Derek Parfit in his book Reasons and Persons:
“Compare three outcomes:
1. Peace
2. A nuclear war that kills 99% of the world’s existing population.
3. A nuclear war that kills 100%.
(2) would be worse than (1), and (3) would be worse than (2).
Which is the greater of these two differences?
Most people believe that the greater difference is between (1) and (2). I believe that the difference between (2) and (3) is very much greater.”
Extinction would be much worse because it prevents the existence of all future generations.
“Compare three outcomes:
1. Peace
2. A nuclear war that kills 99% of the world’s existing population.
3. A nuclear war that kills 100%.
(2) would be worse than (1), and (3) would be worse than (2).
Which is the greater of these two differences?
Most people believe that the greater difference is between (1) and (2). I believe that the difference between (2) and (3) is very much greater.”
Extinction would be much worse because it prevents the existence of all future generations.
👍6👎3
On the day that formerly neutral Finland joins NATO, let us remember once again that this would not have been possible without the strategic genius of Vladimir Vladimirovich Putin.
His achievements are manifold:
1. No more talk about cutting defense spending among European nations. On the contrary, everyone is talking about modernization and increased funding.
2. The Western military industry has received a massive boost. Poland's $10 billion order for HIMARS rocket launchers and ammunition is just the beginning, as Russia is increasingly unable to meet its arms delivery commitments.
3. Ukraine is more united than ever and its self-identity has been massively strengthened. The alleged threat posed by Ukraine has not been neutralized. On the contrary, Ukraine is now capable of launching frequent attacks inside Russian territory.
4. Russia has been severely weakened. Russia has lost most of its considerable soft power over Western nations and access to Western markets. The image of Russia as a superpower has been completely shattered. And without Western technology and with China as its main export market, Russia's future will be that of a Chinese vassal state. A decidedly worse prospect for its national identity than before.
His achievements are manifold:
1. No more talk about cutting defense spending among European nations. On the contrary, everyone is talking about modernization and increased funding.
2. The Western military industry has received a massive boost. Poland's $10 billion order for HIMARS rocket launchers and ammunition is just the beginning, as Russia is increasingly unable to meet its arms delivery commitments.
3. Ukraine is more united than ever and its self-identity has been massively strengthened. The alleged threat posed by Ukraine has not been neutralized. On the contrary, Ukraine is now capable of launching frequent attacks inside Russian territory.
4. Russia has been severely weakened. Russia has lost most of its considerable soft power over Western nations and access to Western markets. The image of Russia as a superpower has been completely shattered. And without Western technology and with China as its main export market, Russia's future will be that of a Chinese vassal state. A decidedly worse prospect for its national identity than before.
👍14🥴2😱1🤮1
“It takes around 6 to 8 years to build a nuclear reactor. That’s the average construction time globally. Reactors can be built very quickly: some have been built in just 3 to 5 years.”
https://hannahritchie.substack.com/p/nuclear-construction-time
https://hannahritchie.substack.com/p/nuclear-construction-time
✍7😢2🤔1
Links for 2023-04-05
1. “There have been an incredible number of gene-editing advances (including in delivery) over the last three days. Here are 7 of them...” https://twitter.com/NikoMcCarty/status/1641811586431213568
2. Eliezer Yudkowsky: “So the actual scary part to me is that GPT4 understands what it means to say, "Compress this in a way where *you* can decompress it."…("Stochastic parrot" my fucking ass.)” https://twitter.com/ESYudkowsky/status/1643428537821720578 (for updates on this see e.g. this tweet and others by the same author: https://twitter.com/gfodor/status/1643415357615640577)
3. SudoLang: A Powerful Pseudocode Programming Language for LLMs https://medium.com/javascript-scene/sudolang-a-powerful-pseudocode-programming-language-for-llms-d64d42aa719b
4. Build an entire data workflow with just a one-sentence prompt https://twitter.com/EinblickAI/status/1641518278668582912
5. “What is the chance your vote decides the whole election? I've found an elegant proof that unless the election is a forgone conclusion, the chance can't be much lower than 1 in the number of voters.” http://www.tobyord.com/writing/decisive-vote
6. Why Starting A Rocket Engine Is So Hard! https://www.youtube.com/watch?v=bAUVCn_jw5I
7. difflogic - A Library for Differentiable Logic Gate Networks https://github.com/Felix-Petersen/difflogic
8. Closed-form parametric equations for plain-knit yarns and the twisted fibers running around them. Also includes C code to generate curves, and displacement/alpha maps for making tiled patterns. http://www.cs.cmu.edu/~kmcrane/Projects/Other/YarnCurve.pdf
9. “The normalization scheme that DeepMind researchers came up with for their "linear recurrent unit" (LRU) is a nice example of how it is possible to predictably engineer circuits in artificial neural networks, when you know what you're doing.” https://twitter.com/CFGeek/status/1640445451387412481
10. Floating Nuclear Power Buoyant on New Prospects https://www.powermag.com/floating-nuclear-power-buoyant-on-new-prospects/
11. “China spent $240 billion bailing out 22 developing countries between 2008 and 2021, with the amount soaring in recent years as more have struggled to repay loans spent building ‘Belt & Road’ infrastructure” [Reuters] https://archive.is/dpRuE
12. "Russia’s economy is in the tank and the reason many of us have thought differently is that Putin literally makes up the numbers." https://www.econlib.org/jeff-sonnenfelds-bombshell-about-the-russian-economy/
13. Satellite images: A web of trenches shows Russia fears losing Crimea https://web.archive.org/web/20230405012231/https://www.washingtonpost.com/world/interactive/2023/ukraine-russia-crimea-battle-trenches/
14. How the UAE paid a Swiss firm to destroy an oil-trading firm using negative PR, in part by editing Wikipedia articles about it [The New Yorker] https://archive.is/v3L8t
1. “There have been an incredible number of gene-editing advances (including in delivery) over the last three days. Here are 7 of them...” https://twitter.com/NikoMcCarty/status/1641811586431213568
2. Eliezer Yudkowsky: “So the actual scary part to me is that GPT4 understands what it means to say, "Compress this in a way where *you* can decompress it."…("Stochastic parrot" my fucking ass.)” https://twitter.com/ESYudkowsky/status/1643428537821720578 (for updates on this see e.g. this tweet and others by the same author: https://twitter.com/gfodor/status/1643415357615640577)
3. SudoLang: A Powerful Pseudocode Programming Language for LLMs https://medium.com/javascript-scene/sudolang-a-powerful-pseudocode-programming-language-for-llms-d64d42aa719b
4. Build an entire data workflow with just a one-sentence prompt https://twitter.com/EinblickAI/status/1641518278668582912
5. “What is the chance your vote decides the whole election? I've found an elegant proof that unless the election is a forgone conclusion, the chance can't be much lower than 1 in the number of voters.” http://www.tobyord.com/writing/decisive-vote
6. Why Starting A Rocket Engine Is So Hard! https://www.youtube.com/watch?v=bAUVCn_jw5I
7. difflogic - A Library for Differentiable Logic Gate Networks https://github.com/Felix-Petersen/difflogic
8. Closed-form parametric equations for plain-knit yarns and the twisted fibers running around them. Also includes C code to generate curves, and displacement/alpha maps for making tiled patterns. http://www.cs.cmu.edu/~kmcrane/Projects/Other/YarnCurve.pdf
9. “The normalization scheme that DeepMind researchers came up with for their "linear recurrent unit" (LRU) is a nice example of how it is possible to predictably engineer circuits in artificial neural networks, when you know what you're doing.” https://twitter.com/CFGeek/status/1640445451387412481
10. Floating Nuclear Power Buoyant on New Prospects https://www.powermag.com/floating-nuclear-power-buoyant-on-new-prospects/
11. “China spent $240 billion bailing out 22 developing countries between 2008 and 2021, with the amount soaring in recent years as more have struggled to repay loans spent building ‘Belt & Road’ infrastructure” [Reuters] https://archive.is/dpRuE
12. "Russia’s economy is in the tank and the reason many of us have thought differently is that Putin literally makes up the numbers." https://www.econlib.org/jeff-sonnenfelds-bombshell-about-the-russian-economy/
13. Satellite images: A web of trenches shows Russia fears losing Crimea https://web.archive.org/web/20230405012231/https://www.washingtonpost.com/world/interactive/2023/ukraine-russia-crimea-battle-trenches/
14. How the UAE paid a Swiss firm to destroy an oil-trading firm using negative PR, in part by editing Wikipedia articles about it [The New Yorker] https://archive.is/v3L8t
This media is not supported in your browser
VIEW IN TELEGRAM
"Introducing DribbleBot: A robot that can dribble a soccer ball on diverse natural terrains. Be it snow, be it grass, be it pavement, or be it sand, it keeps going! If the robot falls, it automatically recovers and keeps dribbling."
