Breza

I write most of my code by hand. I already manage a team, I don't want to spend the rest of my time juggling different agents.

I agree. I feel like we're heading back to the PMBOK days of writing out all the details of your project before you wrote any code.

I agree with you. I use LLMs for a lot of different things. But increasingly, I find myself firing up GPT-0SS-20b on my laptop for smaller tasks. And I know there are better models I just haven't tried yet.

In my experience in the corporate world, figuring out what everybody wants takes a lot more effort than actually doing it. Ten years ago, I could take a detailed request for a small feature, with pseudocode and specific implementation guidelines, send it to a junior dev, and get a result within a week. Now I can get a working result in minutes. But the real delay was usually the series of stakeholder meetings, emails, and IMs necessary to get to that point.

Yeah, just imagine if one of their two biggest customers were to find a way to offload even 10% of their calls to a self-hosted model. That would cost Anthropic billions.

I think of it a little differently. LLMs are clearly poised to have a major impact. And Anthropic and OpenAI are at the forefront of that innovation. But if you look at other technologies that have have huge impacts (trains, semiconductors, you name it) the companies that start off in the lead are rarely the dominant players 20 years later.

Today, Nvidia is bigger than Shockley Semiconductor (founded 1955) and Rivian is bigger than Studebaker (founded 1852). Being first doesn't mean you'll stick around.

I've had a similar experience. Grok is the most confident of any model I use. Unfortunately, it is also the most often incorrect.

I've seen a lot of his interviews and he often makes two arguments: 1. LLMs are not fit for purpose and aren't nearly as useful as people say. 2. It is literally impossible for OpenAI and Anthropic to generate enough profits in the next decade for them to fulfill all of their obligations and return money to investors.

I agree that the first one is sometimes iffy, but even if you think LLMs will replace x% of white collar work by 2030, the second point remains.

Wild how LLMs both weaken software security and supercharge the attackers at the same time

Once WhistlingDiesel posts a stress test of the thrusters, I'll believe they're legit

Last year I went to a steam show and had to share the street with giant steam tractors for the first time. That law suddenly made sense. The visibility and stopping distance aren't great.

Some decisions are important to document. Some are utterly irrelevant. Knowing the difference is a skill.

On one project, we decided we didn't need mobile compatibility. That simplified things greatly.

You're going to want to document that somewhere. And if you also have a stakeholder register, you can link the decision to the person who made it.

There are some things that can't be derived from the code and old school PMP documents are suddenly useful again.

Who are your stakeholders? How are you going to communicate with them? What kinds of approvals do you need to do certain things? Etc. Human devs have long struggled with this stuff and LLMs aren't at better.

I still use R almost every day. It's an amazing language for answering complex questions using data. But yeah, not a great choice for deploying a production system.

I also found myself reading the manual all the time.

I have a toy project I use for learning different tools. Given a CSV containing every MLB game ever played, calculate the win percentage for each team during its first season.

Pandas makes that kind of simple thing painfully unintuitive. Tools like Polars and dplyr and even SQL require code that's far closer to how people think.

Worth reminding readers that pandas stands for "PANel DAta" so it's not shocking it's still useful for non standard data

There are a few packages that still expect pandas inputs and return pandas outputs, so I just convert my Polars dataframes before and after passing them to such a function.

I almost never use pandas directly anymore. Polars is just too good.

This is why my first cloud provider when I was learning VMs was Verda. They let you prepay and everything would shut down the second you ran out of money.

It was a necessary feature when I was learning. I was amazed when I moved up to the big guys and this wasn't an easy option.

I completely agree with you. If I'm running a big VM for model training, I want to be absolutely sure that I can't accidentally leave it on over the weekend. If I'm a fast growing startup, maybe I'll prefer alerts so I can optimize a slow piece of my stack before getting a surprise bill.

I've read all three of those in the past three years and I loved them all for wildly different reasons.

If you want a classic novel that's enjoyable to read and has a moral point to make, I'd add War and Peace. The story ends and then the author has a whole afterward dedicated to "In case you missed the point, let me spell it out for you."

I agree that they are historically important, but if you want to query old thinking in 2070, you'd probably be better off having a modern model analyze archive.org. If that ever goes down, we're sunk.

In my experience as the head of data science in a corporation, everybody uses Windows on their local machines and Linux remotely.

When my daughter turned five and received her final book in the mail, it came with a letter that still makes me tear up.

At work I use a tech stack written years and years ago in .Net that runs on a server we own in a colo. We use a CDN for some latency-sensitive serving, but almost everything is self-hosted. It's amazing how cheap it is for us to run a reasonably sizeable application.

It wouldn't necessarily lower prices. Waymo could compete on quality instead of price. If you offer a better and safer service than the competition (human and AV) why lower prices at all? There's a reason Waymo often charges more than Uber+tip.

I agree with you. Look at railroads, cars, computers, and other technologies that really did end up being world-changing. How many of the original companies are still around?

Sometimes I feel like OpenAI will end up like Fairchild Semiconductor: an early innovator that quickly became an historical footnote.

Yup, every time I'm on a city bus they have a screen that cycles through the different cameras. They really have every angle covered. And some of our buses even have traffic enforcement cameras that issue tickets to cars.

> scheduling constraint

Honestly I think this is the only real advantage of a self-driving bus. Driving a bus is hard work and it doesn't pay well. Getting drivers to cover all the routes every day is an ongoing challenge.

I would honestly be nervous to ride a self-driving bus much of the time. Drivers limit the level of shenanigans that can happen. The subway has emergency intercoms and staff at every single station, but a bus could take you miles away from the nearest transit cop.

Bus drivers don't make much money, at least not here in the United States. I was shocked when I learned how low the average salary is, considering you might be tasked with navigating a massive vehicle carrying dozens of people through narrow streets during rush hour. And I'm not even considering the customer service aspects and public safety aspects of what they do.

All that to say, we are far from the point where a self-driving bus is cheaper than buying a regular bus and paying a driver.

Done. I grew up in suburban Michigan. I moved to Washington DC for college 20 years ago and I'm still here. When looking to buy a house, we only looked at options with public transit access.

I know it's not for everyone, but if you really do value being able to go out without driving, it's a great lifestyle.