JavaScript is not available.

The project is challenging, but I believe meaningful, and can be solved with a combination of research and engineering. Let's make science-fiction science-reality 🚀

Researcher (Robotics) - 株式会社アラヤ

Secondly, I'm looking for a roboticist who has hands-on experience working with hardware, ROS, etc.:

herp.careers

株式会社アラヤでは現在Researcher (Robotics)を募集しています。

Researcher (Neuroscience) - 株式会社アラヤ

Firstly, I'm looking for a neuroscientist who's able to work with and decode signals from (non-invasive) brain-computer interfaces:

herp.careers

株式会社アラヤでは現在Researcher (Neuroscience)を募集しています。

My main work at

, as part of the

, is to, essentially, control robots with my mind. If that sounds like an awesome goal, I'm hoping to hire multidisplinary researchers to build this 🧠🤖

Danijar Hafner

@danijarh

Dec 10

Hi Neuro Twitter, what's the most impressive experiment you know to date where a human learned to send signals via non-invasive brain measurements? Do we have an idea of what maximum intentional bit rate could be achieved given enough training?

Horace He

@cHHillee

Dec 10

Eager mode was what made PyTorch successful. So why did we feel the need to depart from eager mode in PyTorch 2.0? Answer: it's the damn hardware! Let's tell a story about how the assumptions PyTorch were based off of became untrue, and why PyTorch needed to evolve. (1/10)

224

1,358

Building A Virtual Machine inside ChatGPT

317070

@317070

Dec 3

Did you know, that you can build a virtual machine inside ChatGPT? And that you can use this machine to create files, program and even browse the internet?

engraved.blog

Unless you have been living under a rock, you have heard of this new ChatGPT assistant made by OpenAI. Did you know, that you can run a whole virtual machine inside of ChatGPT?

233

2,999

8,392

Tim Connors • 101

@itstimconnors

Dec 4

This fucking guy

818

8,644

60.3K

Any nice solutions for teleoperating robots (for manipulation) using VR hardware? Hardware suggestions? Any software that makes it easier, or does everyone really have to roll their own?

Emergent spacetime from a momentum gauge and electromagnetism

Pablo Morales

@_Pablo_Morales

Nov 28

Our new paper with Patrick Copinger is out🎉 We show that consideration of a topological Berry phase inspired local momentum phase coupled to electromagnetism gives rise to a momentum gauge dependent emergent spacetime.

arxiv.org

Motivated by the fermionic Berry's phase in momentum space, we study a local Abelian phase in momentum space, also known as the momentum gauge, coupled to electromagnetism, for complex scalars in...

Ryota Kanai / 金井良太

@kanair_jp

Nov 27

AI alignmentの問題に取り組みたい若手研究者を探している。

On the link between conscious function and general intelligence in...

Arthur Juliani

@awjuliani

Nov 18

Yesterday I had the pleasure to give a talk at 'Consciousness Club Tokyo' on mental time travel, conscious function, and what they might contribute to future AI development. Thanks again for inviting me,

@HiroTaiyoHamada

youtube.com

【Title】On the link between conscious function and general intelligence in humans and machines【Speaker】Arthur Juliani, Microsoft ResearchSlideshttps://drive.g...

Nov 18

Was about to re-use my own meme, but! These authors actually tested RL problems 🥺 (was worse than Adam but RL was considered OoD for VeLO so 🤷)

Quote Tweet

James Harrison

@jmes_harrison

Nov 18

Tired of tuning your neural network optimizer? Wish there was an optimizer that just worked? We’re excited to release VeLO

, the first hyperparameter-free learned optimizer that outperforms hand-designed optimizers on real-world problems: velo-code.github.io

Show this thread

Kudos to

for being the place to go to get pretrained models, wrapped up in simple APIs. Wanted to test out some new computer vision capabilities, and instead of spending ages figuring out people's repos, I'm up and running in a few minutes!

