正在加载视频...

视频加载失败

Hot take from my panel on decentralized AI at UC Berkeley: Your data is worthless. The idea that you'll get rich selling your data to train AI is a fantasy. The frontier is high-quality synthetic data, not your dumb tweets. Second hot take: the government censorship of AI has...

81,172 次观看 • 1 年前 •via X (Twitter)

35 条评论

Ben Fielding 的头像
Ben Fielding1 年前

strongly agree that you won't get rich selling your data but synthetic isn't the answer, it's part of it the other part is going straight to the user interaction and capturing that, unfettered by labelling incentives analogous to unsupervised + supervised learning

Andrea 的头像
Andrea1 年前

i appreciate your take, but just for the sake of discourse: - synthetic data (continuously) needs real data to avoid drift, perhaps in fewer quantities, but ground truth will always have value. - i actually agree most online data is useless at this point, (we can safely assume the whole itnernet has been crawled many times over), and thats why "dumb tweets" is a strawman example. - proprietary data can be very valuable and AI companies are paying millions of dollars for it every day, especially data that cant be scraped or obtained easily. - like in the audio plugin world (i.e. shit in => shit out), synthetic data is bound by the universe that it has seen and its often in the edge cases that the highest signal comes from - totally agree the value of single data points is near zero, but when aggregated (and high signal enough) it can be worth a very pretty penny and we're seeing this first hand.

Magnus 的头像
Magnus1 年前

A facebook user's lifetime data on their platform is worth up to ~$2,000 in developed economies. You can't get rich selling your data individually but in aggregate that's many billions to sell, which is why Facebook and Google are such valuable ad businesses.

barthazian.eth 巴特 的头像
barthazian.eth 巴特1 年前

your telling me dipwheelers tweets about generational bottoms is invaluable blasphemy

Alex Shevchenko 🇺🇦 的头像
Alex Shevchenko 🇺🇦1 年前

Agree on the first point—models can already infer plenty about a user from their social graph, and they’ve been trained on that data already. But I disagree on the second. Bitcoin’s success wasn’t driven by the crypto-anarchist pitch. It was the ‘digital gold’ narrative: scarce supply, divisibility, easy storage, and transferability. Decentralized AI will win for the same reason—not by selling fear, but by enabling more. That’s the power of open source, paired with the ability to monetize it.

Jongwon 🇰🇷 的头像
Jongwon 🇰🇷1 年前

Repetitive & predictive data is worthless. But on-demand manual data has value in aggregate, albeit with high decay. Tweets are worthless. But if you capture the process of converting thoughts & movements into tweets at high fidelity, that's extremely valuable data, for now.

Ram Ahluwalia CFA, Lumida 的头像
Ram Ahluwalia CFA, Lumida1 年前

Take a look at Reddit licensing deal. Broadly agree, the users won't capture value here, the platforms will.

xTAO - a Bittensor company 的头像
xTAO - a Bittensor company1 年前

$TAO 👀

leochen 的头像
leochen1 年前

Really 🔥 take: Your data becomes more valuable when needed and aggregated as a dataset in scalable, decentralized infrastructure. Of course, an individual data point is nearly useless in a quantum world; only probability matters with a large dataset.

avious 的头像
avious1 年前

excited to listen to this for folks curious about synthetic data, there's an accessible ep with the deepmind genius behind alphago and now alphazero the tl;dr is giving human data to a model actually made it worse lol

Nikita Varabei 的头像
Nikita Varabei1 年前

People will be fighting to give up their data to AI if it brings attention to them

Dabba Network 🟨 的头像
Dabba Network 🟨1 年前

Decentralized AI is the next battleground for freedom

will p. 的头像
will p.1 年前

Had a portco pivot from decentralized data streams to AI swarms, because data was too low quality. That's how agent swarms started, as their code was quickly copied by other projects with no attribution. Same team still shipping with @mirra

Gary Basin 的头像
Gary Basin1 年前

Yep. But most people’s needs will be met by local models. Bottleneck will be search

Andrew Verbinnen💈 的头像
Andrew Verbinnen💈1 年前

Biological data whole different story imo

Poly+predictour.tanay's 的头像
Poly+predictour.tanay's1 年前

Bitcoin was created after the 2008 housing crisis,to take power from the hands of the government and give it back to population. Because goverment used tax payers money to bail out institutions who were greedy during that run and we suffered . Second was money printing problem , U.S treasury prints money and give it to the government to run the country and in return a interest rate is fixed but it piles up and the loan is at a alarming 37 trillion $ plus where the debt is being rescheduled 'n' number of times, just to pass over coming goverment that they will manage it. All politicians in their term target interest rate is low so that economy flourishses and they can be over the news .

John Gillen - WartimeEthereum.eth 的头像
John Gillen - WartimeEthereum.eth1 年前

Hey My dumb tweets are a generational store of value asset

RickyJarret18 的头像
RickyJarret181 年前

$akt fixes this

Kristov Atlas 的头像
Kristov Atlas1 年前

Awesome discussion

3AS 的头像
3AS1 年前

Data might be worthless, but control def isn’t

thenim 的头像
thenim1 年前

LOL. I worked for a trading company back in 2016, that used "high quality synthetic trading data" to train some models to trade futures on CME. Worked amazingly, then turned it on actual exchange, and it proceeded to loose everything. There's a lesson there somewhere...

Matt Luo ClarityText.com: AI apps for agentic web 的头像
Matt Luo ClarityText.com: AI apps for agentic web1 年前

This point starts on timestamp 16:10. Point: Most people don’t have unique human knowledge to contribute to train frontier models.

Monk Zero 的头像
Monk Zero1 年前

Yes, the core value proposition of crypto and its foundation was never technical but political, even philosophical.

Herrsosa 的头像
Herrsosa1 年前

1) but isn't there still data out there in the "real world" that needs to be onboarded to the digital world; not sure which one it would be but aren't we overlooking some "precious" data? 2) does that imply that you see a future for decentralised AI protocols (i.e. TAO)?

Brooke Lacey 的头像
Brooke Lacey1 年前

I feel like our data has been worthless for a while

Garrett 的头像
Garrett1 年前

@JosephJacks_ @jaltucher @samaroo00 @dotkrueger

Chuck Petras 的头像
Chuck Petras1 年前

@BrianRoemmele

SimplyTao 的头像
SimplyTao1 年前

$TAO

Ryan Gomes 的头像
Ryan Gomes1 年前

People only care about Bitcoin because number goes up. No one will use decentralized AI since centralized will always perform better.

Trace Cohen 的头像
Trace Cohen1 年前

Yes so many startups have tried to sell people that your dad has value and it’s not true. Billions of data points + Ai + thousands of sales people is worth something.

syam 的头像
syam1 年前

So the AI era will begin after the government starts censoring and What is the right way for AI to make earning for users?

Nguyen Hoang Le Vi 的头像
Nguyen Hoang Le Vi1 年前

decentralized AI hits different when the gatekeepers arrive

Dani O. 的头像
Dani O.1 年前

🔥🔥🔥

Alex 的头像
Alex1 年前

Decentralized AI will only matter if it’s usable. AGNT Hub is already letting agents act across feeds + chains—feels like the infra we’ll actually need when the clampdown hits.

Swayam 的头像
Swayam1 年前

@hosseeb check this out:

相关视频