Links for 2023-03-29
This is amazing work! These guys invented a low-cost, open-source teleoperation system (top video) using off-the-shelf parts and 3D-printed components to teach an AI fine motor skills (bottom video).
This allows us to learn difficult tasks such as opening a translucent condiment cup and slotting a battery with 80-90% success, with only 10 minutes worth of demonstration data.
Page: Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware https://tonyzhaozh.github.io/aloha/
You can buy the full kit directly from Trossen: https://www.trossenrobotics.com/aloha.aspx
ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks: ChatGPT outperforms crowd-workers for several annotation tasks, including relevance, stance, topics, and frames detection, with 20x less cost. https://arxiv.org/abs/2303.15056
Bi-Manual Block Assembly via Sim-to-Real Reinforcement Learning: “Without manually-designed controller nor human demonstrations, we demonstrate that with careful Sim2Real considerations, our policies trained with RL in simulation enable two xArm6 robots to solve the U-shape assembly task with a success rate of above 90% in simulation, and 50% on real hardware without any additional real-world fine-tuning.” https://arxiv.org/abs/2303.14870
Biologists Say Deep Learning is Revolutionizing Pace of Innovation: "David Baker at the University of Washington sees deep learning driving a technological revolution in biology. Baker estimates the pace of innovation in this field has accelerated 10-fold in the past 18 months as a result of researchers using deep learning and laboratory approaches to confirm the behaviors of newly designed proteins. Jennifer Lum at growth equity firm Biospring Partners said startups across the life sciences sector are working with DeepMind Technologies' AlphaFold2 protein-structure prediction system and other deep learning tools. Some scientists say such milestones will expedite drug discovery and other life sciences advances. Said Baker, "In 10 years it is possible this will be the future of medicine."" [Wall Street Journal] https://archive.is/qZmcg
Replit is teaming up with Google Cloud to reach 1B software creators. How? Make non-devs into devs, engineers into hyper-productive "10X engineers," and 10X engineers into 1000X engineers. https://twitter.com/Replit/status/1640745029080866817
“We introduce CLAIRify: a novel approach that combines automatic iterative prompting with program verification to ensure programs written in domain-specific languages are syntactically valid and incorporate environment constraints.” https://ac-rad.github.io/clairify/
Tool-using AI startup Adept raises $350m https://www.adept.ai/blog/series-b
“We show for the first time that large-scale generative pretrained transformer (GPT) family models can be pruned to at least 50% sparsity in one-shot, without any retraining, at minimal loss of accuracy.” https://arxiv.org/abs/2301.00774
“The year is 2028, and this is Turing Test!, the game show that separates man from machine! Our star tonight is Dr. Andrea Mann, a generative linguist at University of California, Berkeley. She’ll face five hidden contestants, code-named Earth, Water, Air, Fire, and Spirit. One will be a human telling the truth about their humanity. One will be a human pretending to be an AI. One will be an AI telling the truth about their artificiality. One will be an AI pretending to be human. And one will be a total wild card. Dr. Mann, you have one hour, starting now.” https://astralcodexten.substack.com/p/turing-test
CodiumAI generates meaningful tests for busy devs. https://www.codium.ai/blog/codiumai-powered-by-testgpt-accounces-beta-and-raised-11m/
Artificial intelligence 'godfather' on AI possibly wiping out humanity: ‘It's not inconceivable’ https://youtu.be/qpoRO378qRY
Nat Friedman - Reading Ancient Scrolls, Open Source, & AI — “…he has created and funded the Vesuvius Challenge - a million dollar prize for reading an unopened Herculaneum scroll for the very first time. If we can uncover these scrolls, we may be able to recover lost gospels, forgotten epics, and missing works of Aristotle.” https://www.youtube.com/watch?v=qcvMjoJdck4
Existential Risk from Power-Seeking AI https://jc.gatspress.com/pdf/existential_risk_and_powerseeking_ai.pdf
“Editors Nature claim ‘ ‘We live in an era in which there are many attempts to politicize science.’ We find that Nature itself, along with other journals, have recently become politicized.” https://researchersforimpartiality.substack.com/p/political-activism-in-prestige-scientific
I am confident (80%) that all of this will be achieved within the next two decades. But by the end of 2025? Maybe 5% for the conjunction of all capabilities mentioned here.
This happens regularly on the British islands: