668 karma · joined October 5, 2010
AI & crypto may end up in a similar bucket. It is a paradoxical enabler of both freedom and imprisonment. No one gets to build a technology a declare a problem solved. Alfred Nobel made nitroglycerin safe and thus paradoxically ended up being called the Merchant of Death.
Freedom is not free. It is an eternal struggle.
I am grateful for the builders who try to build something for freedom. I do not spurn them when their invention gets corrupted. It is now our job and the job of future generations to fight the corruption.
For now, the human dissenters have a lot of leverage because AI still makes very clear and obvious errors sometimes, and it would be a political nightmare for a decision-maker to be accused of erring on the side of AI by recognized human experts who dissented. I wonder if there will come a time that AI opinion would be on par or even favored over the human expert because the human would be considered more fallible. This doesn't even have to be true - it only has to be sufficiently perceived to be sufficiently true.
The abstract suggests that elites "shape" mass preference, but I think the degree to which this shaping occurs is overblown in many ways (and perhaps underestimated in other ways, such as through education).
AI, even if it is not powerfully "shaped" by the "elites", can push mass preference in predictable ways. If this is true, this phenomenon by itself allows the elites to tighten their grip on power. For example, Trump's rise to power upset (some of) the elites because they really didn't understand the silent, mass preference for Trump.
This could also slow social progress, since elites often cause stagnation rather than progress. AI could generate acceptable, "expert" opinions for the issues that they usually would rely on experts today. I see some signs of that today, where those with authority try to prefer the AI answer in opposition to dissenting, human expert opinions. Human experts seem to be winning, for now.
I work at Google, and I am of the overall opinion that it doesn't matter what you deliver from an engineering perspective. I've seen launches that changed some behavior from opt-in to opt-out get lauded as worth engineering-years of investment. I've seen demos that were 1-2 years ahead of our current product performance get buried under bureaucracy and nitpicking while the public product languishes with nearly no usage. The point being, what you objectively deliver doesn't matter, but what ends up mattering is how the people in your orbit weave the narrative about what you built.
So if "leadership" wants something concretely done, they must mandate it in such a way that cuts through all the spin that each layer of bureaucracy adds before presenting it to the next layer of bureaucracy. And "leadership" isn't a single person, so you might describe leaders as individual vectors in a vector space, and a clear eigenvector in this space of leadership decisions in many companies is the vector of "increase employee usage of AI tools".
In response to your comment, yes, I would largely be in favor of moving forward only with whatever is said in the relevant meetings with the given attendees of a meeting. That assumes a reasonably healthy culture where these meetings are scheduled in good faith reasonable times for all relevant stakeholders.
I always laugh-cry with whomever I'm sitting next to whenever launch announcements come out with more people in the "leadership" roles than the individual contributor roles. So many "leaders" but none with the awareness or the care of the farcical volumes such announcements speak.
Roles fixing FFmpeg bugs would be a hard sell in this environment, imho.
It's hard to commit to a green-field project that is predicated on a level of risk that leaders are hesitant to take on as they would also take on the "counterparty risk" of the expert individual contributors holding them over the barrel to finish the project. The expert individual contributors are likewise hesitant to devote themselves to the task knowing that once leadership considers the "hard bits" to be done, leadership's aversion to risk will try to swap out the expert-individual-contributor roles with much more replaceable roles and ultimately replace the expert individual contributors.
P.S. The parallels between this cycle of mistrust and the modern dating crisis (in the US) does not elude me.
Leaders can often be incompetent since they are often not promoted based on competency or they hang on to power long enough to either have their competence become stale or just lose it do to the passage of time.
Ultimately, AI will decide for us since it will be the crutch of the incompetent leaders. Since these leaders often decide "truth" and "reality" for most people anyways, AI will decide these truths with leaders as a proxy, and power will continue to flow in directions completely unrelated to competence.
Unless there is some sort of blowback, this sort of thing is likely to happen again to someone, and I understand how some people may not want to be involved anymore, and I understand how Mozilla will keep being Mozilla (just like other organizations will continue their current behaviors until some catalyst changes that behavior).
I think people have been really underestimating the power of small organized groups, and I have been putting significant effort into invigorating the small groups around me.
Tao hints at some of the values of small groups, but I think he misses the irreplaceable value that small groups have. We as a society have conceptualized value itself into certain quantities, mostly in dollar amounts (time, money, income, wealth, assets, services, etc) that we have lost our humanity. I find it amusing that Tao describes "a sense of purpose" as a "softer" benefit, and I agree that this is the lens by which society implicitly sees "a sense of purpose" - a sense of purpose does not have a monetary value that can be assigned to it. Even so, what is all the money in the world worth without a sense of purpose?
Tao correctly points out that "large organizations" "offer substantially more economies of scale" and "provide significant material comforts", I wonder if humanity is ready to wonder if it really needs more material comforts at a global scale. Perhaps we can start paring down some of our global organizations into their core functions of providing material comforts and we can start invigorating small organizations to nourish our souls.
