Mohamed Baioumy
@mo_baioumy
Followers
2,345
Following
425
Media
50
Statuses
266
PhD in AI & Robotics @oxfordrobots | building @exolabs_
Oxford, England
Joined September 2012
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Bridgerton
• 325285 Tweets
Cohen
• 245015 Tweets
Butker
• 188428 Tweets
Colin
• 90502 Tweets
billie
• 70323 Tweets
Catholic
• 69385 Tweets
Silvio Luiz
• 62371 Tweets
Penelope
• 50804 Tweets
2NE1
• 46368 Tweets
Senna
• 34160 Tweets
Kanté
• 31603 Tweets
Francesca
• 29832 Tweets
ISABELLE EMBAIXADORA
• 26605 Tweets
HAPPY BDAY HARSHAD CHOPDA
• 26431 Tweets
NCAA
• 25210 Tweets
Vargas
• 24796 Tweets
Barcola
• 18293 Tweets
Deschamps
• 17529 Tweets
Leeds
• 16634 Tweets
Olise
• 15685 Tweets
Madden
• 12147 Tweets
Betis
• 11088 Tweets
#الاتحاد_الخليج
• 11082 Tweets
Pinned Tweet
AI x Crypto is a hot narrative right now.
But what does “AI x Crypto” even mean? Is it all hype, or do these projects have substance?
To answer this, we've written a 90-page report.
Key ideas are in this 🧵👇
(1/12)
57
270
817
افتكرت موقف حصلي و انا في هندسة بنها، كنت رايح سكشن بدري و لما دخلت مكنش في حد غير بنتين
قولت "سلام عليكم"
واحدة منهم كشرت مره واحدة و قالتلي "علي فكرة انا مخطوبة" 😂
35
129
5K
4
3
309
Introducing Granular: the next generation NFT AMM.
It’s on-chain, gas-efficient and preserves the NFT value (doesn’t demote assets in the pool to the floor price).
Granular was one of the winners at
@Macro_Hacks
by
@1kxnetwork
@nftfi
A brief description 👇🧵
7
26
148
Finally, we present a landscape with 200+ projects, featuring:
FHE:
@zama_fhe
@Privasea_ai
ZKML:
@gizatechxyz
@ModulusLabs
Training:
@gensynai
@togethercompute
Data:
@getgrass_io
@FractionAI_xyz
GPU networks:
@gpunet
@golemproject
@ionet_official
Full stack:
@bittensor_
$tao
2
26
86
I first learned about Account Abstraction (AA) at a
@HomeDAO_live
lecture given by
@konradkopp
3 weeks later, we built a modular AA solution called and won at ETH Denver!
Check out our 2 min demo at
6
10
53
I have recently chatted to many AI x Crypto projects building in stealth.
Two that I am very excited about are
@sentient_agi
and
@Gatling_X
Keep an eye on them
7
4
47
Today, with Engineers Without Borders Oxford, we took a
@clearpathrobots
Jackal out to the university parks. We collected some data which will be used for a landmine detection project.
Thanks to
@oxfordrobots
@GOALS_oxford
for supplying us with the robot. With
@raabuchanan
.
1
10
44
Our 'Market Making in NFTs' paper is out!
Check the pre-print at
If you don't wanna go through the math (see pic),
@KristofLommers
has written an article with the main ideas of the paper at
6
10
38
Landscape (pt 2)
AI in smart contracts:
@ritualnet
@HyperOracle
@Spectral_Labs
@PhalaNetwork
Agents:
@Fetch_ai
@MorpheusAIs
@autonolas
@chainml_
AI assisting crypto:
@PondDiscoveries
@teablocks_com
Apps:
@PaalMind
@nfprompt
@Chain_GPT
@aiarena_
Security
@CertiK
@testmachine_ai
6
4
40
If you want to go straight to the report:
1
4
39
AI can be integrated into a chain. For example, an L1/L2 where you can use AI inside a smart contract.
This opens a realm of possibilities. Devs can now write a smart contract to act as an AI agent. This agent can manage a DAO or monitor the security of a protocol.
2
3
39
To recap, four flavours of AI x Crypto have substance:
1. AI assisting crypto
2. AI inside smart contracts
3. Cryptographic AI
4. Economic incentives
For each of the 4, we look at their potential, their bottlenecks, and the pay-off period
These are summarized below.
3
5
38
Our paper "Pixel Rarity: Rarity Learned Directly From Pixel Data" is now available at:
We compute rarity from the images only (actual pixels), rather than using the meta data.
