M. Can Eris (@mcaneris) profile picture

himself, a lawyer by original intention and a computer scientist by final decision

Joined January 2010
Photo by M. Can Eris (@mcaneris)Photo by M. Can Eris (@mcaneris)
From Yoda to Aşık Veysel. Two very wise beings, a Jedi Master and a blind folk poet, might agree on this: perhaps, the thing you are looking for is the thing that is looking. Luminous Beings Are We open.substack.com/pub/mcaner…
1
64
Karl Marx imagined a future where society’s accumulated knowledge would live inside machines. Tonight I found myself discussing that idea with one. Somewhere between 1858 and now, knowledge became machinery. -- Talking with Marx through a Machine open.substack.com/pub/mcaner…
43
The AI agents are working in the other room. I’m listening to Dave Matthews. Only one of us can feel the earth tilt. I wrote a longer piece about AI, Train Dreams, and what remains when the work leaves us. Losing Up and Down open.substack.com/pub/mcaner…
36
OpenClaw was an iPhone moment — without a stage. No keynote. No announcement. Just a quiet realization that apps are no longer the right abstraction.
1
65
I wrote a longer piece on OpenClaw and why skill-first, agent-native software marks the end of the app era. mcaneris.substack.com/p/the-…
2
47
a piece on @openclaw and why skill-first, agent-native software marks the end of the app era. mcaneris.substack.com/p/the-…
35
the future of software is skill-first
45
M. Can Eris retweeted
Mahmut Tanal konuşuyor yayın kesiliyor adam iyi niyetli diye dalga geçip gülüyorlar sondaki yüz ifadesi kalbimi parçaladı… Haddinizi bilin çıkın özür dileyin!!! @szctelevizyonu
352
1,153
151
13,585
630,718
Ekrem millettir.
3,134
7,467
492
74,977
2,140,662
How will AI change programming? Wrote down some of the thoughts I've been toying with these last few weeks.
8
5
2
54
9,952
M. Can Eris retweeted
This is really a 'WOW' paper. 🤯 Claims that MatMul operations can be completely eliminated from LLMs while maintaining strong performance at billion-parameter scales and by utilizing an optimized kernel during inference, their model’s memory consumption can be reduced by more than 10× compared to unoptimized models. 🤯 'Scalable MatMul-free Language Modeling' Concludes that it is possible to create the first scalable MatMul-free LLM that achieves performance on par with state-of-the-art Transformers at billion-parameter scales. 📌 The proposed MatMul-free LLM replaces MatMul operations in dense layers with ternary accumulations using weights constrained to {-1, 0, +1}. This reduces computational cost and memory utilization while preserving network expressiveness. 📌 To remove MatMul from self-attention, the Gated Recurrent Unit (GRU) is optimized to rely solely on element-wise products, creating the MatMul-free Linear GRU (MLGRU) token mixer. The MLGRU simplifies the GRU by removing hidden-state related weights, enabling parallel computation, and replacing remaining weights with ternary matrices. 📌 For MatMul-free channel mixing, the Gated Linear Unit (GLU) is adapted to use BitLinear layers with ternary weights, eliminating expensive MatMuls while maintaining effectiveness in mixing information across channels. 📌 The paper introduces a hardware-efficient fused BitLinear layer that optimizes RMSNorm and BitLinear operations. By fusing these operations and utilizing shared memory, training speed improves by 25.6% and memory consumption reduces by 61% over an unoptimized baseline. 📌 Experimental results show that the MatMul-free LLM achieves competitive performance compared to Transformer++ baselines on downstream tasks, with the performance gap narrowing as model size increases. The scaling law projections suggest MatMul-free LLM can outperform Transformer++ in efficiency and potentially in loss when scaled up. 📌 A custom FPGA accelerator is built to exploit the lightweight operations of the MatMul-free LLM. The accelerator processes billion-parameter scale models at 13W beyond human-readable throughput, demonstrating the potential for brain-like efficiency in future lightweight LLMs.
104
818
270
4,706
2,452,804
M. Can Eris retweeted
New chain unlocked: @modenetwork 🟨⛓️🟡 Cross the Mode Bridge to discover the newest link in the Superchain. layer3.xyz/collections/explo…
286
23,811
40
4,811
205,666
M. Can Eris retweeted
v2 is here –– it's our largest update yet. Here's the alpha 👇
1,332
75,566
279
16,563
1,273,140
M. Can Eris retweeted
🚀✨The journey begins, with a bright light shining at the end 🪂! We're ecstatic to unveil our @layer3xyz campaign, plunging into the depths of the @zksync ecosystem 🌐 Today marks the beginning of an exciting series of events, showcasing KOI and our distinguished partners throughout the next month. Dive into these activities, boost your XP, and secure your portion of token rewards from KOI and our partners. Join us on this thrilling adventure and increase your chances at getting a double surprise at the end 😉 @HoldstationW @mavprotocol @zerolendxyz @RFX_exchange @getclave @SatoriFinance
128
16,091
22
2,443
182,422