sxysun ⚡️🤖
@sxysun1
Followers
4,301
Following
772
Media
70
Statuses
1,004
performance artist 🎭, ai & market design @agenticmarkets , auctions @
Joined December 2018
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
NuNew 3rd Showcase
• 121096 Tweets
Illa
• 77564 Tweets
Puigdemont
• 66265 Tweets
#Number_i_ビアボール60秒CM
• 64509 Tweets
PS 1
• 58934 Tweets
Kamuda Tasarruf
• 35814 Tweets
Good Monday
• 33387 Tweets
$GME
• 31804 Tweets
STOP MISTREATING NEWJEANS
• 21650 Tweets
LINEの新機能
• 18459 Tweets
最高の仲間
• 15907 Tweets
Square Enix
• 15289 Tweets
サマーウォーズ
• 13647 Tweets
#السعوديين_الاكثر_تاثيرا
• 12084 Tweets
ALWAYS BESIDE GULF
• 11485 Tweets
GET WELL SOON GYUVIN
• 10468 Tweets
Pinned Tweet
Stoked to announce an Agentic Markets workshop
@agenticmarkets
at
#ICML
2024!
@icmlconf
📇 Details:
✏️ Call for papers: due May 17th
📅 Conference: July 26/27th
featuring GOAT speakers like Tuomas Sandholm, Gillian Hadfield,
@drimgemp
@KonstDaskalakis
…
7
21
65
Released a new ⚡️🤖 Flashbots Research post on "Speeding up the EVM" with a focus on shared data conflict analysis and parallelization
12
46
256
MEV is homomorphic to the game of 5D Chess♟️
with multiverse (x-domain) time travel (multi-block)
This is why flashbots has always been a video game company 🎮
cross-domain MEV where the two domains are domain A at t=0 and domain A at t=1
5
2
46
36
19
156
⚡🤖 Flashbots Transparency Report ⚡🤖
We take a look at what happened during the last 4 months: megabundles, Flashbots Protect, Eth 2.0 MEV design & workgroup, and research preparing us for the multi-chain future (1/13)
3
35
124
longtail MEV Alpha: build a tool that automatically detects deflationary tokens in AMMs and then perform inventory manipulation attack. Or maybe this txn itself is generated by such a tool 🤯🤯
seriously, this kind of hacks are pattern-matchable and have happened tons of times
Yesterday, the Uni-V2 pool for USDT-TCR was drained .
Let's see how this was possible 🧵.
1/n
10
73
356
3
11
117
"What is the major value proposition of crypto?"
"And why, from just first principles, this value proposition induces MEV?"
those are the central questions I discussed in this whiteboard session with
@barnabemonnot
from Ethereum Foundation:
4
20
115
mev is getting competitive, more searcher alphas will overlap. probly only two MEV alphas will be left for newcomers:
(i) long tail, whose endgame is formal methods driven tools, like symbolic execution
(ii) probabilistic, which resembles quant trading and be as hard as it gets
Although upon reflection if you’re new you probably should not try to use symbolic execution. That is a big rabbit hole. Oops.
1
0
22
5
7
81
The problem was NEVER that apps cannot solve MEV.
Solving MEV is just adding intents capability (i.e., MEV-time apps), which we know how, it's long past escape velocity
The problem has always been that in solving MEV those apps centralize the supply chain, reverting to tradfi
if endgame is that apps try and capture/order most of the MEV rather than some network built on top of the base layer, then is there a point to the whole shared sequencer debate or SUAVE? what am I missing?
3
2
27
10
10
83
Wanna hear non-BS takes on the synergy between crypto and AI?
How can the two communities work together on concrete problems?
come to @ Zuzalu
09:00 - 19:00 (GMT+2) on Sunday, May 7, video link:
7
17
71
Random half-baked thought:
A relatively standard train of crypto is that MEV exist because the ordering protocol allows MEV to be extracted. But one could flip this argument on its head: perhaps one could argue that protocol emerge due to the existence of MEV to redistribute
Random half-baked thought:
A relatively standard train of economic thought is that rents exist because government policy allows rents to exist. But one could flip this argument on its head: perhaps one could argue that governments emerge due to the existence of rents to divide
6
8
95
10
12
73
One of the most thought provoking pieces in this space.