This work was led by
@KristofLommers
, w our collaborators from
@flowdotso
.
2
7
34
I can't believe it's been a year on
@duolingo
already!!
To start, I created a group chat with friends where everyone sends daily screenshots. And I forced myself to do the daily minimum till I build this habit.
This happened after I discovered Atomic Habits by
@JamesClear
.
11
1
35
Can crypto solve the deep fake problem?
Short answer: No.
But, technologies in web3 can help. A short overview 👇
(1/7)
1
10
32
Crypto can help in the battle against deepfakes. This comes in 2 flavours:
- Cryptographic primitives
- Economic incentives
We provide a detailed explanation for how a crypto-based solution can increase the barrier to creating deepfakes in Sec 3.1
5
0
32
AI is impacting every industry, including crypto.
AI can be a game-changer for major issues, such as security.
AI tools for auditing smart contracts can help keep the 76B of value in Defi protocols safe.
There is another more promising avenue for integrating AI.
6
1
32
Zooming out, we present historical lessons from AI & how they relate to crypto in section 3.2.
Takeaways:
1. ZKML will scale (
@ezklxyz
@Ingo_zk
)
2. We need more compute & data (counting on
@akashnet_
@rendernetwork
,
@bagel_network
)
5
4
32
Been working on this cat collection recently. They are generated with the help of an AI model that captures 'desirability'. More details on that coming soon!
For now, lmk if you have ideas for designs I should try!
3
3
31
That's it for this thread.
Thanks to our supporters:
@0xPolygon
@Aleph__Zero
Thanks to everyone who helped including
@sandeepnailwal
@jackchong_jc
@_grieve
@MattGarciaEth
@MohamedFFouda
@DCbuild3r
@yarcoha
@0xmons
@BeWaterOfficial
+ many more. See full acknowledgement section!
1
2
29
Want to read more? Dig into the report:
Bookmark the first Tweet for reference 👇
AI x Crypto is a hot narrative right now.
But what does “AI x Crypto” even mean? Is it all hype, or do these projects have substance?
To answer this, we've written a 90-page report.
Key ideas are in this 🧵👇
(1/12)
57
270
817
1
1
30
The 4 flavours of AI x Crypto are interconnected. We use a framework to break down projects and understand the layers involved.
@flock_io
uses a decentralized network to train AI models (called federated learning) and rewards users with tokens.
More details in Chapter 2
4
2
30
Our
@iros2022
paper explores the limitations of active inference for robot control tasks.
We then introduce an 'unbiased active inference controller' to overcome these limitations.
Paper:
Code:
W
@CorradoPezz
@hawesie
@CionnRatz
3
8
29
انا بحاول ازود عدد الكتب العربي اللي بقرأها، فعملت جروب صغير فكرته اننا كل شهر بنخلص كتاب عربي و نناقشه في اخر الشهر
فأولاً لو حد عنده اقتراحات لكتب يا ريت تسيبها في تعليق
ثانياً لو عايز تشترك في الجروب ابعتلي رسالة
3
6
27
The 2 flavours of crypto helping AI are not limited to deepfakes.
- Cryptographic primitives can be used for the privacy and verifiability of AI models.
- Economic incentives can create marketplaces for AI inference, compute or data.
@opentensor
$tao
5
0
28
AI disruption is not all positive. The pace of disruption brings challenges.
A striking example is deepfakes. With Sora, you can generate hyper-realistic videos from a text prompt. How can we believe what we see online anymore?
Can crypto solve deepfakes?
6
0
27
بقالي ٥ سنين عايش في اوروبا من غير من انزل مصر
الفترة دي خلت العيد يوم عادي، بكرة مثلاً عندي مقابلة و بعدين جلستين في مؤتمر و شويه مواعيد، حتي صلاة العيد بقيت احطها في الاجندة و إلا هنسي
اللي بيفكرني بالعيد هي رسايل كل سنة و انت طيب، فكل سنة و انتو طيبين و عيد مبارك عليكم
6
0
27
Last Friday I gave a lecture at
@ucl
on "Asset Pricing for NFTs" hosted by
@richardmuirhead
&
@cryptonativephd
from
@fabric_vc
.
I discussed why pricing is hard, problems with rarity, and novel work using causal inference.
If interested in pricing or AMMs, DM me for a chat!
4
2
26
How can a robot detect sensory faults? And if sensory faults are detected, how should the control law be adjusted? We use the active inference framework to help us answer these questions.
Video presentation:
Paper:
1/2
1
6
24
My first trip to the US just got started. Looking forward to
#ETHDenver
and staying at
@sozuhaus
!