https://gmargo11.github.io/dribblebot/
https://gmargo11.github.io/dribblebot/
This media is not supported in your browser
VIEW IN TELEGRAM
Meta AI: "Today we're releasing the Segment Anything Model (SAM) — SAM is capable of one-click segmentation of any object from any photo or video + zero-shot transfer to other segmentation tasks"
https://ai.facebook.com/blog/segment-anything-foundation-model-image-segmentation/
https://ai.facebook.com/blog/segment-anything-foundation-model-image-segmentation/
🤯3
Links for 2023-04-06
1. TPU v4: An Optically Reconfigurable Supercomputer for Machine Learning with Hardware Support for Embeddings — Much cheaper, lower power, and faster than Infiniband, OCSes and underlying optical components are <5% of system cost and <3% of system power. TPU v4 outperforms TPU v3 by 2.1x, and The TPU v4 pod is 4x larger at 4096 chips and thus ~10x faster overall. https://arxiv.org/abs/2304.01433
2. Baize: An Open-Source Chat Model with Parameter-Efficient Tuning on Self-Chat Data -- Proposes a pipeline that can automatically generate a high-quality multi-turn chat corpus by leveraging ChatGPT to engage in a conversation with itself. https://arxiv.org/abs/2304.01196
3. “REFINER, a framework for finetuning LMs to explicitly generate intermediate reasoning steps while interacting with a critic model that provides automated feedback on the reasoning. Specifically, the critic provides structured feedback that the reasoning LM uses to iteratively improve its intermediate arguments.” https://arxiv.org/abs/2304.01904
4. “…we are releasing Cerebras-GPT, a family of 7 GPT models from 111M to 13B parameters trained using the Chinchilla formula. These are the highest accuracy models for a compute budget and are available today open-source!” https://twitter.com/CerebrasSystems/status/1640725880711569408
5. “We present Cluster-Branch-Train-Merge (c-BTM), a new way to scale sparse expert LLMs on any dataset — completely asynchronously.” https://twitter.com/ssgrn/status/1640322362100051968
6. unarXive 2022: All arXiv Publications Pre-Processed for NLP, Including Structured Full-Text and Citation Network https://arxiv.org/abs/2303.14957
7. Robotic hand can identify objects with just one grasp https://news.mit.edu/2023/robotic-hand-can-identify-objects-just-one-grasp-0403
8. To Make Self-Driving Cars Safer, Expose Them to Terrible Drivers https://singularityhub.com/2023/03/31/to-make-self-driving-cars-safer-expose-them-to-terrible-drivers/
9. “I’m scared of AGI. It's confusing how people can be so dismissive of the risks. I’m an investor in two AGI companies and friends with dozens of researchers working at DeepMind, OpenAI, Anthropic, and Google Brain. Almost all of them are worried.” https://twitter.com/arram/status/1642614341622181889
10. IQ estimates of public intellectuals and personas https://kirkegaard.substack.com/p/iq-estimates-of-public-intellectuals
11. Calling the Lab-Leak Theory 'Disinformation' Created Disinformation [The New York Times] https://archive.is/6YLW8
12. Do Politicians Ignore NIH Ties with Wuhan Lab to Protect Biodefense "Contract Racket"? https://disinformationchronicle.substack.com/p/do-politicians-ignore-nih-ties-with