Expedient intent implementations are a continuation and magnification of existing centralization of MEV supply chain. And god I see so many bad intent implementations that just centralizes the market.
We of crypto twitter have been yapping about intents for a while now so
@gakonst
and I wanted to write more than a 240 chars about the idea
In particular, it felt important to lay out the *risks* to the eth community along with the sunshine and rainbows
13
42
233
4
4
75
"What are some concrete, approachable ways to study and implement cooperative AI?"
"How will crypto, acting as commitment devices, help its realization?"
Those are the questions we answer at
🍿 watch the full video series at:
1
17
54
📢 New paper alert! 📮
Many multi-agent AI algos assume the use of commitments when achieving security & cooperation, e.g., parameter sharing.
But how robust are those assumptions? In face of real-world incentives at breaking those commitments in AI apps, what dynamics emerge?
NeurIPS 2023 is upon us
At the Multi-agent Security workshop on Saturday,
@sxysun1
we'll be presenting our recent work on using Blockchains as infrastructure to guarantee AI security & cooperation
"Cooperative AI via Decentralized Commitment Devices"
3
10
61
1
12
43
If flashbots were tasked to design Calendly, they would ask everybody to submit a program blindly simulating everybody's availability and everyone elses program's expectations of those availability, where the counter-proposals are private so you have to guess, inside SGX
The sender of a calendly link is able to perform a stackelberg attack but they give up last look
the value of sending vs receiving a calendly link depends on schedule volatility and the computational cost of updating availability (aka time liquidity)
3
5
47
3
2
52
Best explanation so far
3
2
39
"mev is really about information transfer,"
Information is the New Money.
will we see a complete theory tomorrow in
@phildaian
's talk?
bridges are information transfer across trust boundaries; mev is really just about information transfer; thus, implies we can use mev to transfer information across bridges and perhaps get more secure constructions
2
1
27
5
8
46
The Sovereign Orderflow
a thread on the implication of sophistication, and memes (1/n)
4
6
42
THE most understandable Tarun talk I've seen to this day
love the "all it changes is who takes the chairman seat in the communist party"
we need the Hayekian "use of knowledge in society" style free market capitalism: by lowering the barrier of entry to deploy MEV-time apps
4
13
76
7
7
42
what is it with mev ppl and sauna???
Is everybody trying to copy the
@NathanWorsley_
secret sauce?
Randomly bumped into two mev people uncoordinated at Russian bath house 🧖♀️
9
3
40
today's twitter is literally this meme, god all the posts about how RFQs are "cool" and "a step in the right direction of experimenting intents"
2
5
39
Very cool paper on “symbolic mev”
🧵1/5 I am excited to announce that our latest research paper, "Your Exploit is Mine: Instantly Synthesizing Counterattack Smart Contract" (w/
@cszlin
, Marcelo Morales, Xiangyu Zhang, and
@KaiyuanZh
), has been accepted for presentation at Usenix'23!
@USENIXSecurity
@PurdueCS
12
34
187
4
1
30
🪭 This weekend, with
@apartresearch
, Flashbots is bringing a research-athon to you! 🚨🔔
How can lessons learnt from auctions and cryptography help multi-agent security, in the age of autonomy?
(irl locations at Berkeley/Amsterdam, also online)
1
8
28
In 5d chess you can checkmate by traveling back in time (reorg), and you can checkmate by going across timelines to checkmate (locking in the mev of another chain on your chain)
1
0
26
An ideal non-trivial solution to MEV consists of three parts:
1. programmable privacy such that allocation of spec-on-state can be done, and that no agent can profit from gaining knowledge of another's private information
2. efficient/expressive mechanism so negative
It is not possible to "solve" as in the sense of total elimination. But it is possible to "solve" in the sense of doing the best one could be doing.
