Super excited to see @getaxal launch their stablecoin savings platform. One of the first ever MITxH companies making big moves🔥
Announcing Axal – a smarter way to save.
Axal is an automated stablecoin savings platform that earns you higher interest than your bank by aggregating top lending protocols.
Learn more to earn more. ⬇️
4
5
33
7,974
Congrats to Rumi!
Thrilled to announce Rumi’s $4.7M pre-seed raise, led by @a16zcrypto CSX and @EV3ventures.
We’re building alongside some of the sharpest minds in tech, with founders and leadership from @virtuals_io, @elizaOS, @solana, @monad_xyz, @Polygon, and @OmniFDN helping shape the vision.
This is our runway. Now we build.
1
467
Congratulations to Cohort 2 company @RumiLabs_io on their recent fundraise! Read more about how they're building everyone's context-aware personal AI👇
🚨 Rumi Private Beta is officially open.
Start earning just by watching what you already love.
We’re rolling out in private batches through exclusive invites.
Backed by @a16zcrypto CSX, @ev3ventures, and more. We’re just getting started.
291
MIT x Harvard Blockchain Accelerator retweeted
We are hosting Cohort 2 Demo Day IRL along with @MITBitcoinClub expo tomorrow! Come join us and meet the teams 🚀🚀🚀
lu.ma/pcx2cb6g
1
2
788
MIT x Harvard Blockchain Accelerator retweeted
An extremely clear thesis for decentralized AI. Fantastic job by @SPLehman
One key takeaway was just the insane technical work by the DeepSeek team on the hard tech model problems and how this is not only happening at the popular Web2 AI labs. These extremely nuanced decisions on building models will expand as teams create differentiated models through varied tradeoffs in the design phase.
A few areas from the incredible post on why Decentralized AI can win
- Foundational Models: Decentralized pre-training networks could eventually enable a fully trustless, end to end process for building models. Challenges are still heavy communication requirements and limited hardware availability but teams are innovating here.
- Gym: A decentralized “gym” for generating reasoning data leverages crowd sourced contributions where anyone can add and verify reasoning traces, with token incentives for quality submissions and rigorous verification methods; we’re now looking for highly varied, domain-specific reasoning traces rather than just raw data.
- Refinery: Decentralized fine tuning is more feasible because it shifts the heavy lifting to distributed inference. Tons of inference requests not only provide answers but also generate reasoning traces that can be verified and used to progressively make models smarter.
- Peer Collaboration: Peer to peer learning, as seen in approaches like RL Swarm, allows models to evaluate and improve upon each other’s outputs, enabling a global network of models to learn collectively instead of relying on a single centralized model.
- Resource Optimization: Quantization and efficient distributed communication enable participation from nodes with limited hardware, lowering entry barriers and making decentralized reinforcement learning more accessible. TLDR we can start using everyones latent hardware.
1/ "There are decades where nothing happens; and there are weeks where decades happen." At present, nowhere does this quote feel more apt than in the field of AI. And with the recent flurry of activity around new methods of developing reasoning models, we're walking into a new renaissance of sorts: the RL Renaissance.
This thread dives into the newly forming RL paradigm for model improvement, and explores how decentralization might be leveraged to further it 🧵
5
8
45
5,258
MIT x Harvard Blockchain Accelerator retweeted
Excited to "open-source" my conversation with some of my favorite people in DeAI! Recorded as part of the @0xmitharvard accelerator, we go deep into all things decentralized training, data, open v closed-source, and even some fun stuff on brain waves🧠
MITxH Speaker Series Ep 1: Decentralized AI Panel with @alexanderlong, @annakaz, & @caseykcaruso | hosted by @SPLehman
(0:00) Intros
(5:34) What's Working in DeAI, What's Not Working
(8:31) Capital Intensity & Open v Closed Model Development
(12:23) Will Frontier Labs Continue to Open-Source?
(17:14) Quality v Quantity, Are We Running out of Data?
