Andrei Lapets

36 posts

Andrei Lapets

Andrei Lapets

@andreilapets

Boston, MA Katılım Kasım 2015
58 Takip Edilen150 Takipçiler
Andrei Lapets
Andrei Lapets@andreilapets·
Once the ecosystem has gained experience from building many such ensembles (and filtered out the ones that either aren't a fit for popular use cases or do not offer benefits users value), everyone can begin factoring and identifying the common ones. Those will become the building blocks that future PET compilers emit.
English
1
0
1
11
Andrei Lapets
Andrei Lapets@andreilapets·
This commentary captures well Nillion's position with regard to relying on any specific PET. Each differs from the others not only in performance but (sometimes subtly and sometimes dramatically) in the privacy benefits it can add to a use case. In fact, we have learned over the years that most PETs are qualitatively a poor fit for most use cases! We believe custom-built PET ensembles are the only viable way to address use cases for at least the next few years.
English
1
0
1
23
John Woods
John Woods@JohnAlanWoods·
FHE coming to Nillion...
English
20
13
106
14.4K
Andrei Lapets
Andrei Lapets@andreilapets·
Multi-party computation is a compelling example that could support this paper's argument. While a secret-shared lookup table split over two machines light-years apart can reply intelligently to queries from a midpoint, the machines' contents would be seen locally as random noise.
Alexander Lerchner@AlexLerchner

I highly recommend reading The Blind Spot for anyone interested in why observer-independent computation is a category error. And if you haven't read the preprint yet, you can find it here: philpapers.org/rec/LERTAF

English
0
0
0
101
Andrei Lapets
Andrei Lapets@andreilapets·
Privacy-enhancing technologies will not be worthwhile differentiators if users can't confirm apps use them and apps can't credibly advertise them as a feature. Starting today, anyone motivated to help address this gap and to make PETs more competitive can run a Blacklight node.
Nillion@nillion

Nillion Blacklight is a decentralized network that verifies private workloads running on the Blind Computer. It checks that workloads are running correctly and honestly inside secure enclaves (TEEs), without revealing user data or execution logic. This verification is done by Blacklight nodes.

English
0
0
5
311
Andrei Lapets
Andrei Lapets@andreilapets·
APIs transformed the developer experience by standardizing how apps share and leverage data. However, APIs only indirectly involve and benefit the users who subsidize this ecosystem by contributing their data. BPIs can act as a gradual path towards a more user-centric ecosystem.
Build On Nillion@buildonnillion

x.com/i/article/1956…

English
1
3
17
1.2K
Andrei Lapets
Andrei Lapets@andreilapets·
Nearly a decade after the vision was first articulated, a modular web-friendly set of PETs software tools that lets developers disentangle component roles/functions in apps and workflows (thereby allowing them to minimize exposure of user data) is taking shape at @nillion.
Build On Nillion@buildonnillion

x.com/i/article/1950…

English
1
1
13
1K
celestineia
celestineia@celestineia·
I just love how everything is so polite and courteous, asking whats up and shaking hands and all. People should be more like dial up connections.
English
4
8
49
1.2K
Andrei Lapets
Andrei Lapets@andreilapets·
@CryptoMeina @juanaxyz00 Targeted advertising campaigns created in response to the online/offline activity of millions of people can already cause feedback loops that cause them to buy unhealthy food products, for example. Now this can happen even more quickly and with even more steps automated.
English
1
0
2
46
Meina 🌱
Meina 🌱@CryptoMeina·
@andreilapets @juanaxyz00 I thought about this for a while - maybe we are all just data points for robots' ethical dilemmas? x.com/CryptoMeina/st…
Meina 🌱@CryptoMeina

Some fleshed out thoughts: I’m increasingly concerned about AI’s trajectory. It already processes information faster than our brains, and as its compute power grows, it will inevitably surpass us in total knowledge. At some point, an AI system might judge that humans consume more than we produce - so what justification would remain for our existence? Based on what we have seen so far, our unique value lies in the moral judgments we make when faced with difficult situations. Those chaotic data stabilize the system. So, maybe we are all just data points for robots' ethical dilemmas? Consider the Zeroth Law of Robotics, which states that a robot may not harm humanity—or, by inaction, allow humanity to come to harm—and that the welfare of all humanity takes precedence over any individual (or the robot’s own survival). If an AI truly internalizes this law, will it protect us, or will it decide that the best way to minimize harm is to sideline or even eliminate those it deems detrimental to humanity’s overall welfare?