2
0
15
2
3
26
Here are 9 ideas 💡 on some fun projects you can implement in a hackathon to explore how AIs can cooperate using smart contracts, "the robot money"
And, we will sponsor successful teams in those directions out of
@augmenthack
(18-19th) via research grants
A thread 👇
Announcement time: Flashbots ⚡️🤖 are sponsoring a massive research project coming out of the hackathon!
@sxysun1
has led the way with and it will now be a sponsored part of Augment!
1
1
12
2
7
27
Babe wake up new suapp just dropped
Has anyone at Flashbots reached out to Taylor Swift's team to get them to use Flashbot Auctions?
1
3
33
0
4
26
Super glad to see Arbitrum's ordering policy taking a first step towards MEV-awareness 🔥
sharing some of my thoughts on the design and latency race incentives compared with FBA-FCFS ⬇️
Curious about the proposed changes to our transaction ordering policy?
Read on as our co-founder
@EdFelten
walks us through a modified version of the first-come, first-served (FCFS) sequencing model, outlining the why, and the rationale behind it! 🚀
44
51
201
2
5
25
back in NYC, DM open for hang
down to meet and talk about all topics especially L2s/MEV 🤠
2
0
24
Rollups MEV is actually all about price discovery, incompleteness theorem, extortion, and monarchy 🤯🤯🤯
8
3
24
Request for SUAPP: a private shared state compute coprocessor for Base
1
6
22
we are hosting a small meetup focused on LLM routing, one of the few interesting market structures in AI, with
@onjas_buidl
and
@EdgeAGI
May 6th, 7:30PM, San Francisco
An A-tier list of guests is coming (details soon), setting up to pre-game the
@agenticmarkets
workshop at…
Excited to co-host an event on LLM router with the amazing
@sxysun1
🫡
Anyone interested in the intersection of open market mechanism and LLM should definitely register! We already have some amazing guests coming over 👀
Registration link:
5
4
28
0
9
25
Who wants to see a crypto version of the "Elon v Zuck jiujitsu fight?"
heated research roasts are meant to be a challenge on both the intellectual and physical level
the combination of most gigabrain scholars and most sagacious warriors
3
0
24
Inner monologues actually offer great intuition on why an internet of agents can work better than a single one:
Having multiple agents talking to each other is an un-raveling of the inner monologue that goes on in our head when humans are reasoning. Multi-agency just makes the…
If you ever struggle to understand why things are the way they are and why life is so terribly frustrating, remember that 50% of the population don't even have an inner monologue and they get to vote too.
1K
2K
21K
4
4
24
Thus, just like we created government to coordinate on allocating lands, we created ordering protocols to coordinate on allocating spec-on-state. This takes welfare back from Moloch.
1
2
22
which "layers" in the "modular blockchain" thesis are the most important ones (accrue value, etc,.), in future?
@jadler0
@toghrulmaharram
@sreeramkannan
6
0
20
OH at flashbots onsite: research is the sum of vibes
2
0
22
"Why newcomb's paradox is just a flashloan?"
"Why the OG prompt, Bible, prophesizes the negative externalities of modern coordination technologies?"
elaborated in slides for my talk: 🧠Intelligence beyond commitments 🗳️ at
1
0
6
4
5
16
Some ideas that I’m excited to see at
@augmenthack
!
WTF is cooperative ai and why does it has anything to do with decentralized technology?
1
0
17
@tzhen
@0xQuintus
@ks_kulk
@zmanian
@cwgoes
@sxysun1
@tarunchitra
@GwartyGwart
I heard there’s a prediction market for who will win 👀👀
1
0
5
4
0
19
There is no reason one cannot make an AI agent with cryptographical privacy and integrity guarantees that does all of the paperwork of visa for you
(for Americans reading this tweet, this means you get to live 8 more hours)
there already exist agent services like this…
I die a little bit inside every time I go through the Schengen visa process
2
0
21
4
5
17
going be at London next week, who's there and wanna chat MEV? Would be interested in AI topics as well
1
0
18
Orderflow and transaction execution is a hot topic, but poorly understood and most discussions lack an empirical touch:
Who holds what power in the game today?
