Forwarded from AbstractDL
This media is not supported in your browser
VIEW IN TELEGRAM
SayCan: как заставить языковую модель принести тебе колу (by Google)
Идея довольно простая: нужно взять робота, который уже умеет выполнять примитивные команды (найти, поднять, положить и тд), а в качестве мозгов для планирования взять языковую модель (PaLM) со специальным step-by-step промптом:
«How would you {TASK}? I would 1. ..., 2, …»
Дальше допустимые команды для робота ранжируются по перплексии и классификатором успеха (обученным с помощью RL). Таким образом, получилось заставить робота решать довольно сложные и многоступенчатые задачи.
Статья, блог, GitHub
Идея довольно простая: нужно взять робота, который уже умеет выполнять примитивные команды (найти, поднять, положить и тд), а в качестве мозгов для планирования взять языковую модель (PaLM) со специальным step-by-step промптом:
«How would you {TASK}? I would 1. ..., 2, …»
Дальше допустимые команды для робота ранжируются по перплексии и классификатором успеха (обученным с помощью RL). Таким образом, получилось заставить робота решать довольно сложные и многоступенчатые задачи.
Статья, блог, GitHub
Forwarded from 📚kruasan's library
2021-brown.pdf
2.5 MB
Can You Ever Be Too Smart for Your Own Good? Comparing Linear and Nonlinear Effects of Cognitive Ability on Life Outcomes
Forwarded from 📚kruasan's library
📚kruasan's library
2021-brown.pdf
> We found no support for any downside to higher ability and no evidence for a threshold beyond which greater scores cease to be beneficial. Thus, greater cognitive ability is generally advantageous—and virtually never detrimental.
Forwarded from Just links
Omnigrok: Grokking Beyond Algorithmic Data https://arxiv.org/abs/2210.01117
Forwarded from Just links
Nature
Discovering faster matrix multiplication algorithms with reinforcement learning
Nature - A reinforcement learning approach based on AlphaZero is used to discover efficient and provably correct algorithms for matrix multiplication, finding faster algorithms for a variety of...
🤔2🤣1
Forwarded from Evgenii Zheltonozhskii🇮🇱
Nature
Revisiting Lévy flight search patterns of wandering albatrosses, bumblebees and deer
Nature - Levy flights are random walks characterised by many short steps and rare long steps. They display fractal properties, have no typical scale, and occur in physical and chemical systems. On...
Forwarded from Empty Set of Ideas
полезная картинка, раньше я всегда просто пользовался пиклом, теперь буду брать npy или npz для спарсных данных
https://github.com/mverleg/array_storage_benchmark
https://github.com/mverleg/array_storage_benchmark
∅
https://www.youtube.com/watch?v=gvp8khkvoQQ ah yes, esoteric marxism again
https://www.youtube.com/watch?v=2aq820K-7r8
even more schizo stuff
(resonate with some stuff from complexity theory)
even more schizo stuff
(resonate with some stuff from complexity theory)
🙏2
Forwarded from Ordo Nexus
The Distributed Knowledge: Nexus (5/5)
Let's bridge all pieces together. Nexus created the search that is capable to run inside browsers using WASM and be distributed through IPFS. Also, the data for a search engine is also put in IPFS.
IPFS network cares about resilient storing of both parts, search engine and data, making them immortal. WASM brings the search engine back to you for the sake of resilience and privacy.
I'd like to underline it - privacy. All your queries executed on your own computer and nobody can ever learn, exploit your queries against you and make you a subject of advertising.
Now, you do not need to bring your queries to the search corporations but search engine comes closer to your queries.
Moreover, while using such kind of the search, you are becoming an owner of corresponding parts of the search index. Every user continues to spread part of index it has used recently. Search become possible even while you are detached from the Internet.
So, Nexus-on-IPFS mentioned recently is not just another LibGen mirror. Nexus cannot be blocked and is naturally replicated across IPFS nodes. Once released and replicated for multiple nodes, Nexus will stay in the network for eternity.
We have plans to attach to Nexus other severely censored datasets like Sci-Hub, Wikipedia, some national mass medias and several torrent trackers.
All parts of Nexus have open sources. Any of you can use developed tools to create your own database, put it in IPFS, easily pack it together with Summa search engine and hence make your data really interplanetary and unstoppable.
Let's bridge all pieces together. Nexus created the search that is capable to run inside browsers using WASM and be distributed through IPFS. Also, the data for a search engine is also put in IPFS.
IPFS network cares about resilient storing of both parts, search engine and data, making them immortal. WASM brings the search engine back to you for the sake of resilience and privacy.
I'd like to underline it - privacy. All your queries executed on your own computer and nobody can ever learn, exploit your queries against you and make you a subject of advertising.
Now, you do not need to bring your queries to the search corporations but search engine comes closer to your queries.
Moreover, while using such kind of the search, you are becoming an owner of corresponding parts of the search index. Every user continues to spread part of index it has used recently. Search become possible even while you are detached from the Internet.
So, Nexus-on-IPFS mentioned recently is not just another LibGen mirror. Nexus cannot be blocked and is naturally replicated across IPFS nodes. Once released and replicated for multiple nodes, Nexus will stay in the network for eternity.
We have plans to attach to Nexus other severely censored datasets like Sci-Hub, Wikipedia, some national mass medias and several torrent trackers.
All parts of Nexus have open sources. Any of you can use developed tools to create your own database, put it in IPFS, easily pack it together with Summa search engine and hence make your data really interplanetary and unstoppable.