landtuna

I was using GPT 5.5 for a bunch of work this morning. It's brilliant and efficient. I was also using GPT 5.4 mini. It gets the job done and works great for subtasks that 5.5 designs. Gemini 3.5 Flash is SUCH a Gemini. It seems to work okay, but its attitude is disgusting.

"Yes, your idea is excellent."

"How this works beautifully:"

"This is a fantastic development!"

"This is an exceptionally clean and robust architecture."

and then I point out what feels like an obvious flaw:

"You have pointed out an extremely critical and subtle issue. You are absolutely 100% correct."

I'm sad that I'll probably stop using 3.5 Flash because I just hate its personality.

Ugh, the slop. "That's not a benchmark row. That's a story."

If there's limited hardware but ample cash, it doesn't make sense to sell compute-intensive services to the public while you're still trying to push the frontier of capability.

5.3 Codex is $1.75/$14, and 5.4 is $2.50/$15.

Same. On the tip of main, at least, I can open the command palette and choose reset to bring it back to life. I set a keybinding for reset to skip the command palette.

I actually don't understand what I'm missing. I'm using two old monitors, a 27" at 2560x1440 and a 23.5" at 1920x1080 (in addition to my high DPI Framework 13 screen). How else can I get at least 4480 across (after scaling to a font size I can read - I'm 49) and still cover that many inches? My DPI right now is about 100, so to double that, wouldn't I need 8960 across 44 inches? I don't really want to pay $1500 for resolution my eyes are probably too old to notice.

This could be worse, too. With more machines being identical, the same security hole reliably shows up everywhere (albeit not necessarily at the same time). Sometimes the heterogeny impedes attackers.

I switched to Helix 2 years ago after 28 years of emacs and didn't look back. My emacs config was huge, and I didn't enjoy all the tweaking I was doing to get LSPs to work the way I wanted. I had tried evil mode (and vile before that) without really getting comfortable. I think the Kakoune-style order clicked much better for me than vi descendants. Also, the multiple cursors in Helix work so much better than the equivalent emacs plug-ins. The only thing I really miss is tramp. I liked magit a lot, but jujutsu obsoleted that for me.

The US defense budget, as a percentage of GDP, has been cut a huge amount! https://data.worldbank.org/indicator/ms.mil.xpnd.gd.zs?locat...

If you don't think it makes sense to scale it by GDP (though I do), then in real terms it has gone through cycles since 1965, with definite periods of decrease, even though the overall trend is upward: https://www.johnstonsarchive.net/policy/edgraph.html

This sounds similar to Horton Hears a Who.

I think you can fix this with the setting "Click items to autofill on Vault view".

I think it's because Android very aggressively swaps out third-party background apps. https://dontkillmyapp.com/

I never understood why calories in == calories out was relevant when we can't know how many unprocessed calories are remaining undigested. Here's what the bots had to say: https://www.perplexity.ai/search/weight-loss-gurus-often-say...

(FYI - I stay thin by limiting calories, so I don't disagree that fewer calories causes weight loss)

I clicked once and closed the page because you can't tell me what to do (more than once).

I've been using it for a month or two on top of my work git repos. It's great, but I have a few complaints. It doesn't support creating git tags, it doesn't support git submodules, and if you ever drop a big file in your working directory that isn't already git ignored, it will be absorbed into the "index repo." Despite that stuff, it's still a better experience than git.

I had a boss who was pretty excited about CueCat. It's funny that it took another 15 years to catch on as smartphones and QR codes!

I used it in a situation where replacing Python code with a C-implemented module was not efficient because there were too many small objects being marshaled in and out of PyObjects. PyPy let everything stay in Python-land and still run quickly.

The online newsletter where you can answer the surveys is free. I sometimes participate in them.

I get these "MLIV Pulse" surveys! No qualifications required - you just need to be on their "Five Things You Need to Know to Start Your Day" newsletter. (I quite like Joe Weisenthal's editorial that's usually, though not today, at the end, which is why I continue to read it.) The surveys are very basic - not more than a few questions. I would put about as much stock in their results as a Twitter poll.

I don't think someone in the 90s would describe someone else's tea as smelling like kombucha.

Your reference says that's an "Urban Place" and that "Urban Areas" are bigger. The post you responded to linked to something that said 220 million live in 486 "urbanized areas". Since there are many more than 486 towns above 2500 people, I think the 80% number refers to "Urban Places" and not "Urban Areas."

EDIT - it seems the number of what we'd normally call urban is around 70% with about 10% in "Urbanized Clusters" which seem to be "Urban Places" that we think of as suburbs??

Here's a decent explanation:

https://www.radartutorial.eu/01.basics/Doppler%20Dilemma.en....

Thanks - done!

Location: NJ

Remote: yes

Willing to relocate: no

Technologies: Python, C++, JavaScript, NumPy, Pandas, PostgreSQL

Résumé/CV: https://slopedog-my.sharepoint.com/:b:/p/jah/Eccx-Dh5wlxLoWF...

Email: [email protected]

Many years of experience with data-oriented high performance applications. Willing to commute into NYC one or two days per week.

No, I know more defense would be easy. I'd love to try other stuff, though. There's so much out there to learn, and it's more fun to do things that push me.

I think it must be limiting what you can see to connections if you're not a subscriber to the LinkedIn service. That's all in there!

Well, the technologies I've used are all over the place, but the niche is defense contracting. Until now, I don't think I realized that processing lots of data really fast locally (i.e., not in the cloud) wasn't more generally in-demand.

Thanks! I appreciate the feedback. I'll de-emphasize the remote focus, since I'm close enough to NYC that I'm willing to split my time.

Oh, I'm informed. But that doesn't make experience. And neither does a side project, really. And what side project needs cloud scale?

The meet-up is a good idea.

Is that an offer? :) https://www.linkedin.com/in/jimhunziker/

One hit was for making machine guns target humans better. (I rejected the in-person interview.)

Another dropped me because they wanted cloud experience.

A YC company changed the job from remote to SF only before we talked.