1. TPU v4: An Optically Reconfigurable Supercomputer for Machine Learning with Hardware Support for Embeddings — Much cheaper, lower power, and faster than Infiniband, OCSes and underlying optical components are <5% of system cost and <3% of system power. TPU v4 outperforms TPU v3 by 2.1x, and The TPU v4 pod is 4x larger at 4096 chips and thus ~10x faster overall. https://arxiv.org/abs/2304.01433
2. Baize: An Open-Source Chat Model with Parameter-Efficient Tuning on Self-Chat Data -- Proposes a pipeline that can automatically generate a high-quality multi-turn chat corpus by leveraging ChatGPT to engage in a conversation with itself. https://arxiv.org/abs/2304.01196
3. “REFINER, a framework for finetuning LMs to explicitly generate intermediate reasoning steps while interacting with a critic model that provides automated feedback on the reasoning. Specifically, the critic provides structured feedback that the reasoning LM uses to iteratively improve its intermediate arguments.” https://arxiv.org/abs/2304.01904
4. “…we are releasing Cerebras-GPT, a family of 7 GPT models from 111M to 13B parameters trained using the Chinchilla formula. These are the highest accuracy models for a compute budget and are available today open-source!” https://twitter.com/CerebrasSystems/status/1640725880711569408
5. “We present Cluster-Branch-Train-Merge (c-BTM), a new way to scale sparse expert LLMs on any dataset — completely asynchronously.” https://twitter.com/ssgrn/status/1640322362100051968
6. unarXive 2022: All arXiv Publications Pre-Processed for NLP, Including Structured Full-Text and Citation Network https://arxiv.org/abs/2303.14957
7. Robotic hand can identify objects with just one grasp https://news.mit.edu/2023/robotic-hand-can-identify-objects-just-one-grasp-0403
8. To Make Self-Driving Cars Safer, Expose Them to Terrible Drivers https://singularityhub.com/2023/03/31/to-make-self-driving-cars-safer-expose-them-to-terrible-drivers/
9. “I’m scared of AGI. It's confusing how people can be so dismissive of the risks. I’m an investor in two AGI companies and friends with dozens of researchers working at DeepMind, OpenAI, Anthropic, and Google Brain. Almost all of them are worried.” https://twitter.com/arram/status/1642614341622181889
10. IQ estimates of public intellectuals and personas https://kirkegaard.substack.com/p/iq-estimates-of-public-intellectuals
11. Calling the Lab-Leak Theory 'Disinformation' Created Disinformation [The New York Times] https://archive.is/6YLW8
12. Do Politicians Ignore NIH Ties with Wuhan Lab to Protect Biodefense "Contract Racket"? https://disinformationchronicle.substack.com/p/do-politicians-ignore-nih-ties-with
👍4
"Great Awokening is a global phenomenon. No evidence it started in US media. Analysis of 98 million news articles across 36 countries quantifies. Exception: state-controlled media from China/Russia/Iran using wokeness terminology to criticize/mock the West"
https://davidrozado.substack.com/p/gag
https://davidrozado.substack.com/p/gag
👍4
Media is too big
VIEW IN TELEGRAM
Generative Novel View Synthesis with 3D-Aware Diffusion Models https://nvlabs.github.io/genvs/
👍3
Even in the absence of utopia or doom, it is questionable whether your retirement savings will have any value by mid-century. Because you don't need either to bring down our current socioeconomic system.