🚀

and

's work in action! #generative #AI #Contest #StayTuned

Quote Tweet

Space Engineers

@SpaceEngineersG

Nov 10

Spaceship Generator Competition

Competition Details

spaceengineersgame.com/announcing-the

𝐃𝐞𝐯𝐬 𝐥𝐨𝐬𝐭 𝐢𝐧 𝐬𝐩𝐚𝐜𝐞

Today, Nov. 10th, 5 - PM UTC

Twitch: twitch.tv/keencommunityn

YouTube: youtube.com/user/SpaceEngi #SpaceEngineers #Innovation #Space #Xbox #NeedToCreate

GIF

Nov 11

Neat work that shows how many RL problems (including dynamics prediction) can be turned into inference over conditioned sequence modelling tasks. Merging this with the algorithm I presented for Generalised UDRL would be even more general-purpose 🥳

Quote Tweet

Micah Carroll

@MicahCarroll

Apr 29

You know how we train general language models by randomly masking parts of the input? We think this makes even more sense for training a single _general sequential decision model_ that can perform behavior cloning, offline-RL, goal-conditioning & more!

arxiv.org/abs/2204.13326

Show this thread

Hiro Taiyo Hamada

@HiroTHamadaJP

Oct 28

【CC Tokyo】 11/18 10 am - 12 pm 次回のスピーカーは、Arthur Juliani

@awjuliani

Microsoft Research。 On the link between conscious function and general intelligence in humans and machines 意識と汎用知性についてトークしていただきます。 👇登録 us06web.zoom.us/meeting/regist

Nov 3

Very exciting. LLMs enable language interfaces for RL, and now code-LLMs enable symbolic policies. Prediction: high-level embodied policies will enable self-experimentation (ideally via meta-learning) for understanding the physical world, for further improvements in capabilities.

Quote Tweet

Jacky Liang

@jackyliang42

Nov 2

How can robots perform a wide variety of novel tasks from natural language? Execited to present Code as Policies - using language models to directly write robot policy code from language instructions. See paper, colabs, blog, and demos at code-as-policies.github.io long

Show this thread

0:20

32.5K views

gallorob

Oct 29

It hasn't even been a week since its release, and I have already come across a few amazing (non/refined by users) spaceships generated with the AI Spaceship Generator I worked on with

and

@err_more

--love to see it!

Quote Tweet

Arara(宇宙用）

@Arara00751113

Oct 29

AI製の船から錬成しました #SpaceEngineers

Show this thread

研究成果発表：オープンソースの『AI宇宙船ジェネレータ』を開発 | 株式会社アラヤ

ARAYA

@ArayaPress

Oct 27

アラヤの研究開発部のKai ArulkumaranとRoberto Gallottaが、手続き型コンテンツ生成(Procedural Content Generation)のための新しい進化的アルゴリズム(Evolutionary Algorithms)を開発する研究プロジェクトにて「AI宇宙船ジェネレータ」を開発いたしました.

araya.org

画像：生成された宇宙船のビジュアル化【概要】アラヤの研究開発部チームリーダーKai Arulkumaran

GoodAI

@GoodAIdev

Oct 27

🎤 Tonight! Livestream at 19:00 CEST with guest appearance by AI researchers Martin Poliak and

@ArayaPress

on #EA #algorithms + #PCG techniques for in-game content generation 🚀

Quote Tweet

Keen Software House

@KeenSWH

Oct 26

Don't miss our next live stream

𝐃𝐞𝐯𝐬 𝐥𝐨𝐬𝐭 𝐢𝐧 𝐬𝐩𝐚𝐜𝐞

Thursday, Oct. 27th, 5 - PM UTC

Twitch: twitch.tv/keencommunityn

YouTube: youtube.com/user/SpaceEngi #SpaceEngineers #Innovation #Space #Science #Sandbox #Xbox #NeedToCreate

GIF

GitHub - arayabrain/space-engineers-ai-spaceship-generator: Research code for the Space Engineers...

📢 Just released!

's grant research on evolutionary algorithms for procedural content generation has resulted in a standalone application for sandbox game Space Engineers 🚀 Read more here ➡️ goodai.com/ai-in-games-op

GIF

For more details, check the paper. And if you'd like to try it out, or even build upon this research, all of the code from the Space Engineers PCG project is open source 🚀

github.com

Research code for the Space Engineers AI Spaceship Generator project. - GitHub - arayabrain/space-engineers-ai-spaceship-generator: Research code for the Space Engineers AI Spaceship Generator proj...

