bluelightning2k

Is that even true though? As per another comment Gemini CLI and OpenAI Codex have more stars, and they're both much larger companies.

Then Replit, V0, Lovable are all extremely popular.

I don't have a horse in this race just find it interesting. Your Coke vs Pepsi analogy is a good one though.

Realistically - LLMs don't "delete the database". I find it quite unlikely that it proposed drop all out of nowhere. I wonder if what actually happened was a schema migration with an ORM? Prisma is still a pretty common choice, and FREQUENTLY migrations propose/require either a very nuanced path or a reset.

The second theory is an unbounded or inadequately bounded delete statement - essentially deleteMany on a single table.

From a more technical org I'd be interested in a write-up, but my intuition says one of those two paths to deleting technically a single table.

Analogy: wow this Tesla full self driving mode is legit, the car really can drive itself! I'm all in, I've fired my driver and deleted Uber.

OK the car scraped my other car. I made it apologize and PROMISE not to do it again.

OH NO IT HITS THINGS AND PEOPLE EVEN WHEN IT SAID IT WAS GOING TO BE GOOD.

I'd guess it's something like Operator which works on your PC not a remote virtual one.

Is this that machines rising up thing I've heard so much about?

(Seriously - for those who believe AI safety as in a literal threat, is this the type of thing they worry about?)

"I have an idea for a startup. it's like AI but BETTER. Wouldn't you like better AI? Money please!"

SUCH a good comment - made me laugh

People with no product, no specific insight, and no published plan raising $$ trillions, due to elite background.

(Sure -- that background is somewhat self earned.)

This is so weird to me.

Yes, she was there at OpenAI and by all accounts a super respected non researcher.

Clearly she's incredible, her reputation proceeds her etc. but is operational strength (being a good exec) really what gets to AGI?

Feels almost like cargo-culting.

As someone who stuck with it - it's good. And it's not actually just a thin wrapper. Their tab model is really, really good now (I think that was Cursor's big advantage for a long time). And they also price very fairly (read: heavily subsidized, minus Claude).

Reading this made me wish I hadn't named my son Windsurf

It's weird being a Windsurf fan on HN. You get the feeling that you're in the minority or doing it wrong because the dialog is dominated by Cursor and (more recently) Claude Code.

It's also been a lot of random stuff recently with their 3 separate Ross and Rachel acquisition storylines.

Some takeaways:

1. Devin/Cognition definitely have a legit AI dev agent now

2. It's crazy what Google passed on. The fact that it was worth it to them without the traditional best assets is wild. Guess that's what happens when you play on ultra hard mode with an infinite money glitch.

3. I am worried/pre-emptively sad that Windsurf will likely go away or get nerfed, more expensive etc.

As an actual user of Windsurf model, I don't think "tried" is fair. I sometimes use it. It's not as smart as Gemini but it iterates quicker and is very well aligned with their tool calls

It's a staggeringly bad deal. It's a hugely expensive task where unless you are the literal best in the world, you would never even see any usage. And even for those who are BOTH best and well known they have to be willing to lose billions on repeat with no end in sight.

It's very very rare to have winner takes all to such an extreme degree as code llm models

Windsurf has developed their own fron tier model. It's pretty good. It's not sota but it's very well aligned with their tool call formatting etc.

isn't using software like this deeply illegal? or is that a legislative blind-spot? seems like this database should be sent to the FBI and someone can make a career out of prosecuting

Interestingly I think I would over-think this. The interviewer is assuming a single server, running in a VPS type environment. There's also no notion of state persistence/timeout/recovery etc. I think I'd immediately have started factoring those things in.

ALSO while JavaScript is a single threaded environment, the while solution would still basically work due to the scheduler (at least if you yield, await sleep, etc.)

I can top this. The Erudite zone was literally built around a massive chasm. Early on in the game I unlocked "levitate" - and naturally levitated down into the big chasm.

A little way down a loading screen hits for a zone called "The Hole"; a high level raid zone. My levitate was removed by the loading screen, and retrieving my body would require a team of high level players - thus lose (meaning ALL my gear and inventory was permanently lost, and a heavy XP penalty).

I don't think experiences like these are as positive as your nostalgia has led you to believe.

I don't think this is a security flaw.

That's like saying it's a security flaw in the Chrome store that you could enable dev mode, copy the malware and run it that way.

I read through the pricing page and it's super confusing. Like I can understand it but it just seems very awkward.

I much prefer the price, and pricing model, of Windsurf (although Claude 4 not being an option is odd)

This is impressive. But it just strikes me as optimizing the wrong thing. A CTF shouldn't be about submission ops. Surely it would be better for everyone if all teams who send the flag within a submission window share the prize

I can't say I've ever wanted to transcribe code from an image. That seems super niche.

Perhaps the specific idea is to harvest coding textbooks as training data for LLMs?

Two takes here. Cynical and optimistic.

Cynical take: describing yourself as a full stack AI IDE company sounds very invest-able in a "what if they're right" kind of way. They could plausibly ask for higher valuations, etc.

Optimistic take: fine tuning a model for their use-case (incomplete code snippets with a very specific data model of context) should work. Or even has from their claims. It certainly sounds plausible that fine-tuning a frontier model would make it better for their needs. Whether it's reasonable to go beyond fine-tuning and consider pre-training etc. I don't know. If I remember correctly they were a model company before Windsurf, so they have the skillset.

Bonus take: doesn't this mean they're basically training on large-scale gathered user data?

I don't like or agree with this take. You're basically saying - "something good exists, so why try to improve upon it".

Their stated goal is to improve on the frontier models. It's ambitious, but on the other hand they were a model company before they were an IDE company (IIRC) and they have a lot of data, and the scope is to make a model which is specialized for their specific case.

At the very least I would expect they would succeed in specializing a fronteir model for their use-case by feeding their pipeline of data (whether they should have that data to begin with is another question).

The blog post doesn't say much about the model itself, but there's a few candidates to fine tune from.

potato + microwave = potato tree

Interesting choice to have your only demo video be testing a CLI. Unless that's literally the use-case it's for?

Turns out the non profit structure wasn't very profitable

I like deno's security model where you have to enable things. But I think it made a HUGE mistake to the point of being useless.

Realistically, we need to grant access to env variables, network, etc. So those flags are functionally useless as always enabled.

The issue we face is making sure our dependencies don't do something nefarious. Demo doesn't solve that. I would really, really like to be able to specify these flags at the package level.

Couple things:

First, perf you mention is from fundamentally CPU intensive workloads, not a sparse async environment like a typical server.

Second (far more important) the reason we do this is type safety. We don't want types to break at the handover or to maintain them twice. We eliminate a huge category of issues and mental overhead.

This maintainability and simplicity is worth more (to me at least) than perf.

As a bonus there's also directly shared utils, classes etc. which again can be reused across client/server but more importantly stay in sync.

Issue with deno deploy specifically: they have Deno subhosting but the model is you create an end user script, push it, then can run it.

There's no concept of "run this once" or eval. So it's a very heavy, deploy per run, model of iteration which is not suitable for the main use case of giving end users an editor to iterate with