Joined November 2007
- Tweets 3,661
- Following 4,561
- Followers 1,549
- Likes 6,296
271 Photos and videos
Make Rustworkx Converge Again (on pagerank for large sparse graphs)
github.com/Qiskit/rustworkx/…
1
153
Robin Ranjit Singh Chauhan retweeted
20 Jul 2022
2028: Labs ranked by pps (publications per second).
2
22
Robin Ranjit Singh Chauhan retweeted
17 Nov 2025
E73: @danijarh (ex-@GoogleDeepMind Research Scientist) in-depth on Dreamer v4.
Audiogram edition 📡
2
12
962
Robin Ranjit Singh Chauhan retweeted
15 Oct 2025
yes, thankfully RL wasn't invented yet at the time :D
1
1
4
771
3 Aug 2025
Hacked a Canadian university computer network when i was 18, just so I could email my girlfriend in the UK (I asked them nicely first 😇).
27 Jul 2025
Share some lore about yourself that literally no one will care about.
1
4
789
Robin Ranjit Singh Chauhan retweeted
22 Jul 2025
E67: Stefano Albrecht on Multi-Agent RL @ RLDM 2025
@s_albrecht shares the story behind his multi-agent RL textbook, and how @DeepFlowAI turns these ideas into action with LLM-powered agents for business automation.
Recorded at @RLDMDublin2025
1
2
15
4,863
8 Jul 2025
Temperature control for anthropic opus on the console now works! (t=0 ftw).
But now, Opus refuses to answer my pretty tame questions about the biotech industry (sonnet does not refuse).
@AnthropicAI
27 May 2025
Whyever would claude opus temperature be stuck at 1.0 in web GUI? (For the older claude models, it can be adjusted).
@jackclarkSF I kindly wish for determinism, I am a t=0.0 enjoying person 🙏
1
497
Robin Ranjit Singh Chauhan retweeted
12 Jun 2025
TalkRL is at @RLDMDublin2025 , feel free to say hi!
1
1
12
870
3 Jun 2025
Entropy as epistemic humility
29 May 2025
Next, we identified that entropy collapse, the key bottleneck in scaling, stems from the covariance between logp and the advantage of actions.
Since base models already possess strong priors, initial covariance is high, especially on easy samples, causing rapid entropy collapse.
2
304
27 May 2025
Whyever would claude opus temperature be stuck at 1.0 in web GUI? (For the older claude models, it can be adjusted).
@jackclarkSF I kindly wish for determinism, I am a t=0.0 enjoying person 🙏
2
740
Robin Ranjit Singh Chauhan retweeted
4 May 2025
Excited for my first RLDM! Would love to say hi -- and will be looking for Talk RL Podcast interviews
🌸🍒🌸
15 Nov 2024
Save the date!
RLDM 2025, The Multi-disciplinary Conference on Reinforcement Learning and Decision Making, is only around the corner. Visit our website to keep an eye on our submission deadlines👀
rldm.org/
1
15
1,225
30 Apr 2025
"Binoculars for a triclops" -- Sora today no better than DALL-E a year ago in terms of having a "world model" it would need for this seemingly simple request.
3 Apr 2024
In which DALL-E attempts to draw trinoculars.
That is, binoculars for a triclops.
3
361
1 Apr 2025
In 2018 I gave this talk on GPT-2 and predecessors at Simon Fraser, mentioned ULMFiT as the original.
speakerdeck.com/robinchauhan…
29 Mar 2025
I'm glad @levelsio checked this, but sad our contrib has been erased by later big tech co's. Alec Radford said ULMFiT inspired GPT. ULMFiT's first demo predated BERT.
Today's 3-stage LLM approach of general corpus pretraining and 2 stages of fine-tuning was pioneered by ULMFiT.
1
1
240
1 Apr 2025
Even back then it seemed to me the ULMFiT story was starting to get papered over.
133
21 Jan 2025
Want a paid speaking gig at our AGM?
30 min talks on a deeptech topic incl:
- Mat Sci
- LLMs, RL, Agents, AI h/w
- Robotics
- Nanotech
- Energy
- Space
Seeking engaging speakers, ideally from Cali or nearby.
San Jose Convention Center Mon May 5 morning.
Please DM!
1
5
576
21 Jan 2025
You don't have to be based in Cali! Just aiming to minimize travel impacts.
1
198
Robin Ranjit Singh Chauhan retweeted
23 Dec 2024
E61: Neurips 2024 RL meetup Hot takes: What sucks about RL?
What do RL researchers complain about after hours at the bar? In this "Hot takes" episode, we find out!
Recorded at The Pearl in downtown Vancouver, during the RL meetup after a day of Neurips 2024.
3
2
35
7,308
