Vik

Basilisk Trainer

[i2cjak @i2cjak]

fuck i wish we had machines that could make anything out of plastic for less than $500 and also robots in my computer that can make any software i want. it'd be like living in the future

[1X @1x_tech]

NEO’s Hands
An API to the Physical World

[Kimi.ai @Kimi_Moonshot]

Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation.

Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers.

🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth.
🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale.
🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead.
🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains.

🔗Full report:
https://t.co/u3EHICG05h

[Hamonshū @MoriYuzan]

Patience is Bitter, But its Fruit is Sweet (2023)

[Deedy @deedydas]

Gymnastics is the Turing test of video generation models

[Wisdom Theory @wisdom_theory]

The idea that changed the world, by Daniel Kahneman.

[Paul Graham @paulg]

AI will magnify the already great difference in knowledge between the people who are eager to learn and those who aren't.

[ash @ashleyyjoelle]

“why do you have a messy pile of clothes in the chair?”

it’s L1 cache for fast random access to our frequently used clothes in O(1) time

[Physics In History @PhysInHistory]

James Randi on reason and belief 💭

[Beff (e/acc) @beffjezos]

The Milei speech in English, rendered using HeyGen

[AI Product Engineer @aipengineer]

LLM-VM - AI library of the day: LLM-VM by @anarchy_ai_inc lets developers easily build apps powered by LLMs without managing infra. Just provide data/APIs, and it handles prompt engineering, fine-tuning, load balancing between models, and more!🧵

👩🏽‍🏭Builders: @mmirman, @OdedeVik

[Mistral AI @MistralAI]

magnet:?xt=urn:btih:5546272da9065eddeb6fcd7ffddeef5b75be79a7&amp;dn=mixtral-8x7b-32kseqlen&amp;tr=udp%3A%2F%https://t.co/uV4WVdtpwZ%3A6969%2Fannounce&amp;tr=http%3A%2F%https://t.co/g0m9cEUz0T%3A80%2Fannounce

RELEASE a6bbd9affe0c2725c1b7410d66833e24

[Physics In History @PhysInHistory]

Same place 58 years apart.

[Han Guo @HanGuo97]

Introducing LQ-LoRA

Decomposing pretrained matrices into (fixed) quantized + (trainable) low-rank components enables more aggressive quantization.

We can quantize LLaMA-2 70B to 2.5 bits with minimal degradation in instruction-tuning performance.

https://t.co/l6gwA1okUx

🧵1/n

[kache @yacineMTB]

practice. you should be practicing

[Programmer Humor @PR0GRAMMERHUM0R]

superpowersBeLike https://t.co/ML1ho7HlQX