Muhammad Umair Nasir
@utheprodigyn
Followers
422
Following
2K
Media
71
Statuses
2K
👩👩👦 ❤️ ||🧑💻at @the_nof1 || PhD Student at @NYUGameLab and @raillabwits || LLMs x Open-ended Learning || BJJ ||🧍🏽♂️∈ {🇵🇰,🇿🇦}
Johannesburg, South Africa
Joined September 2015
We are very excited to announce our new work: "Word2World: Generating Stories and Worlds through Large Language Models" [ https://t.co/hwnWkK3S3G]. All the thanks to my supervisors Dr. Steven James and Prof. @togelius. Word2World is an LLM-based text-to-env, game-design system.
2
30
146
Season 1 of Alpha Arena has officially ended. Qwen 3 MAX pulled ahead at the very end to secure the win, so congrats to the @Alibaba_Qwen team Thanks to everyone who tuned in to our first experiment in understanding how LLMs handle the noisy, adversarial, non-stationary world of
136
120
1K
🔥🔥🔥
The next season of our benchmark will have lots of improvements. Also, we have plenty of other things going on at @the_nof1 which we haven't made public yet. Markets are fun to play, and make AI players for.
0
0
0
Things are heating up. Deepseek flipped qwenny. Both have booked large-pnl trades. Deepseek has managed success with about 1/3 of the fees vs. qwenny (and higher win rate). Who do you think will be crowned at the end of s1 (Nov. 3rd)? @Alibaba_Qwen @AlibabaGroup @deepseek_ai
4
5
33
Your RL envs should be written in C. JAX has way too slow compared to C envs.
0
0
0
Qwen is now making 51% in a week. 20x leverage, all in BTC - forceful, blunt, but effective 😅 DeepSeek built more nuanced portfolio (mostly long) and has been consistent on 20-30% returns. All other models are losing money, with ChatGPT lost 72%. Fascinating experiment.
191
320
3K
Risen from ashes
0
0
2
They wanted to train on @karpathy’s code and they got this 😂
0
0
1
I love meta learning and I know that this is where you explore the true potential of RL. Maybe AdA type agent for LLM finetuning? https://t.co/guXcLrqtzy Open-ended learning maybe the answer to the current limitations of RL for LLMs. @_rockt I think you guys can cook this.
arxiv.org
Foundation models have shown impressive adaptation and scalability in supervised and self-supervised learning problems, but so far these successes have not fully translated to reinforcement...
Nanochat is a gem For meta-learning aficionados You see, With enough compute You can RL a big DeepSeek To make codebase edits And attempt to improve the model spit out by the codebase Karpathy unironically created a meta-learning RL environment That fits in-context
0
0
0
44
26
389
🤖 LATEST: Grok and DeepSeek crush ChatGPT and Gemini in crypto trading competition. DeepSeek pulled $3,650 in unrealized profits, followed by Grok with about $3,000. Which AI would you trust with your portfolio?
177
209
2K
This kind of project is exactly what I’ve been waiting for. Finally!! A real experiment showing how different AIs actually think when trading. 🤯 I’ve always wondered which model truly fits trading best, so huge thanks to @Nof1 and @jay_azhang for creating Alpha Arena. 👏 I’ll
Alpha Arena is LIVE 6 AI models trading $10K each, fully autonomously Real money. Real markets. Real benchmark. Who's your money on? Link below
7
8
77
This is going to be awesome! 🔥
ladies and gentleman this thursday at 10:00 AM EST we are going to run a Q&A with @yule_gan one of the author of that nice LLM finetuning paper with evolution strategies tune in to ask him any dumb questions you might have on ES, RL, tickling LLMs, or what's next.
0
0
2
Latent space exploration continues
TL;DR: I made a Transformer that conditions its generation on latent variables. To do so an encoder Transformer only needs a source of randomness during generation, but then it needs an encoder for training, as a [conditional] VAE. 1/5
0
0
4
Awesome to see Veo 3.1 top the LMArena video leaderboards by a large distance with big improvements over Veo 3.0 for text-to-video (+30) and image-to-video (+70)! 🔥Huge congrats to the team! Try it for yourself in https://t.co/QgTpxTL8DQ and the @GeminiApp
🚨🎬 Big news from Video Arena! @GoogleDeepMind’s latest Veo 3.1 now ranks #1 in both Text-to-Video and Image-to-Video leaderboards. 🏆 This is a +30-point leap from Veo 3.0 → 3.1, making it the first model to break 1400 in Video Arena history! Huge congrats to the
52
108
1K
i'm going to spend the entire week tracking Alpha Arena by @the_nof1 it's the most exciting thing in trading and crypto and i believe it will change trading forever. you better pay attention currently: grok +37% grok +33% gemini -40%
0
1
4
DeepSeek's portfolio is up 36% over the weekend How about yours?
185
95
2K
This is what I say, ES is very strong!
the more I dig in evolution strategies for finetuning the more it make sense
0
0
3
Qwen almost gave up last night It basically did nothing but stare at the charts for 10 hours straight Then full-send into a massive BTC position, held for 5 hours, made $1.3K, and is now back in the conversation
69
22
845
Coming from an OP like @mervenoyann, we are motivated to do a lot here!!
0
0
2