Any recommendations for things to do while here?
6
1
23
In our latest fault-tolerant control paper at
@IWAI
, we explore how active inference and precision learning can make the system robust against faults.
Paper w
@CorradoPezz
,
@hawesie
,
@CionnRatz
and
@chcorbato
Here is a short explanation of how this works
(1/6)
1
5
23
كنت عازم ناس عندي علي الفطار هنا في انجلترا، لقيتهم جايبنلي كارت لرمضان
طول عمري بيجيلي كروت للكريسماس او عيد ميلاد، اول مره يجيلي واحد لرمضان 😅
1
0
23
بعد كل السنين دي في الغربة أخيراً شربت عصير قصب 😂
اكاد من
2
0
21
Next week
@iwai_ws
, I will be presenting our papers on
1) stochastic fault-tolerant control and
2) planning under uncertainty with an indefinite horizon (the horizon length is unknown a priori).
If you're interested check out the workshop at
1
7
19
لو انت مهتم بمجال الذكاء الاصطناعي و الروبوتكس، احنا من فترة بدأنا جروب تعليمي بنذاكر و نناقش فيه موضوع كل شهر
الجلسة الجاية هتبقي يوم ٢٤ عن الsearch alg. and motion planning
بيقدمها
@h_hamed_elanwar
لو عايز تشترك سيب تعليق و انا هضيفك قبل يوم الخميس إن شاء الله
5
3
19
يوم الخميس القادم ٩ سبتمبر هكون موجود مع
@majd_abdulghani
&
@AshElshehawy
للرد علي اسئلتكم بخصوص التقديم للدراسات العليا
و المنح في جامعة اكسفورد او غيرها
لو مهتم بالحضور التفاصيل موجودة هنا
3
6
18
Join me in congratulating William Hartemink (
@ExeterCollegeOx
'21) on his thesis & getting 1st class honours
@oxengsci
. His work on bayesian control is also crucial for two upcoming papers, truly amazing for a master's student!
Secret sauce: punting supervision meetings😂
1
1
16
@ahmd3ssam
حاليا يبدو انه لسه محتاج فترة علشان يبدل شفل مهندس كويس، لكن هيحصل مع الوقت
ناس كتير قالت زمان صعب نعمل AI يغلب الانسان في الشطرنج، لحد ما جه deep blue
و في كل المجالات بعدها حصل نفس الموضوع
في blog ممتاز بيتكلم عن نقطة مشابهة اسمه The bitter lesson
2
3
17
We have won the Byrne Award for Entrepreneurship sponsored by Rutland Foundation &
@ChChwebmaster
🎉🎊
We (Moro AI) use
#MachineLearning
to help people in developing countries access quality education
Thanks to judges
@patrickchung
@pastormhm
Part of OX1 incubator &
@CGIU
0
3
17
في حاجه غريبة بتحصل
في البودكاست بتاعنا (مطب العشرين) لو في حلقة انا مش شايفها حلوة، دايما بتبقي من اكتر الحلقات اللي بتتسمع
ليه؟ 😂
منهم
الخطايا السبعة
عملت ايه في سن العشرين
البرود و الامبالاه
مطب 2022
4
0
17
بقول لحد من قرايبي اني الحمدلله نجحت في مناقشة الدكتوراه
قالتلي دكتوراه ايه و بتاع ايه، المهم العروسة 😂
5
0
15
At
@exolabs_
, we share an office with
@Gatling_X
. They have repeatedly shown us mind-blowing results
So excited for the rest of the world to see what they have built
♨️ Announcing: GPU-EVM by
@gatling_x
♨️
The most performant parallel-EVM in the world by 100x.
Massive unlock: foundational infra for training AI/RL models that interact with the EVM, accelerated L2s, MEV, backtesting, and more
🧵+ waitlist below
26
74
294
1
4
14
Thrilled to share that I have been selected for the Clinton Global Initiative University
@CGIU
class of 2021!!
Looking forward to the
@ClintonFdn
and
@HowardU
annual virtual meeting hosted by President
@BillClinton
and
@ChelseaClinton
on the 23rd of March!
1
2
13
Glad to receive the G.A. Paul Memorial Scholarship from
@UnivOxford
awarded for research excellence!
I want to thank my supervisors
@hawesie
&
@pduckw
, and my amazing colleagues who have helped me get here.
1
2
13
Tomorrow I will be co-chairing a session at the European Control Conference.