There are various scenarios that will cause some sort of massive disruption, such as rogue actors using advanced AI to develop bioweapons. And even if only 30% of jobs are lost. What happens then? Who will pay the taxes? You don't need artificial general intelligence for this to happen.
https://twitter.com/davidad/status/1643957252070449156
There are various scenarios that will cause some sort of massive disruption, such as rogue actors using advanced AI to develop bioweapons. And even if only 30% of jobs are lost. What happens then? Who will pay the taxes? You don't need artificial general intelligence for this to happen.
https://twitter.com/davidad/status/1643957252070449156
✍2👍2
Links for 2023-04-07
1. ChatArena 🏟 - a Python library of multi-agent language game environments that facilitates communication and collaboration between multiple large language models (LLMs)! 🌐🤖 https://github.com/chatarena/chatarena
2. Generative AI set to affect 300mn jobs across major economies [Financial Times] https://archive.is/SiJmV
3. The surprising ease and effectiveness of AI in a loop https://interconnected.org/home/2023/03/16/singularity
4. JARVIS, a system to connect LLMs with ML community, based on the new HuggingGPT paper. Language serves as an interface for LLMs to connect numerous AI models for solving complicated AI tasks! https://github.com/microsoft/JARVIS
5. Open-sourcing “Baby AGI”, a paired down version of the “Task-Driven Autonomous Agent” at 105 lines of code. https://github.com/yoheinakajima/babyagi (“His story is incredible. He went from idea to code. From code to paper. Paper to viral tweet thread. Thread to open-source project on GitHub.” https://twitter.com/NathanLands/status/1643563075885211649)
6. ChaosGPT: Empowering GPT with Internet and Memory to Destroy Humanity https://youtu.be/g7YJIpkk7KM
7. "Having multiple, incompatible desires is like having a miniature parliament in our heads: One faction argues for one thing, another argues for another, and the clash of perspectives often leads to better decisions." https://tricycle.org/article/working-parenthood-conflict/
8. Is China overhyped as an AI superpower? https://muddyclothes.substack.com/p/is-china-overhyped-as-an-ai-superpower
9. 80% of Chinese talent that contributed to GPT-4 got their graduate degrees in US and ended up staying https://macropolo.org/chinese-talent-generative-ai/
10. Americans: How concerned, if at all, are you about the possibility that AI will cause the end of the human race on Earth? Very concerned: 19%; Somewhat concerned: 27%; Not very concerned: 23%; Not at all concerned: 17%; Not sure: 13% https://today.yougov.com/topics/technology/survey-results/daily/2023/04/03/ad825/3
11. 10 obvious reasons that the danger from AGI is way more serious than nuclear weapons https://twitter.com/ESYudkowsky/status/1643800261599825921
12. Yoshua Bengio: "Slowing down development of AI systems passing the Turing test" https://yoshuabengio.org/2023/04/05/slowing-down-development-of-ai-systems-passing-the-turing-test/
13. AI, MRI and the legal system. What if there were no lies? https://kirkegaard.substack.com/p/ai-mri-and-the-legal-system
1. ChatArena 🏟 - a Python library of multi-agent language game environments that facilitates communication and collaboration between multiple large language models (LLMs)! 🌐🤖 https://github.com/chatarena/chatarena
2. Generative AI set to affect 300mn jobs across major economies [Financial Times] https://archive.is/SiJmV
3. The surprising ease and effectiveness of AI in a loop https://interconnected.org/home/2023/03/16/singularity
