JamilD

NIMBYs often advocate for building “out” rather than “up”, either explicitly or by omission (saying not to build any housing results in this; people need housing somewhere).

It results in sprawl, car dependence, inefficiency, reduced access to jobs/retail/parks. Regardless of intention, selfish or not, I can’t imagine worse results for the environment and our cities.

A law can be "not racist", but still be crafted with certain intentions in mind; it's easy to exploit certain correlations to achieve a certain end result. I don't think anyone would argue that there's something innate about one's skin color that would make them predisposed to living in apartments! But if you want to exclude certain people from your neighborhood, and those people happen to often be from a certain culture/socioeconomic class…

It's also worth noting that the same laws were applied to Asians - exclusionary zoning and housing policies led to the consolidation of populations within certain neighborhoods, like San Francisco's Chinatown: https://www.nbcnews.com/news/asian-america/how-1800s-racism-...

It wasn't always the case that most Asians could (whether financially or politically) move to the suburbs.

A tl;dr, because the idea behind the Decision Transformer is really cool and elegant; they model reinforcement learning tasks in a similar way to autoregressive language modeling.

In language modeling, you want to learn, for example, the next word in a sentence. Given a sequence of tokens t₀, t₁, t₂, predict t₃.

The authors here model RL as a similar autoregressive task, where we want to predict which action to take, given a sequence of previous actions, states, and "rewards-to-go", or the estimated remaining rewards in the trajectory. For example, given (s₀, a₀, R₀), (s₁, R₁), we want to predict the best a₁.

This allows us to train in a supervised way by accumulating trajectories from either training data or random walks.

Then, at inference time, all we do is input into the model (a) the start state s₀, and (b) the desired "rewards-to-go" R₀ we'd like to have.

The output at the next step is an action a₀. We then calculate the next state s₁ caused by taking a₀, and the remaining rewards-to-go R₁, to find the best a₁. We can repeat this with partial trajectories until the rewards-to-go hit 0, or until the trajectory is complete.

> And while it can still be accessed over HTTP, the bottlenecks become the same companies that might comply with censorship requests.

This is something that bothers me about the state of dapps right now. The way many of them connect to the blockchain is via Infura (a centralized service); even Metamask uses Infura to connect to the blockchain. There’s this abstraction that you trust, that you’re working with something totally decentralized, but right now the technical constraints necessitate single points of failure or censorship.

Of course, the data is still there in the blockchain, decentralized over many nodes, but the way we access that information seems very, very brittle.

I recently started doing this too. I tend to always push myself and try to deliver as much as I can, limited by time and energy — it was hard to get rid of this habit, so I tried to create things outside of work to "bound" that time/energy limit.

I started with a challenge of reading and summarizing one ML/CS paper per day, which requires a reasonable enough amount of energy and time that I have to make sure to allocate outside of work. It was a catalyst to help me get back into "reclaiming" a portion of my time and mental energy, and I started getting back into side projects.

Taking back my time has given me a lot more purpose and meaning, and brings me back to a sense of curiosity and optimism about technology that I haven't had in a while. It's amazing what feeling some ownership can do, and how much better I feel putting my time and effort into things outside of work.

Note: initially I tried to "bound my work-related energy" with non-technical goals, like running and mountaineering. It worked to distract me, but I feel like I needed an intellectual bound as well as a physical one.

Compute power requires money though, no? Whoever has the most capital now controls the BTC blockchain, which happens to be mostly miners in China. It’s a lot easier for me to stake ETH than to buy hundreds of GPUs.

And my reward is linear with how much I stake or put down, whereas with economies of scale in mining the reward is more like quadratic. Seems to me that with PoS the rich get richer linearly, with PoW the rich get richer ~quadratically.

If I spend 0.01% of what the Chinese miners spend on my mining setup, I’m not getting 0.01% of the rewards.

Are the benefits of proof of work worth the additional cost over other consensus algorithms, to you?

Why is this a more direct reward than generating energy to sell?

In mining, revenue = efficiency * potential energy * BTC per unit produced energy

In selling energy, revenue = efficiency * potential energy * USD per unit produced energy

I don’t see the difference, other than the fact the former just adds to the pool of required energy (it’s strictly additive to the energy produced by the second)

Many of DeepMind's opinion style papers are like this. Another example of the "handwavy" DeepMind paper: https://arxiv.org/pdf/2102.03406.pdf

It's also worth it to note as well that this isn't a homogenous organization, many DeepMind employees have different opinions on issues like this and an individual paper isn't representative of the entire organization.

From the perspective of Richard Dawkins / The Selfish Gene, there are two distinct incarnations of evolution, driven by the same force (natural selection).

Genetic; unit of selection is the gene, substrate is DNA: responsible for our biological development; our eyesight, innate cognitive capacity, etc. Responsible also for the past ~4 billion years of exponential growth in biological complexity. Survival of the gene drives natural selection, so it requires embodied survival.

Memetic; unit of selection is the idea/“meme”, substrate is our brains: responsible for civilization, invention, cultural development. Responsible for the last ~20,000 years (or within that order of magnitude) of the development of culture, science, and civilization. Survival of the idea is what drives natural selection, not of the biological phenotype/person/brain carrying it.

The options for retirement funds in my 401(k) include "LifePath Index 2065 Account A", but that's as far as it goes for now.

Tom Gruber (one of the cofounders of Siri) had a talk a while back where he spoke about this. https://www.ted.com/talks/tom_gruber_how_ai_can_enhance_our_...