PLEs combine methods from EAs, rec sys, pref learning, RL and ML in general. Seems like a lot, but the goal was a general, flexible framework. Make use of your favourite ML model (linear? NN? GP?), RL sampling strategy (ε-greedy? UCB?), etc.

Manually selecting items to evolve at every iteration may be slow and tiresome, so why not get the AI to do that for you too? So we speed up the mixed-initiative setting via preference learning, using a learned model to imitate human selections for some iterations.

Building upon our earlier work on a hybrid constrained optimisation EA for generating functional vehicles, and an improved version with surrogate fitness models, we now go to the interactive PCG setting, with an app for players to use to generate spaceships.

Final bit of work on the Space Engineers PCG project, with support from

@GoodAIdev

: "Preference-Learning Emitters for Mixed-Initiative Quality-Diversity Algorithms" by

@err_more

and myself arxiv.org/abs/2210.13839

Best of luck to the team! This sort of tooling is very important, and having dedicated maintenance is a great asset for the RL community.

Quote Tweet

Farama Foundation

@FaramaFound

Oct 25

Today we're launching the Farama Foundation, a new nonprofit dedicated to open source reinforcement learning, and we're beginning by maintaining and standardizing all the major open source reinforcement learning environments. Read more here: farama.org/Announcing-The

Oct 25

Having worked in RL with video game testbeds and moving to robotics and neuroscience domains, where previously domain knowledge was key, it feels like collecting more data is the way things are going. Trumps everything else. More data + simple objective + good model architecture.

Nathan Lambert

@natolambert

Oct 24

Open problems in RL research thread (infrastructure, methodological, philosophical, ...). Why does it feel like the field is expanding horizontally (more research, more general progress) without breakthroughs in academic (using the same 3+ year old algorithms)? Add some⬇️

274

Currently writing the background section of a paper that includes 5 subsections with completely different areas of AI 😵‍💫 Culmination of my

@GoodAIdev

grant, with

putting the finishing touches on the project. Exciting reveal soon 🎉

Ayana Niwa

@ayaniwa1213

Oct 17

理研AIP主催の女子中高生向けオンラインセミナーが始まります！数理・情報・Al系の研究者・大学院生が研究について話します。第2回は私も少し登壇させていただきます☺️ 近くに興味のありそうな女子中高生がいらっしゃれば共有お願いします！詳細はこちら ▶️ aip.riken.jp/event-list/sur

GitHub - mamba-org/mamba: The Fast Cross-Platform Package Manager

Oct 7

Just discovered mamba as a faster drop-in replacement for conda, and the improved dependency resolution and parallel package downloads are already saving me a lot of time with complex envs 🐍💘

github.com

The Fast Cross-Platform Package Manager. Contribute to mamba-org/mamba development by creating an account on GitHub.

Oct 7

This also seems to mesh nicely with Information Generation Theory, proposed by

@kanair

. In IGT, consciousness sits on top of a generative model. This could be episodic memory, or more broadly, what the authors call the "conscious memory system".

When writing our paper on conscious function and AI (openreview.net/pdf?id=LTyqvLE),

@awjuliani

talked of episodic memory as more of a generative model than a memory store, and this paper runs quite strongly on this perspective of episodic memory.

Interesting read - a relatively broad theory of consciousness that tries to explain many of the phenomena and possibly unintuitive findings on consciousness from a vast swathe of research. It evolved as a way to use episodic memory.

Quote Tweet

Neuroskeptic

@Neuro_Skeptic

Oct 2

"We argue that consciousness originally developed as part of the episodic memory system" pubmed.ncbi.nlm.nih.gov/36178498/

Today in

: #AlphaTensor, an AI system for discovering novel, efficient, and exact algorithms for matrix multiplication - a building block of modern computations. AlphaTensor finds faster algorithms for many matrix sizes: dpmd.ai/dm-alpha-tensor & dpmd.ai/nature-alpha-t 1/

0:07

735K views

120

2,931

8,008

Oriol Vinyals

@OriolVinyalsML

Oct 2

This neural network architecture that was showcased at the

@Tesla

AI day is a perfect example of Deep Learning at its finest. Mix and match all the greatest innovations to do something drastic and super ambitious. Congrats!

Tesla and 2 others

170

676

6,076