aero142

Software that had a data moat because it was hard to integrate with or migrate off of will have that moat disappear. A web site is a client now. Building data migration too for all of you competitors is easier now.

They will add flooded streets to the training simulation and this problem will go away. Eventually, the corner cases not in the training simulation will be so corner they basically never happen. Waymo can be incredibly successful without dealing with "surprise clown parade" or whatever.

Instagram is gone as well. Everything is fake in different ways. If the video isn't ai generated, then it's influencers acting out a scenario they think will get engagement. I realized that when it's a real video, there is a caption that says some scenario is happening, but there is nothing in the video that shows that is real. I think people are just reposting videos with different captions and testing out whatever invented scenario make the video have the most views.

If had to answer this question 2 years ago, I wouldn't have said software was a "don't see it's bad until later" category, with compilers and it needing to actually do something very specific. However, business slides are full of exacting facts and definitely never contains generic business speak masquerading as real insight /s.

This feels like telling a story after the fact to make it fit.

Yes. They probably are, but it's moderately hard to eat a healthy vegetarian diet if you look at what vegetarian athletes actually eat.

If llms are able to write better code with more declarative and local programming components and tailwind, then I could imagine a future where a new programming language is created to maximize llm success.

If a company can align it's business model with user goals, then it can work in the long run. Apple has somewhat aligned it's integrated hardware sales business model with user privacy. Google and Meta are advertising companies and capturing user data and attention will always drive the business.

I humbly submit '1ndex' and 'ind0x', or '1dx' and 'i0x'.

The selling point software that monitors social media to do support is about brand reputation. If people are saying bad things about your company on social media, then that is hurting your public image, so you should respond to those issues and make sure they get resolved. In practice, this means that the people responding are less likely to be outsourced because the the funding is more than just a support cost. In practice, this means you can often get better support through Twitter than email.

Are there any successful models that weren't trained with RLHF, or using a system with RLHF. I'm curious if this could be done without a fine tune step that would't meaningfully bias this.

This article does a poor job of representing induced demand so it's can't be trusted to critique it. Traffic and congestion is defined by (number of trips) * (distance traveled). So, if 10 people take 1 mile trips every day, or 5 people take 2 mile trips every day, the traffic congestion will be the same. Induced demand says that by building roads, you enabled people to buy a cheaper house further out of the city, and drive further, thus congestion stays the same. This is true for other trips as well. You might drive further to Costco to get cheaper groceries rather than to your neighborhood store if there is a fast road there.

I think there is a legitimate criticism of induced demand that it usually doesn't provide a tradeoff for when you have enough roads. 0 roads in all size cities isn't the answer. At some point a city has enough roads and should focus on mass transit or other transportation. I've never seen an induced demand argument attempt to define this threshold and why.

Adding a date of travel and checking the scheduled tail number would be much more accurate.

Doing REST microservices is incredibly slow because of the amount of work it is to agree on what a "clean" and "consistent" api looks like for each service. It's just such an endless well of trying to establish best practices without refactoring constantly.

It's worth it for your public API, but it's such a huge time sink for internal APIs.

When senior leadership is non-technical, they can't tell when non-customer visible changes are useful or they just have mediocre engineering. They also can't tell when their software stack is such a mess it's beyond fixing.

It’s a good version of scrum. There many more bad scrums.

I would guess they monitor tons of metrics for any change to search and any change that reduces ad revenue gets rolled back or reviewed.

I think the author's point is that tracing is a better implementation of both logs and metrics, and I think it's a valid point. * metrics are pre-aggregated into timeseries data, which makes cardinality expensive. You could also aggregate a value from a trace statement. * Logs are hand crafted and unique, and are usually improved by adding structured attributes. Structured attributes are better as traces because you can have execution context and well defined attributes that provide better detail.

Traces can be aggregated or sampled to provide all of the information available from logs, but in a more flexible way. * Certain traces can be retained at 100%. This is equivalent to logs. * Certain trace attributes can be converted to timeseries data. This is equivalent to metrics. * Certain traces can be sampled and/or queried with streaming infrastructure. This is a way to observe data with high cardinality without hitting the high cost.

Correct. I'm saying that private insurance is good at insuring uncorrelated risks, like car wrecks, but usually, only the federal government can enable insuring correlated risks. Suggesting that mandatory liability insurance can a good policy alternative to regulation is correct, but if you are regulating something large enough with correlated risks, only the federal government can every really provide insurance. Banks pay FDIC insurance, and bank failures are usually covered by this, but it's pretty clear that the pool of money to save banks is backstopped by the federal government in case of a cascading failure. Even for uncorrelated risk, if the payout event is bad enough, it can't be insured. For example, insurance on a nuclear plant is backstopped by the federal government, because no private insurer could ever really insure the worst case scenario.