English
2
0
2
81
Andrei Lapets
Andrei Lapets@andreilapets·
One thing not explicit in this discussion is the privacy risk of a prompt for *others*: family members, friends, colleagues, and acquaintances. We may trust an individual to make informed choices about themselves based on risk feedback (though humans clearly have trouble with this), but do they trust everyone around them to do the same? Do they trust themselves to have the best interests of others in mind when evaluating such feedback?
English
1
0
1
49
Dave
Dave@dave93xyz·
Interesting. We had sort of discarded B2B as the sales cycles for "privacy tech" are long. But if the company is actually still getting the full functionality of openAI models, then there is not much risk for the company...just an extra layer of privacy and no trade off in productivity for their employees.
English
1
0
3
62
Andrei Lapets
Andrei Lapets@andreilapets·
Even supposing privacy feels unimportant at the individual level, there is the question of how this impacts society collectively. We can't be sure there isn't a way to exploit aggregations of small, innocuous facts about the daily routines of millions of people. Without private spaces (to do mundane, unusual, or subversive things), society as a whole becomes a more homogenous and less diverse system, potentially too fragile to handle unforeseen disruptions (including deliberate or malicious manipulation).
English
3
0
4
590
Meina 🌱
Meina 🌱@CryptoMeina·
tbh, I get the worries about feeding data into AI models, but the upside I’m getting still outweighs my privacy concerns. - I’m already filtering what I feed the model (detail-level, context, etc.) - I cross-check different models to see if the answers line up. - Data privacy is an endless game. Sure, you can run a TEE, store stuff locally, even train on your own server—but someone can always hack you. - The real question: what level of privacy do you actually need for the info you possess? What are the potential impact of that information? - Do you really need to encrypt what you ate today or who you argued with yesterday?
English
1
0
4
200
Andrei Lapets
Andrei Lapets@andreilapets·
The latest release of @iabtechlab's PAIR standard addresses industry feedback and improves interoperability with digital advertising solutions like @prebidorg, showing that #privacyenhancingtechnologies like #privatesetintersection are finding adoption in real-world workflows.
IAB Tech Lab@IABTechLab

#IABTechLab is excited to release the latest version of the open PAIR 1.1 protocol! This version clarifies the definitions of several of the terms in the protocol and improves the integration with prebid with a new Open PAIR module. Read more here: okt.to/IXvS7N

English
0
0
4
433
Aztec
Aztec@aztecnetwork·
will RT whatever the top reply to this post is (as long as it won't get the intern fired)
English
495
867
1.2K
48.8K
Andrei Lapets
Andrei Lapets@andreilapets·
It's great to see that the @FTC is examining #multipartycomputation and other #privacyenhancingtechnologies in enough detail to articulate some of the challenges associated with deploying them in production (and citing the @BU_Computing/@thebwwc wage equity effort as an example).
FTC@FTC

While an important tool to prevent intentional or accidental misuse of data, companies making representations about their use of privacy enhancing technologies must continue follow the law and ensure that any privacy representations are accurate: bit.ly/4bn3Ufq

English
0
0
4
857
Andrei Lapets
Andrei Lapets@andreilapets·
@kayabaNerve @AvishaiY @nillion We are assembling a small interest group on the topic of non-interactive ITS MPC at correlation.org. Feel free to reach out if you think there is an overlap (or any related topics/ideas that would be exciting to discuss).
English
1
1
3
114
Luke Parker
Luke Parker@kayabaNerve·
@AvishaiY @nillion I'm eager and willing to be a class group shill given my own experiment on a potentially-robust linear-complexity tECDSA signing protocol, which I'm largely review begging for, depending on them.
English
1
0
0
169
Avishay Yanai
Avishay Yanai@AvishaiY·
The claim by @nillion that has raised the most doubts, is that of an IT-secure MPC with a *non-interactive online-phase*, which apparently (as mentioned in x.com/GarbledCircus/…) contradicts a lower bound. >> That claim was recently revised >>
Avishay Yanai tweet media
Mike Rosulek@GarbledCircus

MPC fans might find this interesting: @nillion seems to be claiming information-theoretic MPC with *constant-round online phase* with preprocessing proportional only to circuit size (or maybe only input+output size?). 🤔

English
4
3
29
6.8K
Andrei Lapets
Andrei Lapets@andreilapets·
@AvishaiY @nillion To evaluate an arbitrary circuit using Nillion's technique, the circuit would need to be converted to a possibly exponentially larger multivariate polynomial ("sum of products"). While this will not always scale, it is notable that this conversion adds no rounds of communication.
English
0
0
7
127

Keşfet

@nillion @celestineia @CryptoMeina @juanaxyz00 @iabtechlab @prebidorg @punk6052 @jambutties