xmonkee

Well, we use flash models extensively (both 2.5 and 3.1) and I cannot overstate this, google cannot fucking serve them without 503s 70% of the time on most days

I think it’s pure economics. Flash models are OP for the price, leads to too much demand, google cannot serve it. This is likely expensive to reduce load and hey, if it still makes money just keep the margin.

One persistent issue I keep having is preview environments for this kind of stuff. I have the full setup, migrations, database seeding, etc. But having it run off a PR is still kind of a mess with spinning up 2 services, databases, redis etc. Do you guys run into this problem?

How do you feel about it now? Like holistically.

It’s probably an offshoot of making Claude more and more suitable for code/cowork.

Technical question: I run my own Pi hole dns server. Is it possible to just add an A record for it? What are you guys doing to get around this?

I am a manager. I am the head of engineering at my company. I still don't understand what is going on.

IDK what everyone is doing anymore. Just why do you need 10 parallel agents doing things. How is this even a possible workflow for a person.

Yeah, I gave it a very fair shot. Got pretty good at it too. But I’m back to Vim too.

Things I loved: no plugins. Native LSP integration. The pickers are a lot faster and nicer than what I can get in neovim after absolutely atrocious configs.

Things I liked: w and e selects the word. I kinda got used to that and miss it in vim now.

Things I loathed: there is no clear mental model of what will get selected on a motion. Something like selecting a paragraph (V } in vim) is replaced with a (gf) which doesn’t ever do exactly what I want.

Overall, the annoyances outweighed the benefits. I wish evil-helix all the luck. I would use it but it kinda sucks on Mac rn since you have to whitelist every library used.

[flagged]

This is why everyone should have at least a basic education in materialist analysis. Our material relations dictate the structure of society, not the other way around. China is a manufacturing economy, that's why it's run be engineers, and America is built on exploiting the productive capacity of the rest of the world, so of course there's lots of lawyers.

I have a lot of mechanical keyboards, and honestly I prefer plastic over aluminum. I don't really understand the allure of all that heavy metal. The new 8BitDo one is a good example of a "premium" mechanical keyboard with a plastic case and it sounds and feels amazing.

I was actually shocked he included those tweets as if they were incredibly benign. I now believe this is an intentional move on his part. He knows the tweets are crazy incendiary, he just wants to filter out the audience early. This trend of catering to far right fan boys while maintaining plausible deniability is happening everywhere.

Same. And love those brutality mods.

Do you not understand that that is also just more finely processed american propaganda? Captain America represents the "real" America, one that will not compromised by the corruption in our system. I think after the 90s it was just too much to show all Americans as the good guys, so now we do this dance where we are still essentially good, and we just have to stand strong against the bad guys within.

you woke up before dawn with your companions to go diving in the freezing cold ocean, in hopes of putting some mussels on your family's table. But suddenly, you die. A man you have never met and whose presence you did not know about has shot you with his rifle. His companions stab your lungs so that your body will sink to the bottom of the sea. Your family will likely never know what happened to you.

Man, fuck these people. Meanwhile hollywood will churn out another hundred films about how Captain America would never let something like this happen because murdering innocents is not a line America would ever cross.

Does jj effectively support stacked diffs on github? I've been paying through the nose for Graphite lately because it does, and the feeling I get is I'm just paying for their nice cli tool. `gt create` and `gt sync` is all I need. I don't really fully agree that their interface is entirely necessary or worth the $3500 per year it costs right now for my team.

This is just very clever marketing for what is obviously just a cost saving measure. Why say we are implementing a way to cut off useless idiots from burning up our GPUs when you can throw out some mumbo jumbo that will get AI cultists foaming at the mouth.

A lot of the CEOs in the US currently come from this uni. It's called IIT, in case you're curious. There's also IIMs which are based on the same model but for business studies. Also no legacy admissions. Also produce a metric ton of CEOs and execs. There are none more elite than these, at least in India. The politicians in my country are rarely highly educated, otoh.

This is such a bizarre and gross take. Yes our history is a history of class struggle. But history does progress. For thousands of years we were supposed to be property of kings so shall we mean revert to that?

I went to an “elite” public university in India which has a sub 1% acceptance rate. It was mostly extremely smart and driven middle class kids from incredibly diverse social backgrounds. Everyone had the time of their lives. And almost everyone now (20 years later) is doing incredibly well in life. They are doing startups, public policy, research, tech leadership etc. There is zero legacy admissions. And yes there is a network effect, of course. You can count on the friends you made at uni, but not because they inherited the influence. You don’t have to lick boots to have a good life.

A standard as in there is a cottage industry of tools and websites built around it now, like this one.

God I really abhor jq and it seems it's becoming a standard. I dislike it cause I'm too dumb to correctly dredge up it's incantations, and once a year I have to go reading their arcane docs. I suppose it's another fertile ground for LLM use.

I guess this serves as an ad for Lumafield's x-ray machines. I have no use for one but those scans are stunning.

I really really like it. I honestly think they should wrap it in a GUI with some conveniences like a mouse based file browser and it can compete well with vscode.

I'm an LLM noob, but how feasible it is to make a research agent that can not only download articles, but read and reference them in it's process?

I don’t know what sort of founder you are or what you are looking for. I have been a founder and am currently an exec at an early stage startup. There is close to zero value in talking to other early stage founders most of the time. I never cared about the “startup scene” because it’s complete irrelevant to my company. Can you get me customers or engineers? That’s all I cared about. I am curious what you hope to get out of these convos.

Dude... thanks for the laugh.

nice, V lang. How come I don't hear about this as much as zyg/odin/jai etc?

The User object in your example is used to parse the data. Its the “somehow” part of your question. There is no way to go from a type to data in typescript (there is no runtime awareness of types whatsoever) so zod solves this by you writing the zod object and then deriving the type from it. Basically you only have to weite one thing to get the parser and the type.

We are using deno to allow our customers to run custom code with imports. This is only made possible with http imports and I for one am very happy it exists.

Does anyone know the impact of the prompt size in terms of throughput? If I'm only generating 10 tokens, does it matter if my initial prompt is 10 tokens or 8000 tokens? How much does it matter?