Do proposed market structures (uni x, solvers, ofas) work as they said in practice?
How does this compare with tradfi?
Excited to share a new Flashbots blog post and dashboard illuminating Ethereum’s order flow market.
Links below!
38
136
553
2
1
16
many in the industry right now seem to mistake shortsightedness for down-to-earth product exploration "in the arena"
The bear has turned ppl into defeatists and many simply give up trying. It saddens me to see those around not only embrace the ponzis, but (even) celebrate it
1
1
17
🎭🕺🪭💃New performance art just dropped for y'all Swifties and econ nerds by
@stephensonhmatt
@lstephanian
and moi:
an auction to mitigate the negative externalities of ticketing bots and resale markets
read on for a tldr, and find more at !
In honor of our favorite celebrity’s birthday,
@stephensonmatt
,
@sxysun1
and I are releasing SUIFT, the Simple Unified Inception for Ticketing auction, a concept for a working suApp. You can view the landing page here:
6
11
41
2
5
17
Not your keys, not your intents. (h/t
@barnabemonnot
)
Private relay/offchain solutions for txn coordination (e.g., all RFQ-based OFAs) is gonna eat all your lunch, what will you do, anon?
3
0
14
Had a blast talking about cooperative AI & contract/commitment games on the panel with
@gordonliao
@stephensonhmatt
@MatheusVXF
it was amazing to see more serious discussions around AI and decentralized infra
Huge thanks to ~1K in-person attendees & 4K+ who joined livestream to our Future of Decentralization, AI & Computing Summit, hosted by
@BerkeleyRDI
!
Check out full program slides&videos including keynotes from Shafi Goldwasser Michael Jordan
@berkeley_ai
:
11
20
96
0
2
13
It is not possible to "solve" as in the sense of total elimination. But it is possible to "solve" in the sense of doing the best one could be doing.
Solving for MEV is a misnomer, you can’t solve an emergent property inherent to distributed consensus. It’s like attempting to solve for gravity, it’s laughable.
15
8
94
2
0
15
Slides for my latest performance art show:
🎭 MEV and Credible Commitment Devices 🗳️
Agenda for tomorrow! Makes up for missing out on SBC a few months ago
2
2
19
2
3
16
what can I say? having multiple credibility tech to choose from is always better than having only one credibility tech to choose from
we proud ourselves in providing differentiated solutions: get your credibility from a TEE instead of a committee
OH: “Whenever Ethereum has a problem they try to solve it with a committee + a synchrony assumption”
“Well Flashbots solves them with a trusted third party where they are the trusted third party so..”
1
3
15
0
0
16
Thank you to the Flashbots mates and stewards! 😎
@phildaian
@tzhen
@ObadiaAlex
@epheph
@fiiiu_
@jparyani
@tkstanczak
@bertcmiller
@taarushv
@thegostep
@metachris
@greg7mdp
@yoelopio
@DanielWorsley_
@hasufl
@pujalight
@lvanseters
@mmrosum
0
0
16
Much more bullish on how crypto games can uncover some interesting cases of mev after playing CATAN with the aid of commitment devices, can just integrate chatgpt and ask to encode natural language agreements into commitments easily and settle/resolve on the delegated domain
2
0
16
Concert tickets meets privacy and mechanism design
Getting concert tickets in the hands of true fans has long been a problem plaguing the event industry.
Tickets are often swiped by sniper bots, who then turn around to resell the tickets at a much higher price.
