Yulia
@vveiln
Followers
388
Following
340
Media
13
Statuses
190
I'm fine with having a boring bio. The comet above is Lovejoy. I do things at @anoma
Berlin, Germany
Joined December 2011
I find it even worse for coding. I don't know if imitation of activity behavior can ever become good enough to perform somewhat complex tasks and it is terrifying how many people find the current state enough to slowly give LLMs control over their lives
0
0
2
The rest of the time I either use it as a rubber duck, mostly ignoring the responses (to even have a chance for helpful response I would have to load way too much context) or as a search tool
1
0
0
One thing that using LLMs really help me with is naming, and even then it is rarely a smooth experience. Though picking from many options is better than struggling to come up with any options.
1
0
0
1) Delegation of thinking may seem attractive but I would rather keep what is left of my cognitive function after staring at dopamine-inducing interfaces all day every day 2) Using LLMs without delegation of thinking is a chore (because it always fucking lies)
this is in fact true. MIT did a complete research on the effect of AI on your cognitive abilities and i’ve never looked at AI the same way since then: > LLM use accumulate cognitive debt > the more you rely on AI the worse you get at thinking without it > you stop exercising
1
0
2
Mandatory off-the-screen-and-touch-grass hours for everyone working in tech
0
0
1
I’m scared of what’s up there. Once you close up all the “simple” bugs (overflows, simple logic bugs) I doubt the vulnerabilities go away. But then you reach the places where the bugs get harder to even understand.
5
5
26
That feeling when you try to design smth, pick a primitive that could work and end up trying to use it in a way that breaks its founding property. Building a programmable privacy system in the world of no accessible programmable crypto is like going for a run on a minefield
0
0
6
I love this thread
@DigiratiGotti @BagsApp Judging from the 10k @ mentions I’ve gotten in the past few hours, I can tell what an earnest community this is that genuinely wants to support me and my work — without even a hint of grift! I checked my Venmo, though, and haven’t gotten anything yet?
0
0
1
Who certainly isn't gonna suffer from the carefree adoption of vibe coding for production code is bug bounty hunters
0
0
1
How much time do you think will pass between quantum computers becoming strong enough to wreck modern cryptography in a meaningful way and people learning about this?
1
0
0
I find it curious how mechanism design - a field of statements about human behaviour and incentives - doesn't imply studying psychology but only economics, game theory, and such. Human interaction is the foundation of everything. Understanding each other is something to learn
3
0
5
And then these snarks get even further away from each other being separated by domains and tooling around them We need runtime trade-offs programmability with a simple stable interface that we can build against
An irritated rant: There's a million snark papers out there with some variant of - split your witness into a matrix, commit separately to each row, and then you can get various proof size verifier time tradeoffs for inner product, maybe by also doing it recursively. The original
0
0
3
Trust doesn't evaporate; when we make something trustless, we assume trust somewhere else. It is good to be aware where the trust assumptions live
1
0
2
It was such an energy boost to see all these happy sleep deprived people presenting their projects they built from scratch in a couple of days
@BerBlockWeek @leonardoalt @georgwiese @pop_eax @NoirLang @aztecnetwork @gabrielaxyeth @ZKPassport @therealyingtong @nico_mnbl @AndrijaNovakov6 @mmkostrzewa @RomarioKavin1 @hyli_org Huge thanks to the 20(!) amazing judges that were with us throughout the weekend. Thanks for your invaluable insights in judging such a great set of projects! @AnnaRRose (@zeroknowledgefm) @aurelcode (@inversed_tech) @billyrennekamp (@triflelife) @colludingnode (@Celestia)
0
0
5
be careful what you wish for, lest it come true
0
0
2
Today's edition of underrated products: Excalidraw Simple yet powerful enough UI, seriously improves communication of ideas without too much overhead
0
0
4
I am skeptical about learning from AI and reading papers with AI. It makes mistakes, and not the kind of mistakes people make. If you verify the output, then it can be useful, but verifying cancels out the efficiency benefit. Good thing I don't mind being inefficient.
2
0
6