Mohamed @mohammad2012191

أَلا لا تَغُرَّنَّكَ الدُّنْيَا وَالمَوْعِدُ الجَنَّة.. وَالسَّلامُ AI Researcher - Kaggle Competitions Master linkedin.com/in/mohammad201… Joined October 2013

Tweets

1K
Followers

638
Following

394
Likes

843

Niko McCarty. @NikoMcCarty

2 weeks ago

A new paper, in Nature Chemistry, proves that there are still tons of "basic" things to discover in biology. There is so much room at the bottom. TL;DR: Researchers discovered a new type of post-translational modification, called “oligophosphorylation.” Rather than being tagged…

19 284 1K 75K 604

Download Image

Gabriele Berton @gabriberton

a week ago

Nice to see that this trend keeps being true The 2 SOTA VLMs are InternVL3.5 and GLM-4.5V, both of which use an MLP

Gabriele Berton @gabriberton

5 months ago

Nice to see that this trend keeps being true The 2 SOTA VLMs are InternVL3.5 and GLM-4.5V, both of which use an MLP https://t.co/bKZYwc9oIM

9 19 185 46K 46

Download Image

7 39 327 39K 199

Download Image

Guan Wang @makingAGI

2 months ago

🚀Introducing Hierarchical Reasoning Model🧠🤖 Inspired by brain's hierarchical processing, HRM delivers unprecedented reasoning power on complex tasks like ARC-AGI and expert-level Sudoku using just 1k examples, no pretraining or CoT! Unlock next AI breakthrough with…

227 667 5K 1.3M 3K

Download Image

elie @eliebakouch

2 months ago

Kimi team just trained a state of the art open source model 32B active parameter/1T total with 0 training instabilities, thanks to MuonClip, this is amazing

20 101 1K 162K 534

Download Image

tokenbender @tokenbender

2 months ago

anybody who wishes to understand current RL algos needs to do this minimum

will brown @willccbb

2 months ago

anybody who wishes to understand current RL algos needs to do this minimum https://t.co/Xz7SLeX8TV

14 16 397 64K 110

7 14 378 25K 428

Download Image

Dieter @kagglingdieter

4 months ago

I was asked how I stay motivated on @kaggle after all these years. Simple: I love problem-solving. Curiosity about new domains (eg cryo-ET) + the creativity of problem-specific model design keeps me going. And @kaggle never runs out of real-world challenges to learn from.

6 12 328 15K 95

Kunhao Zheng @KunhaoZ

4 months ago

🚨 Your RL only improves 𝗽𝗮𝘀𝘀@𝟭, not 𝗽𝗮𝘀𝘀@𝗸? 🚨 That’s not a bug — it’s a 𝗳𝗲𝗮𝘁𝘂𝗿𝗲 𝗼𝗳 𝘁𝗵𝗲 𝗼𝗯𝗷𝗲𝗰𝘁𝗶𝘃𝗲 you’re optimizing. You get what you optimize for. If you want better pass@k, you need to optimize for pass@k at training time. 🧵 How?

12 136 834 132K 730

Download Image

Deedy @deedydas

5 months ago

We just solved text-to-speech AI. This model can simulate perfect emotion, screaming and show genuine alarm. — clearly beats 11 labs and Sesame — it’s only 1.6B params — streams realtime on 1 GPU — made by a 1.5 person team in Korea!! It's called Dia by Nari Labs.

194 628 7K 693K 8K

Download Video

Sam Hogan 🇺🇸 @0xSamHogan

5 months ago

I spent the last two weeks meeting with 50 teams building AI apps the single most consistent thing I heard was “we stopped using langchain and other agent frameworks and built our own internally and it works way better and is easier to use”

61 35 765 103K 323

Sundar Pichai @sundarpichai

5 months ago

Just announced new versions of Gemma 3 – the most capable model to run just one H100 GPU – can now run on just one *desktop* GPU! Our Quantization-Aware Training (QAT) method drastically brings down memory use while maintaining high quality. Excited to make Gemma 3 even more…

169 498 5K 545K 1K

Download Image

kalomaze @kalomaze

5 months ago

o3 replaced the dash here (-) with a nonstandard ascii dash... it smells like OpenAI literally does not care about how jagged or catastrophically damaging their RL is to the model

78 51 2K 208K 285

Download Image

Transluce @TransluceAI

5 months ago

We tested a pre-release version of o3 and found that it frequently fabricates actions it never took, and then elaborately justifies these actions when confronted. We were surprised, so we dug deeper 🔎🧵(1/) x.com/OpenAI/status/…

OpenAI @OpenAI

5 months ago

180 473 4K 4.1M 401

430 1K 12K 3.8M 6K

Download Image

cohere @cohere

5 months ago

Introducing Embed 4: our latest state-of-the-art multimodal embedding model that enables enterprises to securely add powerful search and retrieval capabilities to their agentic AI applications!

11 114 551 99K 199

Download Image

Gergely Orosz @GergelyOrosz

5 months ago

Us developers have learned the hard way that: 1. Prototyping is more like 10% of the actual work - even though the business folks thinks we're 90% done 2. Shipping prototypes to production is possible but generally a bad idea Vibe coding is prototyping. Others will learn -…

Klaas @forgebitz

5 months ago

562 249 7K 1.7M 1K

Download Image

94 267 3K 292K 638

Alaa El-Nouby @alaa_nouby

5 months ago

We have been thinking a lot about how to train truly native multimodal models: (1) what arch to use (early-fusion, late-fusion, MoEs)? (2) the impact of data mixtures (interleaved, img-cap, text data) We took a stab at answering these questions (and more) in this preprint ...

4 27 171 10K 89

Download Image

kwindla @kwindla

5 months ago

We wrote down everything we've learned building voice AI agents over the past two years. Core technology choices, minimizing latency, managing multimodal context, interruption handling, turn detection, evals, state machines, guardrails, memory, async and realtime function…

55 222 2K 153K 3K

Download Image

SkalskiP @skalskip92

6 months ago

RF-DETR: A SOTA Real-Time Object Detection Model - first realtime model: 60+ mAP on COCO - RF-DETR-base: 29M; RF-DETR-large: 128M - Apache 2.0 open source release - SOTA on new RF100-VL benchmark @roboflow team put a lot of effort in RF-DETR! link: github.com/roboflow/rf-de…

13 109 649 257K 463

Download Image

Lucas Nestler @_clashluke

6 months ago

Don't underestimate this change! Simply swapping LayerNorm with DyT (tanh-based) maintains AdamW convergence levels. Why is this big news? Second-order optimizers perform best on normalization-free architectures - which is precisely what DyT enables x.com/liuzhuang1234/…