decision to manage the communication.

I guess the old adage of if something goes 20% faster something was improved if it is 10x faster, it was just built wrong is true here.

Chess being popular is mostly because FIDE had a massive push in the last decade to make it more audience friendly. shorter time formats, more engaging commentary etc.

While AI in chess is very cool in its own accord. It is not the driver for the adoption.

This is such a cool concept. props to you for building it.