/biz/ - Business & Finance

Data is big money and it's here to stay, and AI is going to make it much more efficient gathering data. So naturally there's going to be a need to assuage peoples fear of it being gather and used, and to avoid massive fines. Oasis does this.
Dawn is an expert as well in adversarial machine learning that will help further securing it against new attack vectors that follow while the tech progresses. So the data is kept both secured by machine learning through adversarial machine learning research, stored in a secured silo and protected by differential privacy. All of this Dawn has spoken about and is available online.

>>53636457
>SHAMELESS REBRAND

>>53636493
Argue against it, jeet.

>>53636457
Market your shitcoin somewhere else Ranjeep.

This coin has been a swingers paradise. 6c -> 7.5c -> 6c -> 7.5c in the last day has been pretty fun.

>>53636493
>rebrand
Same "branding" since Ekiden paper in 2018. Oasis was always about machine learning (among other things).

Broes I got 20k bags what am I in fer?

>>53636493
>shameless rebrand
partnership with meta AI goes back to 2019, meta tweeted about oasis last year. this is one of the few projects that actually have anything to do with AI, and ironically one of the last to jump on the hype bandwagon

>>53636576
Did you miss the bottom?

>>53636618
Comfy gains

>>53636457
Yes

More about Dawn's work with AI, don't the fudders scare you.
https://www.youtube.com/watch?v=eMh5YqKopjE
17:03 Dawn’s research and techniques on how to build resilient machine learning systems. Dawn talks about how they studied, how vulnerable they are, how they can be attacked, and what sort of defense can be build (Dawn has also talked about this on Lex's podcast, go to the relevant timestamp for adversarial machine learning 16:50) and in their recent work, they show small modifications can fool deep learning classification system. The stickers on signpost to name the more prominent example. Dawn answers how attackers no longer need high end access to attack the machines. Black box attacks; IE attackers doesn't need to know the architecture of the neural network in the deep learning system. There are two different types of black box attacks. One is where you don't know anything about the victim model, and don't even have query access. Two is where the attackers doesn't know anything, but does have query access. In both cases, attackers can still launch successful attacks, where it fools the machine learning system. In detail about the first, one technique ensemble. Attackers build themselves white box models, for say computer vision, image classification. And then use these ensembles of the white box models to craft adversary examples that can fool the crafted examples, and then transfer over it to the black box. And if the attackers have query access, they can make the attacks more effective, where every information gained is utilized to effectivise the attacks, and adapt to it, and then become successful in such a manner that the next attacks only require a small number of queries.

22:00 Dawn is asked about the defensive side. It's very challenging, with no general solution. One defensive approach might be, which is consistency checks. Which is to look at segmentation. As humans we can look around, separate between different objects. If we see an image, and we're then shown different sub patches, We'll still give the same segmentation. So even if the sub patches of the image is segmented, our understanding of it is consistent cause we've seen the image. So by forcing consistency it makes it harder for the attacker; the attacker can
generate one adversary example to give a wrong segmentation, but if you give it a patch it will give different solutions to the segmentation. So it becomes difficult to be consistent. Not only has the attacker fool the segmentation system once for one image, but also as the deep learning system looks at the different patches of the system it needs to be fooled in a consistent way. This was an effective solution, but more study of it is required.

REBRAND HAHAHHAHAHAHAH

25:31 Dawn’s work in AI and data privacy. Security and privacy can help in machine learning, cause the main power is data. A lot of the data though can be very sensitive, users images, text messages, or personal information. So users privacy needs to be protected. So in collaboration with google if you chain a language model from emails, the original emails, that contained sensitive information from the user, the attack can by querying the trained language model can extract the original sensitive information.
The solution to this problem, was instead of training vanilla language learning model, you train instead a differential private language model. Meaning you add noise to it. (A basic description of the data, while detailed information about the data is withheld. Dawn's model of TEE (see the Cota capital interview from; 17:27) intercept the individual data, adjusts, adds noise and thus the query for the information returns only a summary of it.)

>>53636457

>>53636836
imagine being this retarded when information is literally being spoonfed

27:49: How deep learning can help computer security. Security has a lot of problems, one of the recent works that Dawn did was for IoT devices (note; video was published 28th of September 2018). A lot of these devices uses open-sourced code in their firmware. A lot of these codes, contain vulnerabilities. People discover that the vulnerability gets included in the IoT's firmware. Detection needs to be done, to check if the firmware contains a vulnerable piece of code. The source code for these devices may be the same, but the part in the hardware, and often only the binary code is available. The way to do the similarity check (IE checking to see if the vulnerable code matches the simulation), was to convert the binary code in to code graphs, then checked to see if the graphs were similar. And thus they could detect vulnerabilities in IoT devices.

>>53636685
no, bags are packed. it's a no-brainer, rose is a perpetual hype machine. every new trend, every new cool tech can and will be implemented as a paratime. too many potential bullish catalysts to risk it:
>ai narrative
>sapphire launch(the first flagship network built on top of oasis)
>oasis privacy layer - any evm chain will use this to leverage sapphire's privacy features, there's no current alternative, and won't be for a long time
>trustless bridge to ethereum, basically implementing the zkBridge paper, which will run on oasis, and no this isn't just an asset transfer bridge, this will allow for message passing/data sharing too
>equifax decentralized identity(smpc+sapphire)
all of this in 2023. not even mentioning the countless other things oasis labs is working on

No excuse now "I couldn't get past her accent" you have the video summarised. Big brains involved with AI follows Oasis Labs and dummies here think it's a rebrand. You can start buying, and begin blooming, or you can wilt like so many others, the choice is yours.