We spend so much time, energy, and mental anguish over challenges like war and political gridlock. Our natural inclination to consolidate power into large organizations to overcome these challenges may be counterproductive because these large organizations naturally dehumanize us. And when this dehumanized organization inevitably fail to achieve the original noble goals of e.g. strongarming Russia into peace with Ukraine, we naturally try to push for more power and use different forms of aggression or increase the magnitude of our aggression, not only to try to "conquer" the "problem" but also feel safe in our membership into a big and powerful organization capable of such aggression. This might also drive us to abandon "small organizations" for some "great cause". For example, look at all the time and energy we put into global political coalitions on social media rather than local causes.
Tao may be right that we need to fundamentally rethink how these different levels and sizes of organizations need to engage with one another. When we encounter a problem, even a big problem, a big organization sometimes is not only ineffective but they can be actively harmful.
I did my usual, ran a few miles on the treadmill, did family walks all over, etc. I just felt lethargic - like I didn't sleep well at all, but I powered through it.
I know people who were absolutely bedridden with the flu. I can 100% believe some people can almost entirely brush it off too.
There are a lot of technical challenges in maintaining code health in a monorepo with 100k+ active contributors, so teams and individuals get a lot of plausible excuses for kicking the problem down the road, and truly improving code health is not appropriately incentivized. One common occurrence is a broken monorepo, so one just waits until someone fixes the monorepo, and you retry submitting your code again. It's such a common occurrence that people generally do not investigate brokenness, and maybe the monorepo wasn't broken but your code change actually made things even flakier, but no one would be able to distinguish that from a broken monorepo that eventually got fixed when no one bothers to check anymore.
It took time as different firms adapted to adopt computer technologies in their various business needs and workflows. It's hard to precisely predict how labor roles will change with each revolutionary technology.
Perhaps you feel like this needs to be called out more explicitly rather than just hinted at. I think another concern that the author doesn't _highlight_ but is plenty clear is that this sort of lifestyle probably is more amenable to the able bodied, with their frequent allusions to hunting and fishing.
So while the author doesn't explicitly list out all possible issues, and everyone will have different priorities and opportunities in their life, I think the author is plenty clear on what they're describing, and the concern you raised is easily inferred from what they're saying.
Exactly - money is hype. Hype is money. Which goes into the foundation of money to begin with: confidence and faith. And the levers of power which move that confidence and faith are decreasing in number and increasing in length/leverage.
> Precise but not accurate. Since it is always open to interpretation, it simply cannot be used for verification
Agreed - the main way I try to manage this is to have a very coarse interpretation of statements - almost like a probability distribution of semantic meaning[1]. For example, if someone says "I like spaghetti", most of the probability distribution is located somewhere between "I tolerate spaghetti, but I won't say it that way in case you're about to serve me spaghetti" to "My favorite food is spaghetti, but only if it's served in the general style of my hometown".
I might be misinterpreting what you're saying right now, but we can continue to have this conversation, and I hope you're enjoying it even as I am enjoying it!
I generally agree with the idea that it cannot be used for verification, depending on what you precisely mean (my probably distribution of _your meaning_ has significant overlap with what I would agree with, but not entirely, though to be pedantic, it never entirely overlaps since there's always a chance of misinterpretation). I think that's why it's important to establish relationships and baselines of communications. Repeatedly successfully interpreting each other's words can provide a strong baseline for future communicative success. It's important not to project one relationship onto another, which we are _very_ prone to do. E.g. we should not project how we communicate with one sex to another sex, or necessarily even one person to another person!
[1] Maybe this is the next representation of semantic meaning that LLMs will use somewhere in its internals? Instead of a single embedding, it describes some probability distribution around an embedding with the greatest probability?
Communication is so automatic for us that we may not even realize or understand how we communicate. So many people don't even take the time to understand their own thoughts, and understanding our own communication is many times harder because you have to inspect your own thoughts and be constantly building a model for how your communication is affecting the minds of the people around you. You had a specific idea for how communication works, just because you're both speaking the same language (e.g. English), in many ways, you might not have been speaking the same language since the ideas one person intended to transmit in good faith still did not end up in the mind of their audience.
> Sam’s Letter to Employees.
> OpenAI is not a normal company and never will be.
Where did I hear something like that before...
> Founders' IPO Letter
> Google is not a conventional company. We do not intend to become one.
I wonder if it's intentional or perhaps some AI-assisted regurgitation prompted by "write me a successful letter to introduce a new corporate structure of a tech company".
I was even more blown away when the whole thing becomes old, forgotten news the moment the customer stopped threatening to cancel the contract, even if the underlying problem remains.
This really reinforces the lesson that the main power structures of the world does not listen to "reason" - it only responds to incentives, whether they're airlines or ginormous tech companies.
Hacker News has also repeatedly noticed the same thing whenever a big tech customer issue hits the top of Hacker News, and the comments point out that the only way that customer got help was when it got enough attention to cause a reputation risk to the big tech company.
At a general level, I believe there are other options - changes/features need to meet some level of usage or it is scrapped out of recognition that supporting all these features make bugs more likely, performance likely to degrade, increase difficulty of adding features, make the product more difficult to use, etc.
The cynical spin I would put on this idea is that software performs as poorly as it can get away with. MSFT is feeling the need/pressure to have Office load faster, and they will try to get away with preloading it.
Otherwise, there is a strong pull towards bloat that different people will try to take credit for as features even if the cumulative experience of all these "features" is actually a worse user-experience.
Could the DoJ behave sufficiently separately from the office of the President to conduct a proper prosecution?