> Tom Gruber, co-creator of Siri, wants to make "humanistic AI" that augments and collaborates with us instead of competing with (or replacing) us. He shares his vision for a future where AI helps us achieve superhuman performance in perception, creativity and cognitive function -- from turbocharging our design skills to helping us remember everything we've ever read and the name of everyone we've ever met. "We are in the middle of a renaissance in AI," Gruber says. "Every time a machine gets smarter, we get smarter."

I'm a big believer in Intelligence Augmentation, in the Engelbart tradition, over completely independent agents. I don’t think my ideal AI is one that has its own sense of agency and autonomy, rather I think of technology as an extension of myself, one that enhances my autonomy.

There's a weird methodology quirk here. Looking at the data, words like "violence" seem to directly decrease the overall happiness score. Which is weird, because use of that word doesn't necessarily indicate that someone is feeling less happy.

Mentioning negative events therefore seems to directly decrease the happiness score, so a conclusion like yours is almost tautological -- when a negative event happens, people talk about it, so the algorithm infers that people are unhappy.

Thanks for those links! Didn't know Amadeus had something similar.

Just like with Sabre, it seems all the "juicy" stuff is in the Enterprise API (which is opaque on pricing and setup -- you have to contact the sales team to get access).

Do you have any insight into whether it's possible for individual developers to get access to that enterprise API, and what the fees look like?

A few years ago I tried writing a Python wrapper around SABRE's API to get pricing, route, and schedule data, which seemed to work reasonably well. It likely doesn't work anymore, but it was a fun exercise. https://github.com/Jamil/sabre_dev_studio

I wish I had access to the GDS data to get realtime seat/award availability, but I couldn't find any pricing information to get that information through Sabre's API.

Does anyone know how much that costs, or if there are any services which provide it as an API? I use ExpertFlyer for personal use, but ideally I'd want to get that information at the source…

I wonder how we escape this trap.

Cars are 1-2 ton machines that can kill you as a cyclist or pedestrian, and humans are awful at operating them. The solution is therefore apparently to put more of those machines on the road (for self-protection) instead of reducing the risk altogether?

Likewise. Once in a while I force myself to do backprop by hand to make sure I still understand all the concepts and can walk myself through the calculus.

It's interesting how we can use age (selectively) as a discriminator, but within a certain range and in certain contexts, it's seen as "bad" to discriminate.

Most wouldn't say the concept of a retirement age is discriminatory, nor the concept of age of consent or minimum age to purchase alcohol, or pricing insurance rates differently based on age.

Somehow this application of age discrimination seems to cross the line for many people, though. I'm genuinely curious as to why.

> only do what maximizes their own financials

So if a company feels like they can acquire more customers or employees, or gain positive publicity by taking certain actions and expressing their right to free speech, shouldn't they be able to? This is why businesses take political stances in the first place. A company would be doing a disservice to shareholders by _not_ taking an action which could improve financials.

Before gay marriage was legal, for example, companies would stand up for gay rights and offer gay employees equal benefits, because it would help with talent acquisition and expand their pool of possible employees. This is a "political" stance with clear selfish, financial implications.

You have the right to not shop or procure services at a business which does not support your ideology, but others have their own right to choose companies based on the actions they take. That's the beauty of the free market.

Regulating or otherwise stifling companies' speech and actions in this way simply hinders progress and the marketplace of ideas.

Undergrad courses are 100-level, graduate courses are 200-level, advanced graduate courses are 300-level.

A lot of people underestimate how important this is for us. So many countries in which I'd love to live are ruled out immediately, and we simply don't have the flexibility to only take into consideration cost of living, natural beauty, family, etc.

Relatively speaking, are very few places in the world where you can live a "normal" life as an LGBT person.

When OS X Snow Leopard was announced, this was the intro slide: http://2ig18m1zutag3bjpoclbo8am-wpengine.netdna-ssl.com/wp-c...

I’m in the middle of reading “The Outlaw Ocean”; really eye-opening to see how much people can get away with in international waters, away from any scrutiny or accountability.

You do. There are so many referenda, at least in the US, on seemingly inconsequential things that end up having a huge effect on the layout and organization of a city. Transportation taxes, light rail lines, zoning, affordable housing minimums… in aggregate these policies can remake a city.

With respect to point (2), the argument doesn’t really work for SF; rent control only applies to units built before 1979. New developments can raise the rent as much as they want.

You could argue the threat of possibly having rent control discourages new building, but it’s a much weaker argument.

Your points in (3) I think are much, much more relevant to the SF housing crisis than rent control.

> I'm sure Uber engineering are tackling tough technical problems

They definitely are, but I think the point is that those technical problems aren’t core to the service. The value that Uber provides to you is a platform for independent, cheaper taxis – the routing algorithms and ML add to that, but the technology isn’t core to the product or the value of it.

For tourism and appeal for businesses it can have a clear ROI; both of those are important for Hong Kong.

Not just technology, but the landscape of our cities too; suburbia, car culture, skyscrapers in the urban core; there haven’t really been radical transformations there either.

Their business model is interesting:

> The difference is that Walmart won't require a monthly subscription to access these shows, but will instead generate money through "new advertising technology." This is described as "shoppable" content, with television shows that can directly lead to customers buying products online: "Not only will viewers see ads, they will have the option to buy products seen in shows, such as paper towels or soft drinks."

The article implies so:

> They quarantined Chinatown, preventing any movement of food in, or of its inhabitants — many of them cleaners, cooks and labourers — out. […] Outside Chinatown, the middle-class white population was outraged by the disruption.