This resonates with me. Sometimes I build up some artifacts within the context of a task, but these almost always get thrown away. There are primarily three reason I prefer a vanilla setup.
1. I have many and sometimes contradictory workflows: exploration, prototyping, bug fixing debugging, feature work, pr management, etc. When I'm prototyping, I want reward hacking, I don't care about tests or lint's, and it's the exact opposite when I manage prs.
2. I see hard to explain and quantify problems with over configuration. The quality goes down, it loses track faster, it gets caught in loops. This is totally anecdotal, but I've seen it across a number of projects. My hypothesis is that is related to attention, specifically since these get added to the system prompt, they pull the distribution by constantly being attended to.
3. The models keep getting better. Similar to 2, sometime model gains are canceled out by previously necessary instructions. I hear the anthropic folks clear their claude.md every 30 days or so to alleviate this.
I wonder if this maintains the natural language capabilities which are what LLM's magic to me. There is a probably some middle ground, but not having to know what expressions, or idiomatic speech an LLM will understand is really powerful from a user experience point of view.
Very different places. Lived downtown for 15 years. Never been to Mercer island as it pretty much only residential and hard to access. Those can be good or bad depending on the person.
In Seattle there are a ton for town homes being built, split lot and connected, most without any hoa. When they need work I bet you’ll see a lot of neighbors suing neighbors to fix and replace shared elements. I would never move into shared living space that didn’t have some sort of hoa to manage any the shared space.
In Seattle there are a ton for town homes being built, split lot and connected, most without any hoa. When they need work I bet you’ll see a lot of neighbors suing neighbors to fix and replace shared elements. I would never move into shared living space that didn’t have some sort of hoa to manage any the shared space.
As someone living in a condo with and hoa I completely disagree. I get yearly budgets, bi-yearly building inspection reports. The hoa has a fiduciary responsibility to the building and is required by law not to let things decay. Without the hoa communal living wouldn’t work, and not everyone can have a single family home, especially in urban centers.
I think a lot of modern construction sucks. My condo was built in 1980 and I almost never hear my neighbors. It is also a fairly well designed building where both vertical and horizontal neighbors have aligned spaces for sleeping eating and living. Along with side shard walls being used for closets and bathrooms.
I do not. But I have neighbors who do. My wife also grew up in a condo. Every living situation has different trade offs. Maybe condos don’t work for some people. I think for most people, particularly young people they make a great first home.
I think people think they don’t like condos. I currently live in a condo I own and while not without challenges overall I like it. I think people have unrealistic pessimism for hoas. In Seattle there are a ton for town homes being built, split lot and connected, most without any hoa. When they need work I bet you’ll see a lot of neighbors suing neighbors to fix and replace shared elements. I would never move into shared living space that didn’t have some sort of hoa to manage any the shared space.
We use websockets and solve a lot of the state management problem called out here by keeping very little state on the server itself. The primary thing on server is a monotonically increasing integer we use to stamp messages, this gives us total order broadcast which we then build upon: https://en.m.wikipedia.org/wiki/Atomic_broadcast
Here are some code pointers if you want to take a look:
The deltamanger in the container-loader package is where we manage the websocket. It also hits storage to give the rest of the system a continuous, ordered stream of events:
The main server logic is in the Alfred and Deli lambdas. Alfred sits on the socket and dumps message into Kafka. Deli sits on the Kafka queue, stamps messages, the puts them on another queue for Alfred’s to broadcast:
https://github.com/microsoft/FluidFramework/tree/main/server...
I've had good luck filing complaints with the CFPB to get things fixed on my credit reports for some of the other credit reporting agencies: https://www.consumerfinance.gov/complaint/
To understand Net Neutrality you first need to have an understanding of how the internet works. The internet is basically composed of a bunch of different networks. These networks interconnect with each other (see Border Gateway Protocol (BGP) for more details).
There are basically three different kinds of networks:
1) Networks that primarily host content; azure.
2) Networks that primarily do transit; level three.
3) Consumer Terminating Networks; Comcast.
I admit this is a generalization, and some networks fit multiple of these categories, but lets not get too bogged down in the details here, as I don't think it's necessary to understand net neutrality.
Given these types of interconnected networks lets go through an example. Let say a consumer wants to watch Netflix. The consumers pays Comcast to access Comcast's network and get some bandwidth. Netflix pays their hosting network, and possibly a transit network for bandwidth. Comcast either directly, or through another networks connects to Netflix's hosting network. Neither Netflix or Comcast necessarily pays anything to these in-between networks. This works because these in-between networks trade about the same amount of traffic (see Peering). This is basically net neutrality; you pay Comcast for some bandwidth, Netflix pays their host and the content flows freely.
Without net neutrality Comcast can change you, Netflix, or both more to connect to each other. It can make Hulu free to connect to. It can make youtube inaccessible unless you get the super premium package. No one is just paying for bandwidth anymore, someone is paying for individual services on top of bandwidth. This allows Comcast to either pick the winner by making their services cheaper, or lets the company which pays Comcast the most be most available to their customers.
So what it really boils down to is, are you paying for bandwidth that you choose how to use, or not.
I've thought about this quite a bit, and think the tech community is going about this wrong. The public doesn't like being told it can't be done, but I can't explain why, because it's too complicated.
Even though we know it's impossible, we should disregard that and lay out common sense tenants that such system would require, even if it can't be feasibly built. We could then base our arguments on those tenants, and those are public fights I think could be won, because its things non technical people could understand.
For instance, one tenant could be any key escrow system must be open source. We can't base it off keeping the code secret, as then if the code is ever stolen or leaked, the whole system is compromised.
If you can win those arguments, and it just happens such a system can't be built due to the laws of mathematics, you then fall back to arguing which tenants you should break, and ideally breaking any of the tenants would be unpalatable.
1. I have many and sometimes contradictory workflows: exploration, prototyping, bug fixing debugging, feature work, pr management, etc. When I'm prototyping, I want reward hacking, I don't care about tests or lint's, and it's the exact opposite when I manage prs.
2. I see hard to explain and quantify problems with over configuration. The quality goes down, it loses track faster, it gets caught in loops. This is totally anecdotal, but I've seen it across a number of projects. My hypothesis is that is related to attention, specifically since these get added to the system prompt, they pull the distribution by constantly being attended to.
3. The models keep getting better. Similar to 2, sometime model gains are canceled out by previously necessary instructions. I hear the anthropic folks clear their claude.md every 30 days or so to alleviate this.