4. JARVIS, a system to connect LLMs with ML community, based on the new HuggingGPT paper. Language serves as an interface for LLMs to connect numerous AI models for solving complicated AI tasks! https://github.com/microsoft/JARVIS
5. Open-sourcing “Baby AGI”, a paired down version of the “Task-Driven Autonomous Agent” at 105 lines of code. https://github.com/yoheinakajima/babyagi (“His story is incredible. He went from idea to code. From code to paper. Paper to viral tweet thread. Thread to open-source project on GitHub.” https://twitter.com/NathanLands/status/1643563075885211649)
6. ChaosGPT: Empowering GPT with Internet and Memory to Destroy Humanity https://youtu.be/g7YJIpkk7KM
7. "Having multiple, incompatible desires is like having a miniature parliament in our heads: One faction argues for one thing, another argues for another, and the clash of perspectives often leads to better decisions." https://tricycle.org/article/working-parenthood-conflict/
8. Is China overhyped as an AI superpower? https://muddyclothes.substack.com/p/is-china-overhyped-as-an-ai-superpower
9. 80% of Chinese talent that contributed to GPT-4 got their graduate degrees in US and ended up staying https://macropolo.org/chinese-talent-generative-ai/
10. Americans: How concerned, if at all, are you about the possibility that AI will cause the end of the human race on Earth? Very concerned: 19%; Somewhat concerned: 27%; Not very concerned: 23%; Not at all concerned: 17%; Not sure: 13% https://today.yougov.com/topics/technology/survey-results/daily/2023/04/03/ad825/3
11. 10 obvious reasons that the danger from AGI is way more serious than nuclear weapons https://twitter.com/ESYudkowsky/status/1643800261599825921
12. Yoshua Bengio: "Slowing down development of AI systems passing the Turing test" https://yoshuabengio.org/2023/04/05/slowing-down-development-of-ai-systems-passing-the-turing-test/
13. AI, MRI and the legal system. What if there were no lies? https://kirkegaard.substack.com/p/ai-mri-and-the-legal-system
"OpenAPI agents: We can construct agents to consume arbitrary APIs, here APIs conformant to the OpenAPI/Swagger specification."
Here an agent plans and executes 5 different API calls to complete a user request: https://python.langchain.com/en/latest/modules/agents/toolkits/examples/openapi.html#st-example-hierarchical-planning-agent
Here an agent plans and executes 5 different API calls to complete a user request: https://python.langchain.com/en/latest/modules/agents/toolkits/examples/openapi.html#st-example-hierarchical-planning-agent
Media is too big
VIEW IN TELEGRAM
Here is a reminder that Meta AI is working toward general-purpose embodied AI agents that can assist humans in both the virtual & physical world: https://ai.facebook.com/blog/robots-learning-video-simulation-artificial-visual-cortex-vc-1/
It can already follow voice commands to carry out complex sensorimotor functions while navigating around people and furniture in a completely new house.
It can already follow voice commands to carry out complex sensorimotor functions while navigating around people and furniture in a completely new house.
🥰1
"When an "imaging artifact" turns out to be one of the wildest things we've ever observed in the universe.
A supermassive black hole.
Ejected from its galaxy.
Compressed gas in its wake.
AND FORMED A 200,000 LIGHT-YEAR LONG STRING OF STARS."
https://twitter.com/SpaceSciSteph/status/1644050461698453504
A supermassive black hole.
Ejected from its galaxy.
Compressed gas in its wake.
AND FORMED A 200,000 LIGHT-YEAR LONG STRING OF STARS."