There is an issue with correlated vs uncorrelated risks. This is the reason that only the federal government can offer FDIC insurance for banks and disaster insurance in Florida. When banks fail, they tend to all fail at the same time and there is no insurer that could hold enough reserves to actually pay out if they all fail together. Private insurance can cover fire insurance for homes, but is unlikely to have enough in reserves to pay if a category 5 hurricane floods half the homes in Florida.

I'm a bit late responding, but I'll answer anyway. I did read the article and believe I am disagreeing with the author's core point.

In the roofing example, the roof installer is passionate about roofing shingles, but the home owner doesn't have to care about it because he found a roofer who does care. If the roofer is good with people and passionate, but doesn't know the difference between types of shingles, you aren't going to get a good roof. In this case, understanding people is NOT more important that understanding shingles. The expertise of the roofer is what allows the homeowner to not have to care about shingle. But the expert roofer is the rare item. Every random person on the street wants to not have to care, but the expert roofer is rare. I bet business is booming.

For the team example, it's a straw-man that I don't want to conceded to the author. Sure, in some imaginary land, there is a motivated and skilled team that is force to use terrible technology but somehow overcomes this. This is not a useful description. In my experience, skilled teams are experts at choosing and understanding their tools to make technology. For skilled craftspeople, working with others is important and required, but so is being a skilled expert. The expert part is usually more rare than the people part.

The problem I have with the author is that they have picked two poor examples to support a misleading title. In my experience, companies are fully of people who say they understand people, but there is a skilled group with their doors closed making thing work. One day those people leave, and the people running around saying people are more important than technology are suddenly confused why things stopped getting done.

Do you have any examples of deploying to Kubernetes with something you consider real tools? I agree with what you are saying but have found very few real world companies using real languages for this.

Exactly. Good products and companies are abstractions that allow a homeowner to not have to learn about shingles. The roofer should definitely understand what shingles are going to provide rain and show protection in the customer's environment. The roofer doesn't have to know everything about the chemistry and material that is in the shingle though. That abstraction is provided by the shingle manufacturer. They need an expert that knows about adhesives and gravel and how many years of sun exposure they can handle.

I think this article is bad because it empowers a lot of people to dismiss expertise as unimportant rather than knowing where that expert is. I think the important thing for companies is to have clear understanding of what abstraction your product provides your customers, and which portion of creating that product are you the expert in. You should then delegate the portions that you don't want to be the expert in, and make sure you hire and retain the experts that DO know about that. "The most important thing is understanding the customer." is bad advice on its own. "The customer wants a flying car." Well, do you have an expert in making cars fly? If not then understanding the customer isn't that valuable.

I really appreciated the economist view of parenting as trade-offs and acceptable risk she presents. The way medical information is communicated to general public is very binary and once a particular practice is recommended, it is treated as an absolute must. I'm sure this makes sense for communicating to the general public, but I am very capable of understanding risk and probability and making a weighted choice for what makes sense.

Also, the baby-industrial complex loves this binary view because it allows them to sell products that are "better".

It seems the improvement in the models is the size of the training set. Hence the "Large" in large language models. I am assuming that it is better at programming because there a is a ton of training data available on Github, etc. There isn't a similar set of data on solving a variety of physics problems.

I've been asking friends in non-programming engineering fields how ChatGPT does in their area of expertise, and I believe programming is the area that ChatGPT is the most accurate. Finding solution to general engineering problems seems blatantly wrong in almost all cases, whereas in programming, it seems to be able to generate mostly correct code for simple, boiler-plate like tasks.

How do these companies ensure that the hard drives get cleared of company data? I would expect most companies handle information that is legally restricted and have an obligation to ensure employees don't keep a stray copy of customer information.

I think it will be viewed as a common vice like alcohol and gambling. I wonder if we will see something like a temperance movement.

Perhaps cognitive behavior therapy could help with your anxiety.

I've been part of an on call rotation at my current job and previous job. Both had the same rules because I and other engineers insisted on it.

* You are only on-call for systems you can directly change.

* The person who is on call has full flexibility to work on anything that improves the life of the on-call person for that time.

* Anything that wakes people up in the middle of the night gets prioritized above new feature work.

* There is a pager set of monitors, and a message only set of monitors. If a page goes off, and there is nothing the on-call engineer can do about it, it gets moved to the message only channel or removed, because it is a bad monitor.

I discussed this list of rules when I interviewed and the job description included being on-call. It wasn't a negotiation. If those aren't followed, I remove the monitors. I'm sorry you had a terrible work environment, but I encourage everyone to have professional standards. I hope you find a place where you can.

Yes. It's like the results when people realized you could have a classifier trained to match a person's face, reversed to generate a new face based on the classifier. There are a few extra steps, but the web is just recipe sites and product reviews that look like what the google ranking algorithms idealized site looks like.