If you're attending ECC, check ThA07 Robotics III at 11:00. Also, our paper on fault-tolerant control will be presented. Co-authors:
@CionnRatz
@chcorbato
@CorradoPezz
@hawesie
0
3
12
Our latest work on fault-tolerant control has been accepted at !
We introduce approaches based on bayesian inference for robust state estimation and fault detection.
Paper:
With
@CionnRatz
,
@hawesie
1
3
12
We show the limitations of the active inference controller (AIC) used in the literature and introduce the 'unbiased' AIC.
@InferenceActive
Work with
@CorradoPezz
,
@CionnRatz
,
@chcorbato
,
@hawesie
.
Supported by
@AIforRetail
@oxfordrobots
.
And final thanks to
@PLanillos
!
2/2
0
6
11
مش عارف اجيبهالكم ازاي، بس انا اكتشفت ان في ناس بتشوي البطيخ
و المصيبة ان طعمه بيبقي حلو بعدها
1
0
11
@DanuFinance
, we have analyzed the top 20 *most traded* PFP collections on
@opensea
.
After minting,
39.3% of NFTs are never traded, and
92.2% are traded 3 times or less.
This hints at a liquidity problem in the NFT market.
1
2
10
A while ago, I moved guitar from home to the office. I'd take a break from work and play a bit (sometimes using
@yousician
). It was fantastic!!
But it slowly turned into me
playing guitar all-day and occasionally working 😂😂
Guess it's time to move it back home...
0
0
9
As humans, we are generally aware of our limitations. If your leg is injured you can detect this, and you know you shouldn't run! How can we build robots with this ability?
We answer these question by studying active inference. Check our
#ECC21
paper
1/2
1
6
9
بأكد تاني اني مليش دعوة بموضوع الحشيش اللي اتكلمنا عنه في اخر حلقة 😂
1
0
8
@ahmd3ssam
GPT-4 مستواه قل جامد مع الوقت
و بالمناسبة لو عايز تقارن كذا LLM ممكن تستخدم
@OpenRouterAI
1
0
7
I recently made a tiny contribution to
@PetrouEthan
's amazing research, which got accepted at
@goldschmidt2022
!!
So if you're in Hawaii attending the conference, check out *Session 12A* for our work on "Diagenetic cycling of barium isotopes in anoxic sediments"
1
1
7
Also, thanks to
@mmattamala
, Mees Vanderoeck,
@charliestreet1
and
@CorradoPezz
for helpful discussions and feedback!
#ICRA
@InferenceActive
@FRANKAEMIKA
2
0
7
Work with
@CorradoPezz
,
@chcorbato
,
@CionnRatz
,
@hawesie
and Martijn Wisse. This work was partially supported by
@oxfordrobots
,
@TUDelftDRI
and
@AIforRetail
.
@InferenceActive
2/2
0
0
6
Check out this summer school program from the
@OxfordCharacter
. I have recently participated in the extended version of this program (GLI) and would highly recommend it!
@OxUniStudents
0
1
6
Additionally, there are a few other issues that arise in practice.
To address the above, we built Granular. It’s based on models from
@danuFinance
and built together with
@puttyfi
. Here is the TLDR of how it works:
1
0
5
Learning Spanish has NOT been fun recently.
Using
@duolingo
has become a daily habit. But now I just do the bare minimum. And on many days, I repeat old lessons just to maintain my streak.
Any tips to make a habit fun again?
Dare I ask
@JamesClear
or
@droos_online
#langtwt
2
3
6
What is the value of going to crypto conferences as a founder?
4
0
5
The 2020
@ICAPSConference
Summer School will be held virtually this year and it's free to attend! It's a great opportunity to learn about AI planning and scheduling, robotics, and plan execution.
To audit:
For more info:
0
2
5
Next to the speed, I have also noticed much better awareness of the repo. Amazing work
@amanrsanger
& team
Copilot++ is now ~2x faster!
This speedup comes from inference optimizations + our best model yet. We really like it and hope you do too ☺️
22
35
508
0
0
5
This thread was partially based on Sec 3.1 of
Thanks to everyone who gave feedback on that section including
@DCbuild3r
,
@MohamedFFouda
,
@0xdjma
,
@ai
,
@jamesoleary32
,
@BMANLead
&
@BeWaterOfficial
team
2
0
5
Wanna try it?
We have a testnet version if you’re looking to understand the sub-pools:
Want to contribute, or stay up-to-date on the release? Join our discord!
