TimPC

I personally think the line should be mostly output based. You should be able to train on any copyrighted work by having a single reader license (e.x. purchasing a book or e-book) for that work and no other special licenses. You shouldn't be able to download pirated works for training but you shouldn't need special licenses to train instead of read.

But if your model produces outputs that too closely match their inputs and a company can show it that is a copyright violation and you can be sued for it.

I think the reasons not to self-host are the difficulty of implementing any type of dynamic scaling. For instance when clicking on this article I get a result that their site is down. Instead of benefiting from the traffic their engagement got them, they are crashing.

One thing that's nice in LangGraph is the flexibility in what a node is. A node doesn't have to be an agent it can be any State manipulation function. So you can create nodes that exclusively do preprocessing or postprocessing and get your run loop as close as possible to just ainvoke on your graph repeatedly.

You can have a subclass of your Node class be an AgentNode class and then subclass that for each type of Agent and then when you declare your Graph object you pass in the data to instantiate the AgentNode with the type of data it needs. It is a bit weird that LangGraph doesn't have a default Node class but it sort of makes sense that they want you to write it in a way that makes sense for how you use it.

I do highly recommend abstracting your graph into Node and Edge classes (with appropriate subclasses) and being able to declare your graph in a constant that you can pass to a build_graph method. Getting as much code reuse as possible dramatically simplifies debugging graph issuses.

When I made this decision 20 years ago I landed on Software Engineer because people calling themselves that generally earned more than people calling themselves programmer, coder or developer.

It's debatable to what degree they are as most of the studies have massive confounders. Couples that stay together tend to be more affluent than divorced couples because of only needing to pay for and maintain one place of living instead of two. And poverty is a massive confounder in the studies, kids in poverty do a fair bit worse than kids in middle class lifestyles.

Location: Toronto, ON, CA Remote: Preferred Willing to relocate: Mostly No Technologies: Python, various ML frameworks and tools LinkedIn: https://www.linkedin.com/in/timcapes Email: capestim@gmail.com

Strong AI Engineer/Consultant/Leader with a fifteen year demonstrated track record in both big companies and start-ups. My ideal role is putting together an AI Strategy, AI Roadmap and then building an organization to execute against that roadmap. I am also open to interesting and challenging IC work in AI. I am generally looking for Principal Engineer or Director level roles. I am open to consulting or full time.

You probably don’t ban Facebook as a whole but if they fail to crack down on gambling links that violate advertising laws or allow gambling companies to advertise in spite of those laws they probably face heavy fines from regulators.

This is the same issue where poker companies used to advertise their play money sites and use the play money sites to link to separate real money sites. The loophole exists although it is certainly closeable.

I think in this case it's unlikely because I don't think the problems the tool solves correspond 1-1 with reduced staffing or other resources. The tool mostly seems to provide ongoing diagnosis at a level of detail the clinical team doesn't have regular bandwidth for (they might make one diagnosis of the patient per time they visit the patient rather than on an ongoing basis). It doesn't really reduce the amount of time staff can spend with patients. They can't get rid of doctor diagnosis entirely so they can't really reduce time per patient in any effective way.

Just another example of FIFA's absurd corruption. They know altitude is unfair and they allow increasing it unnecessarily to unreasonable heights.

I guess I’m pretty valuable then because my hit rate on AI projects I’ve lead is over triple the industry average.

They used to give a pretty good stock package but it lead to chaos in the company when the stock went down 10x and people were paid 50% stock.

You don't understand how tax brackets work. You never make less money by moving up a bracket. The bracket rate only applies to income within the bracket.

The red and blue agents are effectively unlimited sources of true and false examples so you can get far more efficient scale than you can by pre training with labelled inputs. It’s also far more targeted on correct/incorrect rather than a notion of answer quality which doesn’t directly get at hallucination vs reality.

