Breza

Networking. For an embarrassing amount of time, I knew very little about it other than some IP addresses looked different from others. In 100 hours, you can learn enough to be useful as a dev without going down rabbit holes.

Consider buying a split keyboard. It was a game changer for me. By itself, you'll learn to always put your hands on the correct sides of the keyboard.

If these tariffs continue for 30 years, it's hard to know if we'd still be the largest single market in the world.

I wonder if there will be a gap between (1) AVs being so common that driving for Uber doesn't make financial sense, and (2) AVs being able to operate in all driving conditions. I imagine people at a concert all fleeing for AV taxis after receiving alerts of unexpected fog coming.

I wonder if there should be a two-tiered structure to replace the traditional drivers license. In flying, you can get a Private Pilot License, or you can get a Sport Pilot Certificate, which is easier to get but has fewer privileges. It would be interesting to see a state replace a drivers license with the Sport level license, which would only let you drive a vehicle up to 6,000 pounds (Cadillac Escalade), have other restrictions, and have a higher insurance rate. Then the higher level license would require additional training (somewhere between current DL requirements and the monthlong requirement for a CDL) and let you drive the full limit of 26,000 pounds and you'd get discounted insurance.

Very well put. I also read books to learn concepts. I usually only turn to documentation to remember the nitty gritty details that are easy to forget. Consider loading the first 100 rows of a large CSV. About as basic as code can get. Yet as I move between languages and libraries, it's hard to remember all the different syntaxes, so I find myself looking up official documentation.

read.csv("large.csv", nrows = 100)

read_csv("large.csv", n_max = 100)

pd.read_csv("large.csv", nrows=100)

pl.read_csv("large.csv", n_rows=100)

CSV.read("large.csv", DataFrame; limit=100)

This is a massively important point. My dad took a programming course back in the day and had to use punchcards. High-level languages make code easier to understand and debug compared to gazing at a stack of "Do not fold, spindle or mutilate." LLMs are the opposite. If you only learn how to use LLMs, you are not going to be able to understand what's happening or why things aren't working.

I've found LLMs making inexplicable mistakes when writing statistical code, but the code will still run and return a confident result. If you can't look at the code, you won't see, e.g., your hourly time series being aggregated on a daily basis.

This is such a great analogy! Exactly how I feel when using AI tools. I have had some incredibly productive conversations about high-level design where I explain my goals and the approaches I'm considering. But then the actual code will have subtle bugs that are hard to find.

A great service. Commonly used by long distance hikers.

> get food stamps

Great advice all around, but I'll emphasize this part. SNAP is becoming a de facto indicator of eligibility for all sorts of programs, from reduced price transit passes to Medicaid, depending on the state.

Same here. I stopped using it partly for this reason.

I strongly disagree about Netflix. It came out when I was in high school without a car. Being able to get whatever DVD I wanted without having to bum a ride from my parents--and also never have to pay late fees--was a major game changer.

In this particular lotto, 4% of all possible numbers were bought in a typical period. So it's not truly riskless, but there was definitely a positive expected value.

Yeah, I imagine they save the list of all the tickets they've sold, get a hash of the values, and then generate the winning number.

Obligatory Seinfeld reference... https://en.wikipedia.org/wiki/The_Bottle_Deposit

A Washington Post columnist has been tracking the media coverage of Hitler's first 100 days. It's interesting how much people viewed him as bumbling or harmless or or or. https://www.instagram.com/petula_d/

Now I'm imagining someone mailing postcards to a dozen friends every month saying "We live in a totalitarian dictatorship!"

Once the postcards stop appearing, you know you're on the way to the message becoming true.

Let's look at a single example of Elon Musk's alleged crimes. He bought Twitter stock in 2022 and did not follow disclosure laws, which led to him getting an extra $150 million. It took the SEC more than two years to investigate him and decide to proceed with charges. That's more due process than most people get. https://www.sec.gov/enforcement-litigation/litigation-releas...

Congress needs to be responsible for restructuring the government. Even if you support what President Trump is doing, it'll all be reversed on day 1 of the next Democratic administration. Avoiding this kind of yoyoing every four years is one of the reasons the Founding Fathers made all proposed major changes to the government receive public debate and approval at least five times (House+Senate committee then House+Senate floor then president).

At work, developed our own suite of benchmarks. Every company with a serious investment in AI-powered platforms needs to do the same. Comparing our results to the Arena turns up some pleasant surprises, like DBRX hitting way above its weight for some reason.

I've experienced this problem but I haven't come across papers about it. For this context, it would be interesting to compare the accuracy of transcribing one page at a time to batches of n pages.

I suggest picking up Knuth's Concrete Mathematics. It helps you get used to the math in TAoCP as well as his writing style, which manages to be simultaneously rigorous, serious, and humorous.

GPT-2, GPT-3, GPT-4.

I completely agree! I'm currently using Gemini's "2.0 Flash Thinking Experimental with apps" model.

I used to agree before one of the "Sonnet" models overtook the best "Opus" model

I just got rejected by Gemini then Claude gave me the exact recipe.

Garmin packs an absurd amount of functionality into a watch

Same here, just replace Apple with Fitbit. I had probably six Fitbits over the years and made the jump to Garmin last year. It's an amazingly better experience.

I had a similar experience, but my employer was fortunate to have our lease expiring in late 2020, so we just took the dive into being all-remote. Since then, I've had direct reports from the UK to Oregon. I live in a major city with a lot of talent, but being able to increase our pool to a global search has been amazing, and I think we all get more done by working from home.

Great attitude! I recently built a tool for my wife that uses an LLM to automate a task. Is it production ready? Definitely not. But it saves her time even in its current state.