A.J. Kourabi

8 followers · 5 following · 627 views

on the atlas · 39

highlights · 265

The 0% threshold is effective but has high cost. If AI is going to transfrom society and the world over the next decade, then it must be applied. The national security risk of not applying it, is too grave.
Fab Whack-A-Mole: Chinese Companies are Evading U.S. Sanctions – SemiAnalysis
Note even the easy tools are imported in huge volumes because “easy” in semiconductors is more complicated and harder than most entire industries.
Fab Whack-A-Mole: Chinese Companies are Evading U.S. Sanctions – SemiAnalysis
By August 2022, AWS had over 20 million Nitro parts installed over four generations, with every new EC2 server installing at least one Nitro part. The primary cost benefit of this “dongle” is that it offloads Amazon’s management software, the hypervisor, which would otherwise run on an existing CPU. The most commonly deployed CPU across Amazon’s infrastructure was, and still is, an Intel 14nm 24-core CPU. Even to this day, other clouds, such as Microsoft Azure eat as many as 4 CPU cores on workloads that aren’t the customers’. If this held true across all of Amazon’s infrastructure, that would…
Amazon’s Cloud Crisis: How AWS Will Lose The Future Of Computing – SemiAnalysis
There are enough chips being shipped to China + manufactured domestically to create the world’s largest AI training cluster.
Fab Whack-A-Mole: Chinese Companies are Evading U.S. Sanctions – SemiAnalysis
(1) Staying at the forefront of AI development will likely require being able to build five gigawatt clusters2 within five years.
How to Build the Future of AI in the United States - Institute for Progress
we examine the technological challenges to building the next generation of AI data centers in America. We make five conclusions:
How to Build the Future of AI in the United States - Institute for Progress
There’s the well-planned, evolutionary, sustaining innovation, which is where you have a road map and, based on this road map, you change the way you do business in a fairly predictable manner
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
Note how the radical, disruptive research requires accepting a performance drop before a performance increase, and so cannot be found via a locally greedy optimisation.
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
Instead all current pressures and incentives align with satisfying current peers.
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
encouraging different systems to collectively explore diverse parts of the unknown rather than low-risk herding
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
‘Despite dramatic advances over the past few centuries, in recent decades biotechnology hasn’t met the expectations of investors—or patients. Eroom’s law—that’s Moore’s law backward—observes that the number of new drugs approved per billion dollars spent on R&D has halved every nine years since 1950.’
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
Remaining dynamic: do not give tenure to researchers
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
Organise as self-organising start-up style small teams, not hierarchical bureaucracies
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
Invest long-term in promising world class scientists instead of individual projects
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
Identify unique and long-term visions These long-term visions become talent attractors
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
Fund scientific communities not scientific factories: “focussed ecosystems pursuing tech visions”
S&T - 2019 Working notes titled 'Research ecosystems need structural diversity' - learning from past sci/tech revolutions at Bell labs/PARC/LMB to design new institutes
Borealis offers dynamic programmability over all gates used, shows true photon-number-resolved detection and requires a much more modest number of optical components and paths
Quantum computational advantage with a programmable photonic processor | Nature
We have successfully demonstrated quantum computational advantage in GBS using a photonic time-multiplexed machine
Quantum computational advantage with a programmable photonic processor | Nature
We carry out Gaussian boson sampling4 (GBS) on 216 squeezed modes entangled with three-dimensional connectivity5, using a time-multiplexed and photon-number-resolving architecture. On average, it would take more than 9,000 years for the best available algorithms and supercomputers to produce, using exact methods, a single sample from the programmed distribution, whereas Borealis requires only 36 μs.
Quantum computational advantage with a programmable photonic processor | Nature
The cooler science becomes, the more people become scientists for the wrong reasons.
Market Failures in Science
Scientists are incentivized to produce results their funders will like.
Market Failures in Science
Teaching is its own skill, distinct from research, and most scientists are bad at it. We shouldn’t force them to do it.
Market Failures in Science
Universities get paid by amount of grant money, not scientific output.
Market Failures in Science
Work everyone agrees someone should do but which nobody does, because it’s inadequately rewarded in money or reputation.
Market Failures in Science
In particular, we believe current social processes in science are designed to support work in existing fields, but strongly inhibit work critical to the creation of new fields. What programs would dramatically increase the rate of production of fruitful new fields