I disagree they will become commodities because most of my use cases are more sensitive to accuracy than cost. We typically have usage volume that isn't absurd and for large enterprise customers our LLM budget is a rounding error. Meanwhile our product saves them many hours of a data engineer. If we can pay double to get a 10% performance boost we will do so gladly. You can already see this in LLM pricing where they have cheap models that deliver low performance a My bet is that 80% of profits will be made on the workloads that are sensitive to accuracy and workloads where running an LLM at all gets you most of the benefit will become very commoditized.

They are becoming like SAP though where initially a company buys one service and it soon finds itself buying every adjacent service from them. If they manage to do that successfully they will be quite profitable.

The Henry George Theorem also requires the public good to be local. Military spending for example is a public good that doesn't show up in land value increases.

Also in some cases (often with environmental spending) there is an external downward pressure on land value that is offset by an increase but you don't necessarily get an increase from the original neutral value if the pollution didn't occur.

100% LVT isn't 100% of land value it's 100% of land rents. If the bare land would rent for $2,000/month the tax is $2,000/month. You can't realistically raise the tax beyond this amount because you'd effectively be making land economically unhabitable outside of cases where the improvements were so large that you weren't significantly hindered by having some of the improvement rent absorbed.

At a 100% LVT the price of land is $0 (or more accurately equal to the price of the improvements on it). Above 100% LVT the price of land is negative so building any improvement immediately loses money and a whole host of other negative consequences.

The claim is that 100% of land rent in the US is smaller than income tax + sales tax + property tax + corporate tax. You can no longer replace every tax in the system with land rents unless you shrink the size of government.

This is no longer actually possible. LVT at 100% generates less revenue than all other methods of taxation combined across all levels of government. This was the original idea but government is now too large for it.

Assymetry makes a certain amount of sense. Employees don’t take $0 for a long time and generally aren’t having as large a pay cut as founders afterwards. Most of the founders I’ve worked with have had the seniority to justify the top salary in the company and have typically had pay at or near the bottom. Someone operating at that extreme getting to trade equity doesn’t necessarily mean that everyone should get to.

The problem with CO2 levels is that no one likes the solution not that we don't have one. I highly doubt adding AGI to the mix is going to magically make things better. If anything we'll just burn more CO2 providing all the compute resources it needs.

People want their suburban lifestyle with their red meat and their pick-up truck or SUV. They drive fuel inefficient vehicles long-distances to urban work environments and they seem to have very limited interest in changing that. People who like detached homes aren't suddenly affording the rare instances of that closer to their work. We burn lots of oil because we drive fuel inefficient vehicles long distances. This is a problem of changing human preferences which you just aren't going to solve with an AGI.

It actually does both. The increase in CO2 causes temperature increase which is well documented in the literature. But increase in temperature also causes the release of CO2 which further increases the temperature.

I think this is changing. Previously you had to train something massive off a huge data set. But now it's moving towards having a pipeline of pre-trained models that are trained on massive data sets and then smaller models that you train in house to tweak results from that pipeline. Any start-up should be able to get its hands on enough data to train a LoRA for example. There are good enough open source components to build a moat out of a good pipeline with one or two components in the pipeline trained in-house and the rest pretrained.

A story about broken hiring processes that I’m sure many of us can relate to.

True but the number of hours I work in SW + the number of hours I think about my job outside of work is still probably less than a typical doctor's working hours so I feel like we come out ahead anyways.

I'm at a director level and I routinely look at code and even more frequently look at product in a late stage start-up.

If the researcher makes a neutral contribution and someone else you could hire with the same resources would make a strongly positive contribution you're always going to favor the strongly positive contribution.

Temporary numbers are net not gross so if a tourist enters and leaves the country in a single year they count as 0 not 1.

A very common pattern is to write a patent and a research paper after the initial patent filing. When the patent office doesn't do their job you can still usually learn the invention from the paper. They key thing is companies wouldn't let employees publish without the patent part.