drubinstein
@dsrubinstein
Followers
505
Following
118
Media
13
Statuses
122
Making models go brrr | Engineering @reflection_ai
Joined March 2024
Excited to finally share our progress in developing a reinforcement learning system to beat Pokémon Red. Our system successfully completes the game using a policy under 10M parameters, PPO, and a few novel techniques. Blog posted below
12
33
400
As someone who spends way to much time in the PyTorch Profiler; https://t.co/ipNpGQFKnD ^^ I really like this ^^
github.com
A debugging and profiling tool that can trace and visualize python code execution - gaogaotiantian/viztracer
6
24
276
Welcome to the team!
🎉 Next week, I am excited to join @reflection_ai as a Member of Technical Staff to help build the open intelligence ecosystem of the Western world. It's the most exciting opportunity to help software builders in our time, and will shape many years of AI Engineering in the
0
0
4
It's amazing how much more you can debug and measure once you figure out the parent process's pid.
1
1
5
We didn't qualify for the Gen1OU #pokeagent competition at Neurips. Found some bugs way too late. Everything will be open sourced as a part of #pufferlib so you can try training Pokemon Gen 1 battling at 1M SPS!
1
2
12
Attempting to beat the Neurips 2025 #PokeAgent challenge with a 2M parameter model in collaboration with @cooperunion and @jsuarez5341 . Will it work? I hope so! Regardless, all work will be open-sourced.
1
1
13
I joined Reflection a year ago to take a big swing at working on the cutting edge of artificial superintelligence. It’s been an incredible year and looking back, I’m extremely grateful for the opportunity. Day after day, our ambition grows and now we are venturing onto our next
Today we're sharing the next phase of Reflection. We're building frontier open intelligence accessible to all. We've assembled an extraordinary AI team, built a frontier LLM training stack, and raised $2 billion. Why Open Intelligence Matters Technological and scientific
7
1
64
ISMIR-a classifier that determines if a paper is suitable for music information retrieval conferences. Is it just a prompt in chatgpt? Yes. Why hasn't anyone published the idea to #ISMIR2025 ? @keunwoochoi ?
1
0
0
Lesson: Always remember to check your machine's time zone when scheduling cron jobs.
1
0
2
GSPO is pretty awesome. The paper has been living rent-free in my head the entire weekend.
Proud to introduce Group Sequence Policy Optimization (GSPO), our stable, efficient, and performant RL algorithm that powers the large-scale RL training of the latest Qwen3 models (Instruct, Coder, Thinking) 🚀 📄 https://t.co/n6jKHJOcnI
1
0
2
Unfortunately, my notifier is not schedule or delays aware. Apparently that's an important feature. The B32 today was detoured due to road work (and also nearly 10+minutes late when initially departing).
1
0
1
B32 automation update: I rigged up a VM running the detection script in cron and posting it to ntfy. ntfy is not "ideal," but it's functional!
1
0
1
I'm very excited for @reflection_ai 's product announcement! The product team here been hard at work building the next generation of AI Agents.
Engineers spend 70% of their time understanding code, not writing it. That’s why we built Asimov at @reflection_ai. The best-in-class code research agent, built for teams and organizations.
1
0
12
Engineers spend 70% of their time understanding code, not writing it. That’s why we built Asimov at @reflection_ai. The best-in-class code research agent, built for teams and organizations.
100
181
2K
🚀 Launch day! The NeurIPS 2025 PokéAgent Challenge is live. Two tracks: ① Showdown Battling – imperfect-info, turn-based strategy ② Pokemon Emerald Speedrunning – long horizon RPG planning 5 M labeled replays • starter kit • baselines. Bring your LLM, RL, or hybrid
7
38
165
I'm looking for something a little fancier than an email and as convenient as Slack webhooks without having to create a Slack workspace. Any advice?
1
0
1
However, vibe coding cannot seem to solve how to setup a notification system beyond email. The entire space feels antagonistic to hobbyists.
1
0
1
Good news is that Gemini amazingly coded up the query script for me including Pydantic BaseModels within minutes. Looks like AI Agents are amazing for simple data processing tasks. https://t.co/Iw4syKu1FN
github.com
Contribute to drubinstein/catch-32 development by creating an account on GitHub.
1
0
1
Anyway, this should've been an easy problem. - Write a script that queries the MTA Bus Time API to check if the bus is running. - Run the script on a schedule between common commuting hours. - Write a notification system so my coworkers can know if the bus will actually show up.
1
0
1