(32:49) Are Decentralized Models Pushing the Frontier or Copying Others?
(34:42) Brain Data as a New Frontier for AI/ML
(42:05) Data v Models as a Moat
(45:49) IP Rights in Model Development
(48:13) Privacy Techniques in Training
(52:20) Invasive v Non-Invasive Approaches for Understanding the Brain
(54:06) Private Weights, Private Data in Decentralized Training
1
4
14
1,358
MITxH Speaker Series Ep 1: Decentralized AI Panel with @alexanderlong, @annakaz, & @caseykcaruso | hosted by @SPLehman
(0:00) Intros
(5:34) What's Working in DeAI, What's Not Working
(8:31) Capital Intensity & Open v Closed Model Development
(12:23) Will Frontier Labs Continue to Open-Source?
(17:14) Quality v Quantity, Are We Running out of Data?
(32:49) Are Decentralized Models Pushing the Frontier or Copying Others?
(34:42) Brain Data as a New Frontier for AI/ML
(42:05) Data v Models as a Moat
(45:49) IP Rights in Model Development
(48:13) Privacy Techniques in Training
(52:20) Invasive v Non-Invasive Approaches for Understanding the Brain
(54:06) Private Weights, Private Data in Decentralized Training
1
3
26
4,528
MIT x Harvard Blockchain Accelerator retweeted
1/ "There are decades where nothing happens; and there are weeks where decades happen." At present, nowhere does this quote feel more apt than in the field of AI. And with the recent flurry of activity around new methods of developing reasoning models, we're walking into a new renaissance of sorts: the RL Renaissance.
This thread dives into the newly forming RL paradigm for model improvement, and explores how decentralization might be leveraged to further it 🧵
17
103
843
141,757
This session is happening in just a few hours! Make sure to register in advance!
🔗lu.ma/88rh9tfm
The first event in this year's MITxH speaker series is later this week! Join us for an open and honest discussion on the state of Decentralized AI with @caseykcaruso from @topology_vc, @annakaz from @vana, & @alexanderlong from @PluralisHQ! Register below👇
2
132
MIT x Harvard Blockchain Accelerator retweeted
The first event in this year's MITxH speaker series is later this week! Join us for an open and honest discussion on the state of Decentralized AI with @caseykcaruso from @topology_vc, @annakaz from @vana, & @alexanderlong from @PluralisHQ! Register below👇
1
2
29
1,075
MIT x Harvard Blockchain Accelerator retweeted
Really excited to be moderating this panel with three of my favorite people in DeAI. Love talking with them not because they're blindly bullish, but because they bring the right amount of skepticism to the space.
Sign-ups initially limited to MITxHarvard folks but will open up more slots as space allows! Link to sign up below🔗
The first event in this year's MITxH speaker series is later this week! Join us for an open and honest discussion on the state of Decentralized AI with @caseykcaruso from @topology_vc, @annakaz from @vana, & @alexanderlong from @PluralisHQ! Register below👇
1
2
16
1,635
The first event in this year's MITxH speaker series is later this week! Join us for an open and honest discussion on the state of Decentralized AI with @caseykcaruso from @topology_vc, @annakaz from @vana, & @alexanderlong from @PluralisHQ! Register below👇
7
16
79
14,568
Register here: lu.ma/88rh9tfm
1
2
242
MIT x Harvard Blockchain Accelerator retweeted
🔖 MITxH Happy Hour co-sponsored by @0xmitharvard + @satlayer
📆 Feb 27, 2025
🕐 4 PM - 6 PM MST
RSVP👇 lu.ma/uqsqi2tj
1
3
7
631
Meet up with MITxH @ ETHDenver! We'll be hosting a happy hour sponsored by Cohort 1 alums @satlayer to bring together the MIT and Harvard blockchain communities. Come stop by if you're an alum, current student, or faculty! 🍻
Registration link in comments👇
2
3
784
Apply to join the event here: lu.ma/uqsqi2tj
1
128