A Vision of Metascience
how much demand is there for discipline switching?
A Vision of Metascience
The underlying thesis is that there are many such people who have extremely unusual combinations of skills, skills unlikely to be found in academia, but which may enable important discoveries
A Vision of Metascience
Rather than basing the way we fund science on custom or anecdote, we can shift to a norm in which science funders build research into the way they support research, allowing them to measure and improve their own effectiveness over time.
Developing the science of science - Works in Progress
particularly in the field of international development.
Developing the science of science - Works in Progress
But despite how much society spends on science, we know surprisingly little about how different structures, incentives, and organizational models shape the rate and direction of science.
Metascience Working Group - Using science to improve science
I think that you should engage in self-reflection, not engage in being obtuse.
Why I'm personally upset with Nick Bostrom right now
Instead, think about why you currently believe this thing, and try to accurately describe what led you to believe it.
How To Write Quickly While Maintaining Epistemic Rigor - LessWrong
nd technological winters of hype cycles, a shared language around technology development could help us smoothly ramp up long term projects to build the future.
Technology Readiness Levels
A common language around technology maturity could enable people to swap useful mental models more quickly between different domains. At the same time, TRLs can help frame the difference between domains and help you realize when it doesn’t make sense to impose an idea from one domain onto another.
Technology Readiness Levels
TRLs illustrate that planning and project management are still important and perhaps there is room to create new development methods depending on a domains unique constraints.
Technology Readiness Levels
However, it also means that you can’t just port agile to other domains and expect it to work as well.
Technology Readiness Levels
This is doing TRL2 terribly. TRL2 is secretly hard because it’s very easy to just check the box and go on. TRL2 is the design step.
Technology Readiness Levels
reduced risks from emerging technologies,
Epistemics and institutional decision-making - 80,000 Hours
The more I think about this, the more I am astounded by the fact that some of the things we most cherish and value in life cannot be pursued directly
Matthew Jordan’s review of Games: Agency as Art
More rigorously testing existing techniques that seem promising. Doing more fundamental research to identify new techniques. Fostering adoption of the best proven techniques in high-impact areas. Directing more funding towards all of the above.
Epistemics and institutional decision-making - 80,000 Hours
What’s more, there’s reason to think that focusing on institutions directly might be a more effective way to improve decision-making than a broad approach to improved education, as it targets a smaller set of people who already have a lot of influence, and focuses more on institutional processes which can often have a big impact on how high-stakes decisions actually get made.
Epistemics and institutional decision-making - 80,000 Hours
A structure that fits the familiar pattern of "masses recognize elites who oversee experts" Any intentionally designed large-scale structure that mediates human interaction (including things like financial markets, social media platforms and dating sites) Widely spread and standardized social customs in general
What even is an institution?
Literally every thing that I labeled as clearly involving interaction had a higher percentage of people considering it an institution than every thing I labeled as not involving interaction. The single dot in the center is my hypothetical example of an island where people with odd-numbered birthdays are not allowed to eat meat before 12:00; I didn't want to give it 100% because the not-meat-eating is a private activity, but the question still strongly implies some social or other pressure to follow the rule so it's also not really 0%. This is a place where Spearman's coefficient outperforms Pe…
What even is an institution?
Does it have roles that take on a life independent of the individuals that fill them?
What even is an institution?
Masses recognize elites, who oversee experts, who pick details.
What even is an institution?
Masses recognize elites, who oversee experts, who pick details.
Overcoming Bias : Elites Must Rule
Being around intense people lets you sink into your intensity, without needing to dim your drive.
intensity - by Isabel - Mind Mine
Prioritizing building AI systems that could do especially helpful things, such as contributing to AI safety research and threat assessment and patching security holes.
Racing through a minefield: the AI deployment problem
Establishing governance that is capable of making hard, non-commercially-optimal decisions for the good of humanity.
Racing through a minefield: the AI deployment problem
Plus there are some powerful sources of error you need to learn to discount. Are you really interested in x, or do you want to work on it because you'll make a lot of money, or because other people will be impressed with you, or because your parents want you to?
How to Work Hard