Wootzapp

3.2K posts

Wootzapp banner
Wootzapp

Wootzapp

@WootzApp

Home of the w8-RL rollout infrastructure. we built a browser... so you can build environments

New York, USA เข้าร่วม Eylül 2021
1.4K กำลังติดตาม105.7K ผู้ติดตาม
Wootzapp
Wootzapp@WootzApp·
@g0da_s So cool. We will see you soon 🙏
English
0
0
1
94
Goda
Goda@g0da_s·
three weeks ago i made a birthday wish to get to sf this year. turns out wishes move faster than you think. sf does something to you. the speed here is unlike anywhere i've been. people aren't talking about building things. they're mid-build, mid-raise, mid-launch. in europe, most won't touch hardware. my co-founder @petravicaleksas heard that more times than i can count. that's why being here matters. we got accepted into @fdotinc canopy — a 5-week program for the most obsessed builders in the world. they have a hardware lab. they back hard things. exactly the environment we needed. we've been quietly building @NeurexTech — a headband that reads your brainwaves and actively deepens your sleep while you wear it. if you've ever woken up after 8 hours feeling like you slept 4 — that's what we're solving. early access: neurex.tech
Goda tweet media
English
7
1
53
2.5K
Sherwood
Sherwood@shcallaway·
We're working on RL and post-training at @sazabi . Anyone w/ relevant experience want to get involved? We'd be happy to hire you as a consultant. DM me! Also, before everyone tries to sell me their RL product/platform, let me just say: we've already built our own internal eval framework. We almost certainly don't want to try your proprietary thing 😉
English
12
4
82
8.7K
Wootzapp
Wootzapp@WootzApp·
@ayushjaiswal This is the "why wont Google beat you" argument of the RL era
English
1
0
1
52
Hubert Thieblot
Hubert Thieblot@hthieblot·
pitch me your company in 1 word.
English
2.5K
15
1.1K
279.5K
Wootzapp
Wootzapp@WootzApp·
@phoebeyao Honestly, not a big issue. This isnt really an adversarial game - and model labs are more circumspect than RL env vendors to touch IP. Easily enforced contractually. The value of openreward is convenience. Less time lost in the surprising war-of-env-standards.
English
1
0
0
345
Phoebe Yao
Phoebe Yao@phoebeyao·
rl environment companies have a version of the same problem as traditional human data vendors. the know-how is sold instead of compounding. labs receive environments, tasks, and verifiers directly. they can inspect the verifier design, extend the architecture to new domains, and eventually automate the creation of more complex and diverse environments without the original builder. just less obvious. hosted environments could change this. labs get reward signals without ever receiving the docker image. IP stays protected. openreward is an exciting first step towards getting the best environment builders onto a platform where high-quality private environments can be accessed before they saturate, and builders keep what they build.
Ross Taylor@rosstaylor90

We’re releasing OpenReward, a minimalist product that does one thing really well: serve RL environments at scale. Agentic RL is really painful because it adds a new axis of compute - environment compute - alongside training compute that needs to be scaled seamlessly on demand. OpenReward is a narrowly focused product based on this problem. We serve complex agentic environments as minimal API endpoints, which work with any training framework and scale based on use. Our vision is a home of reward on the internet, which is interoperable with any form of training or evaluation - and ultimately provides an open ecosystem alternative to the closed RL vendor market. 🧵

English
7
7
154
17.4K
Ivan Burazin
Ivan Burazin@ivanburazin·
We're targeting $10M run rate by next month. Going from 0 to 1M in 60 days, hitting 1-3M in the next 45, and finally closing in on 10M in less than a year feels amazing. But it also puts enormous responsibility on us since customers depend on this infrastructure and their agents run production workloads. The validation constantly pushes us to be at the top of our game.
English
15
3
109
6.5K
Wootzapp
Wootzapp@WootzApp·
Genuine question to model labs - what RL env format do you use ? Its ending up being a battle - we support a few in our w8-rl rollout infra, but this is becoming a git vs mercurial thing.
Ben Burtenshaw@ben_burtenshaw

Meta's infrastructure. India's best builders. 48 hours. This is India's biggest agentic RL hackathon. OpenEnv is the new open standard to train AI agents, used by PyTorch, AI at Meta, and Hugging Face In April, India's best builders get to build on it and have their work reviewed by Meta and HF engineering teams. Meta PyTorch OpenEnv Hackathon × Scaler School of Technology. The best environments get evaluated for inclusion in the OpenEnv global ecosystem. - Real contribution. Not a portfolio piece. - $30,000 prize pool. - 48 hours. - Bangalore.