14
11
84
1
1
15
The bullcase for gaming is that it allows you to have a better signal on what kind of mechanism works and then apply those mechanism in real world, not to actually take the on-chain world for real and turn humanity into another episode of black mirror
3
3
15
You can’t swim in the mempool if you can’t swim in the cold plunge pool
3
0
15
"a protocol should only see what it's willing to defend with its life" 🔥
recommend everybody read, first concrete discussion on how protocol commitments should evolve with the meta-game in response to anti-mechanisms to keep its credibility, and "what" to put in-protocol
I released a new post, "Seeing like a protocol", where I explore how we engineer protocol credibility, and how it fares in a world where the protocol is more and more embedded in a dense net of other domains. Check it out and let me know what you think!
5
35
136
0
5
15
great overview slides & refreshing take that goes well with Michael Jordan's statistical contract theory talk:
"how should contract theory and mechanism design evolve in when facing AI, can we borrow some lessons from crypto"
Thanks for organizing a great convening of minds
@dawnsongtweets
!
Sharing slides from the session on Cooperative AI and Decentralization for those interested:
0
0
4
1
2
13
a dark take, but increasingly convinced that formal methods/verification work's purpose is just to label data for for future AGI to train on and solve deduction
sorry LEAN/Coq nerds
1
0
14
listening to the rollup sequencing debates are like watching couples fight tooth and nail over how to spend their imaginary Powerball winnings
3
0
13
To understand is to generalize. To generalize is similar to be able to do *some* compression.
This is why all "foundations of X" are studies of generalizations, e.g., structuralism, cat, set, induction, etc,. which are studies on understanding itself
2
0
13
Solana's meta-strategy is very different from Ethereum and many other crypto
I think
@jito_labs
mitigation to MEV prob works in Sol given the culture and path dependency there
BUT it would NOT work if done in Eth or in general any highly permissionless systems
Jito Labs has decided to suspend the mempool offered through the Jito Block Engine due to negative externalities impacting users on Solana.
The decision has been made after deliberate conversations with the Jito Labs team and key Solana ecosystem stakeholders.
174
233
1K
3
1
14
OH at Neuralips: "China does RLHF on the data level, AI alignment is not a problem if you already aligned the humans"
0
0
10
Why? indeed.
although disagree with the latter part, MEV is much more than non-credibility, focusing on credibility is really like... supposing that there already exists a designated good "mev-aware" mechanism, I would say there isn't from first place
Why MEV is getting so much mindshare:
Blockchains promise to change the world by serving as a credible third party when there is none (eg all of finance)
But this isn't the case in practice bc non-credible (incentive incompatible) middleman are at the center of it all
1
3
52
5
1
13
What if we make a protocol in which the difficulty to extract MEV increases superlinearly to the extractable MEV on that chain so that rational actors would never extract because no incentive, can this fair ordering aktually work?🤯🤯🤯
2
0
13
this is a MEV subtweet
OH: “After DA fees compete down to zero, the only residual value left to charge a proportional transaction fee on in a multi rollup world is a future on the right to own a lock or mutex”
7
1
61
1
0
11
Rent in
@AnthonyLeeZhang
thread arise from the fact hat some lands are more valuable, ie, the conflict of preferences. Similarly in crypto, some specification that the next VM state satisfy needs to satisfy are in more demand
1
0
12
the endgame of stackelberg attacks are just vampire attack designs coupled with ponzinomics
0
0
10
with
@DavideCrapis
@stephensonhmatt
@soispoke
@barnabemonnot
and Jonathan P-P, we attempt an answer: , borrowing lessons from MEV (where those pitfalls are faced daily, at scale)
I'll be at
#NIPS23
to present the work with
@stephensonhmatt
, come find us!
1
4
12
zkml targets very specific sections of the market (high security app/crypto-native audience).
It's an unexciting section, but since there is a lack of proven crypto+ai use cases atm, the most profitable play is to ride the narrative of decentralized AI. It's fair game imo.
non-consensus take, but bewildering to me the crypto x ai space spends so much time on specific types of verifiability (zkml). constant regurgitation of the same narrative by "important" people in the space. computational integrity *is* important but in a vanishingly small set of…
6
8
84
4
1
12
Some apps on Ethereum do do this today: by moving more components to off-chain centralized relays and/or vertically integrate with builders
Other apps don't: they can't afford off-chain relays or integration cuz don't have a convenient way of exerting mev-time control
why do you think apps on Ethereum don't do this today?
