ml for systems at nvidia opinions are my own
- Tweets 208
- Following 385
- Followers 39
- Likes 1,696
what is @chatgpt suddenly making mermaid charts for all explanations (i support!) but not actually rendering them in the UI (either on the web or the ios app)?
cc @thsottiaux
1
17
Mohit Garg retweeted
DO NOT use /fork or /side in Codex, it will have a different session_id and thus different prompt_cache_key.
the docs literally say you should only use it if you want to *prevent* cache hit.
plz fix it @thsottiaux
10
13
213
52,130
Since we are all enjoying GPT-6 Astra, I decided to write up a short article on distributed training based on my notes from reading the HF Ultra-Scale Playbook. There is a reason why Jensen stressed on 100k+ GPUs and NVLink72 in the congratulatory tweet.
1
99
Just ordered! Can’t wait for mine to skate around my house
We have a huge news to share today!
Today we are unveiling the first truly accessible RL robot - welcome Microduck
A 25 cm tiny open-source biped with 15 actuators and packed with sensors (camera, speaker, LiDAR, NFC, bluetooth, wifi, etc) that you train yourself with reinforcement learning.
It's also playable out of the box with more than half a dozen fun and playful pre-trained policies to have it walk, sit, crouch, roller-skate, pick up objects with its articulated beak, and recover on its own.
And all for less than $400.
See all the details, play with the simulator and order it at: pollen-robotics.com/microduc…
(video with sound on 🔊)
2
62
When the codex app was released, I was hesitant to switch from claude code. I think I just liked the idea of working out of the terminal and opus 4.8 was quite good at understanding what I wanted to be done.
I would try codex once in a while and slowly started to notice the benefits of the polish of codex and how it fixed some of the issues I was facing when working on remote machines via the terminal. However, I was never able to get gpt 5.5 to just do what I meant without a lot of hand holding.
Now that 5.6 Sol is here, I can (happily) say that I am no longer a claude code user. The current Codex + 5.6 Sol combo is goated
@thsottiaux _/|_
153
Mohit Garg retweeted
The value of fast iteration in AI is overrated. The best results are obtained by knowing the right things to do and doing each thing with neurotic precision and attention to detail.
26
35
439
43,879
Tried @ChatGPTapp agent mode today and all I can say is that the @Cloudflare “prove you are a human” check is the real turing test
95
Just read the @paulg essay on cities. In 2008, he said Cambridge was the intellectual capital of the world. I wonder if he still thinks that’s the case? Especially with the rise of AI and SF seemingly emerging as the idea hub for that.
83
why are the amazing folks at OpenAI giving interviews for blogs that are behind paywalls? the article just got interesting and there is no way to pay for an article on a pay-per-view basis :/
53
4. deepseek
it was the worse. no data at all, just gave me the sources where i can look. to it’s credit, it did give me the code i guess
@deepseek_ai
1
59
how are we making sure that llm benchmarks are valid i.e there is no benchmark leakage? for example, (complete?) swebench seems to be available for download
48
Mohit Garg retweeted
Impressive work by these young photographers! Using iPhone XR, more than 260 schoolchildren from around Chennai, India captured images of their community for the annual #ChennaiPhotoBiennale. #ShotoniPhone
264
1,139
7,514
#ChangeForAChange #RollBackBITSPilaniFeeHike
If the VC needs money, we will provide it to him with a personal touch. We will put all the change we have in front of his office. Happy Counting!
3
2