Chin-Yuan Yeh @jimmy_cyyeh

Ph.D. student. Still being made in Taiwan. Joined October 2021

Tweets

678
Followers

15
Following

343
Likes

1K

Jeremy Cohen @deepcohen

3 days ago

Even with full-batch gradients, DL optimizers defy classical optimization theory, as they operate at the *edge of stability.* With @alex_damian_, we introduce "central flows": a theoretical tool to analyze these dynamics that makes accurate quantitative predictions on real NNs.

18 185 1K 179K 1K

Download Video

Ethan Mollick @emollick

2 weeks ago

Some new theoretical economics papers looking at the implications of AGI. These two papers argue that a true AGI-level AI (equivalent to a human genius), if achieved, would eventually displace most human labor and reduce the economic value of remaining human work to near-zero.

146 324 2K 347K 2K

Download Image

Lin @Speculator_io

2 weeks ago

How to set up your portfolio for the future: • Chips: $NVDA $TSM $AVGO $ASML $STX $MU • Networking: $CRDO $ALAB $RMBS $CLS • Data Centers: $ORCL $NBIS $IREN $APLD $CRWV • Quantum: $IONQ $RGTI $QBTS • Space: $PL $RKLB $ASTS • Crypto: $COIN $HUT $MARA $CIFR $GLXY • FinTech:…

94 473 2K 230K 3K

Guan Wang @makingAGI

3 weeks ago

Hierarchical reasoning works well on large language models!🎉

33 185 1K 93K 1K

Download Image

Renç Korzay @renckorzay

a month ago

OpenAI fired this 23-year-old from their Superalignment team. But he turned his insider knowledge into a $1.5B fund that's outperforming Wall Street by 700% this year. He says maybe ~200 people in SF understand what's *actually* happening in AI right now. Here's his thesis: 🧵

357 2K 18K 4.5M 27K

Download Image

Rohan Paul @rohanpaul_ai

a month ago

Another paper claiming really BIG result. The First method to achieve 99.9% on AIME 2025 with open-source models! 🤯 DeepConf uses a model’s own token confidence to keep only its strongest reasoning, with GPT-OSS-120B while cutting tokens by up to 84.7% compared to standard…

20 155 826 70K 879

Download Image

Rimsha Bhardwaj @heyrimsha

a month ago

🚨 BREAKING: NVIDIA just exposed the dirty secret about LLMs. Their new paper proves SLMs outperform massive models in real-world applications. AI researchers are quietly pivoting overnight. 10 wild findings that change everything:

88 608 4K 471K 5K

Download Image

Keith Sakata, MD @KeithSakata

2 months ago

I’m a psychiatrist. In 2025, I’ve seen 12 people hospitalized after losing touch with reality because of AI. Online, I’m seeing the same pattern. Here’s what “AI psychosis” looks like, and why it’s spreading fast: 🧵

2K 14K 95K 7.6M 33K

Download Image

Abhinav SB @Abhhinav_SB

2 months ago

Implemented the research paper “memorizibg transformers“ into code with my own custom modifications in architecture for better performance.The whole model built from scratch with pytorch. Hugging Face link for model and training code is here: huggingface.co/abhinavv3/GPT_…

2 3 10 5K 12

Avi Chawla @_avichawla

2 months ago

KV caching in LLMs, clearly explained (with visuals):

31 189 2K 430K 4K

Avi Chawla @_avichawla

4 months ago

Transformer vs. Mixture of Experts in LLMs, clearly explained (with visuals):

22 188 2K 398K 3K

Prarthna @_Prarthna

4 months ago

@fchollet The "invention machine vs. automation machine" distinction is powerful because it addresses the fundamental disconnect. The model isn't thinking its way through the puzzle; it's following a learned heuristic that eventually snaps. However, the paths of automation and invention…

0 1 0 160 0

Leddy @LeddyLLC

4 months ago

1. Fix your posture Your joints aren’t failing, your alignment is. Years of slouching compress your spine, hips, and knees. Stand tall. Shoulders back. Pelvis neutral. Here's how to test your posture & an exercise to help fix it:

16 279 3K 630K 5K

Download Video

ILIAS ISM @illyism

4 months ago

this project stores millions of text chunks inside a video file (mp4) then runs sub-second semantic search on it - no vector DB, no servers - uses 10x less RAM & storage - no internet required it's called Memvid and it just broke my brain

282 707 9K 1.2M 11K

Download Image

Krish Goel @krshgl

5 months ago

5 undergrads, no lab, no supervision, and 1 year of "figuring it out" later, we're hyped to present - "Chronocept". It's the world's first benchmark to instill a sense of time in machines. This our lives' best work, and we hope to say this about everything we do hereinafter.

22 71 1K 89K 743

Download Image

elvis @omarsar0

5 months ago

LLMs Get Lost in Multi-turn Conversation The cat is out of the bag. Pay attention, devs. This is one of the most common issues when building with LLMs today. Glad there is now paper to share insights. Here are my notes:

98 632 4K 752K 5K

Download Image

Josh Tanke @josh_tanke

5 months ago

Two teenagers built a nutrition app using just a camera. No funding. No team. No ads. 12 months later? $8M in ARR. Just snap a meal photo—get calories, macros, and portions. No barcodes. No logging. This is how Cal AI quietly took over food tracking.

398 257 5K 9.5M 5K

Download Image

Jil Theo @theo_jil

6 months ago

If you're tired of your life, read this: Joe Dispenza wrote one of the most dangerous books I’ve ever read: Breaking the Habit of Being Yourself. It shows you exactly why you're stuck and how to break free. Here are 15 insights that’ll punch your old self in the face (in a…

108 1K 8K 1.1M 18K

Download Image

Ray Dalio @RayDalio

6 months ago

x.com/i/article/1909…

2K 9K 47K 15.0M 35K

Unwind AI @unwind_ai_

6 months ago

RAG is not Memory. AI agents need long-term memory to maintain context and learn continuously. 5 frameworks for AI agent memory. 100% opensource. 1. Graphiti builds temporally-aware knowledge graphs for AI agents that change over time with evolving relationships and context.