https://twitter.com/SpaceSciSteph/status/1644050461698453504
🔥10❤1👍1
Tweet: https://twitter.com/karpathy/status/1644183721405464576
See also:
Looped Transformers as Programmable Computers: “We present a framework for using transformer networks as universal computers by programming them with specific weights and placing them in a loop…Our work highlights the versatility of the attention mechanism, and demonstrates that even shallow transformers can execute full-fledged, general-purpose programs.” https://arxiv.org/abs/2301.13196
See also:
Looped Transformers as Programmable Computers: “We present a framework for using transformer networks as universal computers by programming them with specific weights and placing them in a loop…Our work highlights the versatility of the attention mechanism, and demonstrates that even shallow transformers can execute full-fledged, general-purpose programs.” https://arxiv.org/abs/2301.13196
This media is not supported in your browser
VIEW IN TELEGRAM
ANT DEATH SPIRAL: Ants in many species find their way round by following pheromone trails left by nestmates. Occasionally, though, the trails accidentally loop, and the ants end up walking in a circle till they die of exhaustion. https://en.wikipedia.org/wiki/Ant_mill
✍4
Links for 2023-04-08
1. Better Language Models of Code through Self-Improvement — “The results show that our framework significantly improves PLMCs' performance in code-related sequence generation tasks, such as code summarization and code generation in the CodeXGLUE benchmark.” https://arxiv.org/abs/2304.01228
2. Anthropic’s $5B, 4-year plan to take on OpenAI — Anthropic plans to train a powerful model with billions in new funding https://techcrunch.com/2023/04/06/anthropics-5b-4-year-plan-to-take-on-openai/
3. MATHEMATICS AND THE FORMAL TURN: “Since the early twentieth century, it has been understood that mathematical definitions and proofs can be represented in formal systems systems with precise grammars and rules of use. Building on such foundations, computational proof assistants now make it possible to encode mathematical knowledge in digital form. This article enumerates some of the ways that these and related technologies can help us do mathematics.” https://www.andrew.cmu.edu/user/avigad/Papers/formal_turn.pdf
4. "Ending an Ugly Chapter in Chip Design: Study tries to settle a bitter disagreement over Google’s chip design AI" (independent replication of RL chip design work shows competitive with other approaches despite omitting pretraining) https://spectrum.ieee.org/chip-design-controversy
5. SemDeDup: Data-efficient learning at web-scale through semantic deduplication https://twitter.com/davisblalock/status/1641222279509196801
6. Moderate drinking is still good for you: "How to lie with statistics advanced edition...This meta-study claims to overturn the long established pattern that moderate drinkers have lower mortality than abstainers. After adjusting for confounders there was "no significant reduction in mortality". Very misleading..." https://twitter.com/0xdoug/status/1643993363623084038
7. Don’t Take Batteries to the Moon or Mars, 3D Print Them When you Get There: 3D printing batteries using local materials https://www.universetoday.com/160703/dont-take-batteries-to-the-moon-or-mars-3d-print-them-when-you-get-there/
8. "Our study suggests that sociological theories explaining class outcomes in terms of social origins have little explanatory power, and should be reformulated to consider genetics." https://osf.io/preprints/socarxiv/mncet/
9. “In a renewed attempt at a grand unified theory of brain function, physicists now argue that brains optimize performance by staying near — though not exactly at — the critical point between two phases.” [published in 2018] https://www.quantamagazine.org/brains-may-teeter-near-their-tipping-point-20180614/
10. Adding 0.075% by weight of graphene to polyacrylonitrile carbon fibres can increase their strength by 225% and Young modulus by 184%. Strangely, the effect reverses if we add more than 0.1% graphene, by making the fibres weaker... https://www.science.org/doi/10.1126/sciadv.aaz4191
11. John von Neumann asks, “Can we survive technology?” https://geosci.uchicago.edu/~kite/doc/von_Neumann_1955.pdf (Quotes: 1. “I should like to mention, as an example of this, the phase of computing machines. It is probably true that since 1945 the over-all capacity of these machines has nearly doubled every year. This is astounding because over a period of ten years it means a thousandfold increase.” 2. “The carbon dioxide released into the atmosphere by industry's burning of coal and oil-—more than half of it during the last generation—may have changed the atmosphere's composition sufficiently to account for a general warming of the world by about one degree Fahrenheit.”)