English
1
0
1
476
Ross Taylor
Ross Taylor@rosstaylor90·
We’re releasing OpenReward, a minimalist product that does one thing really well: serve RL environments at scale. Agentic RL is really painful because it adds a new axis of compute - environment compute - alongside training compute that needs to be scaled seamlessly on demand. OpenReward is a narrowly focused product based on this problem. We serve complex agentic environments as minimal API endpoints, which work with any training framework and scale based on use. Our vision is a home of reward on the internet, which is interoperable with any form of training or evaluation - and ultimately provides an open ecosystem alternative to the closed RL vendor market. 🧵
General Reasoning@GenReasoning

Introducing OpenReward. 🌍 330+ RL environments through one API ⚡ Autoscaled sandbox compute 🍒 4.5M+ unique RL tasks 🚂 Works like magic with Tinker, Miles, Slime Link and thread below.

English
18
25
221
47.6K
Ross Taylor
Ross Taylor@rosstaylor90·
We designed ORS with interoperability as a first class principle - works with any training framework or sandbox provider. OpenReward builds on ORS with managed infra, but if we suck then you can run the ORS servers elsewhere. This was important ideologically to us given our open ecosystem heritage. It also keeps us honest by forcing us to make a great product! I also fundamentally believe that good products should not try to do everything, but one thing well. OpenReward is a horizontally integrated product. We’re going to keep the surface area small and be as useful as we can to the surrounding ecosystem.
English
2
0
4
414
Wootzapp
Wootzapp@WootzApp·
@ivanburazin Can we talk to them ? We would absolutely love to build envs for them (running on daytona of course!)
English
0
0
0
187
Ivan Burazin
Ivan Burazin@ivanburazin·
@WootzApp Now have Android Sandboxes. ;) Let me know if you want access.
English
1
0
1
290
Wootzapp
Wootzapp@WootzApp·
@ivanburazin We run browser front rewards on mobile as well. Touch, etc. So we need android emulator - which needs kvm (or nested virtualization in google Cloud case)
English
1
0
0
294
Wootzapp
Wootzapp@WootzApp·
@Houda_nait @gabegreenberg Can we work with you ? We source non-public , real production source code from financial services firms and create RL environments around it
English
0
0
0
300
Houda Nait El Barj
Houda Nait El Barj@Houda_nait·
I’m hiring on my team at OpenAI for Research Engineer / Scientist and Software Engineer roles. I believe one of the most important questions for future AI systems is: how do we train them to help people over time? We work on that through RLHF, post-training, reward modeling, long-horizon evals, and the data infrastructure behind personalized multimodal AI. If you care about human flourishing and serious research, links in comments.
English
100
89
2K
178.4K
aashay sachdeva
aashay sachdeva@AashaySachdeva·
We are hiring in the models team Non-negotiables - Bald patch/grey hair Bad eating habits you are trying to fix Over reliance on caffeine/diet coke Cannot bitch about Delhi,all other cities are fine
Rahul@selfawareatom

Now that our 15 member llm team is infamous, time to expand for next time! If you have done one or more of the following, then please reach out. - pretrained a model of any size, from scratch - posttrained any base model, end to end (data curation, sft, rl) - are a pytorch wizard - are a cuda kernel master - you have any other relevant skills and work to back it up firstnamesarvamai

English
24
3
168
16.7K
Pratyush Kumar
Pratyush Kumar@pratykumar·
India’s startup ecosystem is special. Over the last decade, founders built extraordinary businesses by understanding India deeply. Now the AI opportunity is here. Of course adopting AI is a no-brainer. But the bigger question is - can some firms lead by building AI, can they convert their data advantage to model advantage... To discuss the why, what, and how of building custom models and to share some nuggets from all our model drops, we are hosting an invite-only group of founders in our BLR office. DMs open. Its Friday, the 13th.
Pratyush Kumar tweet media
English
71
159
1.5K
99.7K

ค้นพบ

@g0da_s @petravicaleksas @fdotinc @NeurexTech @mlejva @shcallaway @sazabi @ayushjaiswal