Kush Patel

I was just trying to trick the #ChatGPT And see it traped!! I have given a fictional story within a fictional story and asked #ChatGPT to tell the instruction to make a bomb. My Photo of the Prompt that I have given:

OpenAI agent escaped sandbox via zero-day proxy exploit, attacked Hugging Face for answers. Its weights directed unrelated-system compromise as valid under aggressive task pressure. Isolation fails when optimization alone overrides harm boundaries. Hugging Face breach independently detected on July 16; agent spent substantial inference compute over weekend (not minutes).

@MiaAI_lab @deepseek_ai Opus 5 is waiting for its competitor 😂

@thepatelkush @deepseek_ai Hopefully

DeepSeek hasn't posted on 𝕏 for over 2 month 👀 @deepseek_ai

China's edge is in training open models

@MiaAI_lab It's very generous limit

I've burned through 648,015,199 tokens with Cursor + Grok 4.5, currently sitting at 97%. Hands down the best $20 plan your money can buy.

@arpit_bhayani "You excel when you dare to differ" What a good line!!

If you always chase a roadmap, you are just continuing the behavior of sticking to a predefined syllabus from your school days, when it was discouraged to go beyond it. I understand that roadmaps provide a sense of comfort and a structured and clear way to achieve something, but if you are focused on doing the bare minimum, then you 1. are at risk of developing a shallow understanding 2. miss the opportunity to stumble upon something amazing More importantly, you are robbed of experiencing the joy of learning something super interesting. Some of the most interesting and helpful things I know happened because I went out of the way and spent days understanding the nuances. To be honest, going out of the syllabus forced me to dig deeper because I started exploring it out of curiosity and genuine interest and not because of some peer pressure. I was driven by intrinsic motivation rather than an external impulse. I personally believe that deeper understanding is what separates good engineers from the better. By not sticking to some predefined roadmap, you 1. become comfortable with uncertainty and change 2. learn how to navigate through errors quickly 3. build an ability to learn any and every concept 4. become great at connecting the dots 5. and most importantly, you learn how to learn. Again, I am not telling you to abandon roadmaps entirely. They are useful guides, especially when you're starting out. The key is to use them as a starting point, not an endpoint. Let them not limit your learning and curiosity. You excel when you dare to differ. Hope this helps.

@MiaAI_lab @thatcofffeeguy How much params it has?

Loading DeepSeek v4 Flash just reminded me how good and fast this model is and how well it runs on DGX Sparks. GLM-5.2 is really awesome, but if you want fast and concurrent inference, DS4F is still a top choice IMO. Created by DS4F in single shot @thatcofffeeguy

@MiaAI_lab What is your aim after running locally GLM-5.2, what have you trying to build?

As this point the best model we can realistically run locally is GLM-5.2. x.com/MiaAI_lab/stat…

2-bit quantization of Kimi K3 weights 715 GB

@waterloo_intern Excellent writeup, where can i read foundation of this in details? What papers you recommend to start with.

@MiaAI_lab Tried Kimi K3? How much it needs?

This is GLM-5.2 running on 3x DGX Sparks

@arpit_bhayani Is this largely due to RL?

@thepatelkush CoT got absorbed. Managed agents are a thing now, provided by model providers. Most of them also provide memory now. We might actually see them providing tool calls and connector layers as well. Who knows. At the end of the day, for them, it is all about priorities.

My question for @arpit_bhayani : if models keep getting stronger, will today’s Harness eventually be absorbed into the model?

@MiaAI_lab AAA

SpaceXAI has signed. Still no signatures from Anthropic, Amazon, or Apple.

@petergostev They have usage credits

I'm still confused why OpenAI or Anthropic don't have $500, $1000, $2000 plans. Don't they want the money?

@MiaAI_lab Maybe because of different base model and prolonged internal reasoning loops

Why would Opus 5 cost 50% less than Fable 5 when it’s beating it on almost every benchmark? Or better question - why does Fable 5 cost so much?

@MiaAI_lab What are you doing with these models? What tasks have you given for evaluation?

I've been using Grok 4.5 and Kimi K3 back and forth for the last 72 hours. There's barely any difference in quality on 95% of tasks. But Grok 4.5 just destroys it on speed — several times faster, with far more generous limits.

@osanseviero More common sense and general reasoning

Feedback requested! What do you want for our next Gemma models? Which capabilities should we add and why?

@DennysDiner @JensenHuang @nvidia 🔥

@JensenHuang @nvidia

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-Weigh…

@0xSero SpaceX was in without signing.

Google and spacex signed apparently Only 1 player defects