12. “Today, we stood up to China.” - Kevin McCarthy on his meeting with Taiwan President Tsai Ing-wen. https://twitter.com/HouseGOP/status/1644078557897936901
1. Better Language Models of Code through Self-Improvement — “The results show that our framework significantly improves PLMCs' performance in code-related sequence generation tasks, such as code summarization and code generation in the CodeXGLUE benchmark.” https://arxiv.org/abs/2304.01228
2. Anthropic’s $5B, 4-year plan to take on OpenAI — Anthropic plans to train a powerful model with billions in new funding https://techcrunch.com/2023/04/06/anthropics-5b-4-year-plan-to-take-on-openai/
3. MATHEMATICS AND THE FORMAL TURN: “Since the early twentieth century, it has been understood that mathematical definitions and proofs can be represented in formal systems systems with precise grammars and rules of use. Building on such foundations, computational proof assistants now make it possible to encode mathematical knowledge in digital form. This article enumerates some of the ways that these and related technologies can help us do mathematics.” https://www.andrew.cmu.edu/user/avigad/Papers/formal_turn.pdf
4. "Ending an Ugly Chapter in Chip Design: Study tries to settle a bitter disagreement over Google’s chip design AI" (independent replication of RL chip design work shows competitive with other approaches despite omitting pretraining) https://spectrum.ieee.org/chip-design-controversy
5. SemDeDup: Data-efficient learning at web-scale through semantic deduplication https://twitter.com/davisblalock/status/1641222279509196801
6. Moderate drinking is still good for you: "How to lie with statistics advanced edition...This meta-study claims to overturn the long established pattern that moderate drinkers have lower mortality than abstainers. After adjusting for confounders there was "no significant reduction in mortality". Very misleading..." https://twitter.com/0xdoug/status/1643993363623084038
7. Don’t Take Batteries to the Moon or Mars, 3D Print Them When you Get There: 3D printing batteries using local materials https://www.universetoday.com/160703/dont-take-batteries-to-the-moon-or-mars-3d-print-them-when-you-get-there/
8. "Our study suggests that sociological theories explaining class outcomes in terms of social origins have little explanatory power, and should be reformulated to consider genetics." https://osf.io/preprints/socarxiv/mncet/
9. “In a renewed attempt at a grand unified theory of brain function, physicists now argue that brains optimize performance by staying near — though not exactly at — the critical point between two phases.” [published in 2018] https://www.quantamagazine.org/brains-may-teeter-near-their-tipping-point-20180614/
10. Adding 0.075% by weight of graphene to polyacrylonitrile carbon fibres can increase their strength by 225% and Young modulus by 184%. Strangely, the effect reverses if we add more than 0.1% graphene, by making the fibres weaker... https://www.science.org/doi/10.1126/sciadv.aaz4191
11. John von Neumann asks, “Can we survive technology?” https://geosci.uchicago.edu/~kite/doc/von_Neumann_1955.pdf (Quotes: 1. “I should like to mention, as an example of this, the phase of computing machines. It is probably true that since 1945 the over-all capacity of these machines has nearly doubled every year. This is astounding because over a period of ten years it means a thousandfold increase.” 2. “The carbon dioxide released into the atmosphere by industry's burning of coal and oil-—more than half of it during the last generation—may have changed the atmosphere's composition sufficiently to account for a general warming of the world by about one degree Fahrenheit.”)
12. “Today, we stood up to China.” - Kevin McCarthy on his meeting with Taiwan President Tsai Ing-wen. https://twitter.com/HouseGOP/status/1644078557897936901
👍1
This media is not supported in your browser
VIEW IN TELEGRAM
Towards ML-enabled cleaning robots https://ai.googleblog.com/2023/04/towards-ml-enabled-cleaning-robots.html
"The results from this work demonstrate that complex visuo-motor tasks such as table wiping can be reliably accomplished without expensive end-to-end training and on-robot data collection. The key consists of decomposing the task and combining the strengths of RL, trained using an SDE model of spill and crumb dynamics, with the strengths of trajectory optimization."
"The results from this work demonstrate that complex visuo-motor tasks such as table wiping can be reliably accomplished without expensive end-to-end training and on-robot data collection. The key consists of decomposing the task and combining the strengths of RL, trained using an SDE model of spill and crumb dynamics, with the strengths of trajectory optimization."
🔥3