WAGMI

Honestly kinda amazed, not only at how dead /biz/ is overall but how dead ROSE is as a proportion of discussion. It feels like people really need price action nowadays to tell them what to think, the people have changed.

>>53637125
rose sucks Dawn Song's hairy ballsack

is it just me or is that a double top? on the hourly? you guys are about to get rekt.

bump

i told you. you should've listened to me.

REEEEEEEEEEEEEEBRAAAAAAAAAAAAAAAAAAAAAAND

>>53637125
>>53636922
>>53636890
>>53636846
>>53636831
>>53636819
>>53636457
>>53636471


Thanks op for this.

I did listen to this podcast and kinda understood like 50% of it, so it helps you breaking it down more after all these months.

I also wanna just say i appreciate you /biz/ bro there are a few of us still around trying to help some poor loser make it

Data is the bedrock to expanding the utility and value of AI. As AI continues its rise to prominence in the next decade their will be a growing demand for access to privated/siloed data. Oasis and Dawn Song are the clear leaders in this field which is made clear with their enterprise partnerships with projects like Meta AI and BMW. It is amazing that the average crypto cuck can't see this sort of thing, but I guess all that matters to these redditors is price action and hype. In the end the superior technology and team that is focused on long term use cases vs short term price action will win. Paying for partnerships and obscene grant incentives is not a sustainable business model long term

is 100k still the make it stack?

100k is Sui stack
1M is make it stack

>>53639677
>>53639620
1mil sui
10mil make it
always has been. check the archives if you don't believe me. you can find a ton of threads of anons saying 1mil sui.

Lmao rose cucks rebrand

>>53639985
>a bunch of tranny jeets saying it makes it so

Pic related

They were exposed as SCRT shills and they all came from the same Discord. You're either one of them or you're an idiot who fell for literal tranny propaganda.

>>53639985
There never were a one sui stack, some people said 1M some 100k, i remember morensaying 100k though and i first bought it 2 years ago

Sui stack costing 80k at ico is a bit much imo

>>53636922
Ay, hvad arbejder du med?

a lot of words ITT

>>53636636
>MUH META

Rosechuds are still clinging onto this cope? Top lel.

>>53636922
dear god that accent though. thanks for the spoonfeed

>>53636457
Doing gods work anon! Thank you

>>53637125
TEE is a dead end, that's why no one is using Rose
lot of hoopla about a flawed yet expensive solution. TEE is vulnerable, end of. Until they find something better it's dead in the water

>>53636457
People still don't look deeply at privacy protocols. I'll add Rail and Rose to my list.

>>53641770
try again, the ones that attacked secret remain bullish and mentioned Ekiden. A paper co-authored by Ari. Buy ROSE and bloom, or remain unblossomed.

Where did all the tranny fudders come from? It was never this bad.

>>53642566
scrotum fans most of them, though they've been btfo by the ic3

>>53641770

So I think this is a good topic. Many people in web 3 look at confidential computation and privacy through the lens of anonymity and anonymous transactions. There are other aspects of confidential compute though which are valuable such a confidential smart contract state. Essentially the ability to create smart contracts which hide parameters from network to improve the sophistication of the smart contract. This is the area that the Oasis Privacy Layer and the Oasis Sapphire network can really empower web 3. In these areas TEE risks are relatively low (the incentive to attack is not high), however the increase in UX is extremely useful.

Gaming- almost all compelling traditional games require some aspect of hidden state in the game. In card games you must not know the composition of the deck or which cards are in the other player’s hands. In strategy games there is fog of war, hiding certain areas of the map from player view. Role playing games often incorporate puzzles/riddles with hidden solutions. These features are impossible on 100% transparent blockchains today, meaning that in order to incorporate these aspects into game play they must be taken off-chain, and thus losing the high-integrity and trustless guarantees that makes a game web 3 in the first place. TEE networks offer the ability for games to no longer have to make this trade off. They can have confidential state while also being entirely on-chain. This improves the sophistication of what is possible with smart contracts without really being at risk of being exploited. Performing an Aepic attack to view the fog of war in a strategy game is pretty useless. The end results being that these TEE based networks offer vast improvement in the sophistication of web 3 gaming, while providing high confidentiality guarantees, and even in this instance of an exploit would still behave exactly how every other blockchain network currently does, so there is only upside and no downside

>>53644470
Also there is the very important point of "once the game is over, the data you are leaking is now worthless." There are tons of applications where leaking the encrypted state a year after the fact doesn't matter, you need confidential integrity for only a short period of time. For stuff like on-chain KYC with social security numbers, yeah you need top level guarantees, I wouldn't personally trust my info on a TEE in that case. But for a DEX or poker game I don't give a shit about a possible TEE exploit, it mitigates MEV or cheating or whatever immediate issue would be at hand in a transparent chain.

>>53644620
Exactly. There are many use cases where it allows you to do things you simply couldn't on other chains, while not really being a security risk. Unfortunately this is lost on most people. There is so much increased utility for things like Gaming, MEV, NFT auctions, Confidential DAO voting, etc.

Is the tiny pump Rose had over the passed couple of days a sign that the market as a whole is going to dump?

what are your price targets for this year anons?

>>53646156
7c

>>53646167
thanks and good luck