It's certainly possible.
17
2
19
2
2
10
Believing in decentralization as a goal not as a means is like Mathematical Platonism, we’d like that to be true and often fall into the mental model, but when asked to justify, we dismiss it. Human brain is addicted to explanation.
2
2
11
A followup discussion of the 3EV fable.
Talking about the missed bits in the discussions about “Moloch” within crypto.
0
3
10
If we decentralize all the way down, mev will just become incentive design for program synthesis
4
0
11
preference elicitation&expressivity (direct auction conditional bids etc) = vertical alignment in AI
preference aggregation&credibility (mafiaEV sound semantics of txn commitments etc) = horizontal alignment in AI
related:
looking like we should be treating MEV as an alignment problem, in the AI sense
1
1
18
2
1
11
Just arrived at Singapore 🇸🇬, gonna be here for next 2 weeks, any ppl want to gang up and chat?
9
0
9
@tzhen
@0xQuintus
@ks_kulk
@zmanian
@cwgoes
@tarunchitra
@GwartyGwart
We’ve have been busy prepping the jiujitsu mats and last minute cage rentals, good news is that there are enough folding chairs ay the club already
1
0
10
It is interesting how parallelization designs like EIP-2930 and Solana Sealevel (which are inherently coarse-grained) can lead to dev/smart contract programming paradigm shift (which provide free? oracles for relatively fine-grained state witnesses) 1/4
However, I want to highlight another way in which this is true:
We may see that, once a change like EIP-2930 is introduced, developers will begin to architect contracts with parallelism in mind to enable even more efficient MEV extraction than what e.g. backtests may suggest!
1
0
4
1
2
10
To push this thesis, Flashbots will be offering Research Grants in the
@augmenthack
hackathon in Paris (EthCC).
We have many ideas on what concrete explorations can be done in the cryptoXcooperativeai direction and would love to see them realized!
"What are some concrete, approachable ways to study and implement cooperative AI?"
"How will crypto, acting as commitment devices, help its realization?"
Those are the questions we answer at
🍿 watch the full video series at:
1
17
54
1
3
10
A great post to re-read in light of recent discussions around enshrinement
I released a new post, "Seeing like a protocol", where I explore how we engineer protocol credibility, and how it fares in a world where the protocol is more and more embedded in a dense net of other domains. Check it out and let me know what you think!
5
35
136
0
0
9
What I really like about this industry is the trickster of it, this thread of interactions is a great illustration.
“Oh nooo, our walled garden has been permissionlessly innovated on”
5
4
73
0
0
4
@aeyakovenko
MEV profit grows superlinearly as volume increases, some external data can be found on
0
0
10
in winter, you shouldn't be exploiting the tried-and-true strategy, you should be exploring new stars, and it's common to fail more when your exploration rate is higher, that's fine.
Don't be a coward and only dare to play the safe choice
0
1
9
Flashbots whitehat community has continued to deliver amazing results. We executed 100 successful rescues in the last 6 months, shout out to the amazing
@epheph
@0xdeadf4ce
@amanusk_
@zeroXbrock
@sipox11
mov, and
@MevRefund
for pulling off these rescues!! 😘 (10/n)
1
0
9
POV: you are trapped in a quivering boat, lights out, time is chaos. Beneath, monsters of the Styx hunger for souls.
Gazing down: you're a matrix of 0s and 1s. Stark realization: You're not on ancient waters, but an Ethereum transaction in MEV-time mempool, stalked by MEV bots.
@ralexstokes
@sxysun1
The time between one frozen snapshot and the next...
The world where the beacon doesn't shine light...
Where uncertainty reigns & otherworldly creatures may suck your life
1
0
5
1
0
10