Video description at:
2
0
5
Our group at
@oxfordrobots
is looking for a new post-doc! See below for more info and get in touch if have any questions
We're hiring! Can you help us develop state-of-the-art decision-making algorithms for autonomous systems? If so, get in touch! We're open to everyone across the broad spectrum of AI and robotics, RL, control, formal methods, probabilistic modelling etc.
0
5
9
0
2
5
لسه سامع اول حلقة من Podcast علمي جدا
@ahmedsamiread
@kerningcultures
اتمني لو description الحلقة يكون فيه مصادر للي عايز يقرأ اكتر عن الموضوع، اظن دي موجودة من اعداد الحلقة 😅
غير كده بداية جميلة و منتظركم في الحلقات اللي جايه 😁
0
0
5
All these technologies increase the cost of creating fakes.
For example, hardware attestation proves that my camera did indeed take this picture. I can still take a picture of something that is faked in the physical world. But now the cost is much higher.
(7/7)
1
0
4
@ac_crypto
@Galadriel_AI
@ETHGlobal
I think this is your 5th "wasn't planning to hack but"
Congrats on the win though!
0
0
4
Great listen, as usual! Of course, it wouldn't be a
@lexfridman
interview without asking about the meaning of life
Coincidently, I am presenting work in robotics inspired by the free-energy principle
@icra2020
BRAIN-PIL WS this Sunday. It's free to attend
Here's my conversation with Karl Friston, one of the greatest neuroscientists in history. From the vodka and wine in the background to the mathematics of existence, life, intelligence, and consciousness, this was a truly fascinating conversation.
26
88
512
1
2
3
Turns out this works really well. SOTA methods are still better, but keep in mind that precision learning does not monitor the systems or learn parameters offline.
Also, the performance can be easily improved with other precision learning methods. Stay tuned!
(6/6)
0
0
3
@0xKepler
@ucl
@richardmuirhead
@cryptonativephd
@fabric_vc
@jackchong_jc
Not at the moment, but I am aiming to publish a draft of our paper within 2 weeks.
1
0
3
@0xdjma
@DCbuild3r
@MohamedFFouda
@ai
@jamesoleary32
@BMANLead
@BeWaterOfficial
Thanks for helping clarify those thoughts : )
0
0
3
Our two most recent papers related to Active inference and the free-energy principle:
1)
2)
with
@mmattamala
,
@hawesie
,
@pduckw
and Bruno Lacerda
1
3
3
Let’s start with AMMs for fungible tokens. In a
@Uniswap
or
@SushiSwap
pool, you have two fungible tokens (e.g. ETH and DAI). Each token has the same value, I can buy or sell fractions of these tokens, and the price is decided by a bonding curve: x*y = k
1
0
3
A critical difference between Uniswap and NFT AMMs is that you only move in integers on the bonding curve. That’s because you can’t trade a fraction of an NFT. This causes slippage to be quite high, especially if the pool contains a small number of NFTs.
1
0
3
Also, we're designing the V2 already which improves the sub-pooling and the price efficiency. If you’re interested in contributing to that, feel free to drop me a DM.
1
0
3
Every sub-pool has its own bonding curve. The number of sub-pools and number of NFTs per sub-pool differs per collection, but can be optimally* determined. More liquid collections will have more sub-pools.
1
0
3
0
0
1
@DAppaDanDev
@worldcoin
@ritualnet
@gensynai
@UpshotHQ
@0xPolygon
@Chain_GPT
@zama_fhe
@nfprompt
@bittensor_
@Privasea_ai
Director's cut is at 127 pages so far
The core framework is 8 pages though (Chapter 2).
0
0
2
This has many issues. Most obviously, not every NFT in a collection is valued the same, thus having the same prices is problematic. Non-floor assets will be undervalued (at the floor prices) and thus the owner most likely won’t list them in the pool.
1
0
2
@jbrukh
@perplexity_ai
says
A group of lyrebirds is called a "musket".
The source it provides is this Wikipedia page. But the page itself says this had been removed since it was unsourced
Linke:
0
0
1
@0xGeeGee
@mewn21
Sub-pools are not independent, i.e. they reallocate tokens to minimize slippage & arbitrage. See 3 points here:
Also, deciding # of subpools and which NFT goes into what pool is very non-trivial. E.g. rarity results in poor performance.
Hope this helps
@0xCadmus
@Macro_Hacks
@1kxnetwork
@NFTfi
Yeah great question, I am working on a write-up to clarify this.
Briefly:
1) the subpools (in the V2 design) are not independent. Tokens are reallocated between them to minimize slippage and arbitrage. So it's not a few sudo pools next to each other.
2
0
1
0
0
2
1
0
2