David Gomes
181 posts

David Gomes
@davidgomes
Working at @cursor_ai (previously @neondatabase and @singlestoredb)
Madrid 🇪🇸 Katılım Şubat 2012
400 Takip Edilen4.4K Takipçiler

@NickADobos The glass interface is impressive. Real question though: does it maintain context better than cursor's native mode? Tested both last week, glass felt faster but memory seemed shallower.
English

Cursor is now the codex app. Neat.
Cursor@cursor_ai
We're also sharing an early alpha of our new interface. cursor.com/glass
English

@davidgomes @midego1 @cursor_ai If your taking feedback, can you stop disable my own key randomly. Not just on every window open, but after 7 hours or so it resets to off
English
David Gomes retweetledi
David Gomes retweetledi

I personally find this quite inspiring — Cursor is much more sophisticated than what most companies can do today, but compounding your learnings into a model that excels at your use case is the ultimate way to build an AI product.
Cursor@cursor_ai
Composer 2 is now available in Cursor.
English

@midego1 @cursor_ai That's good feedback! I personally think most people will prefer Fast. It's a great tradeoff.
English

@cursor_ai @davidgomes But why you make fast (3x more expensive) the default? This feels like a criminal move.

English

@iserflott @austinnickpiel I have the same question and I work there 😅
English

@austinnickpiel What’s that ui ? In GitHub it doesn’t say anything when it doesn’t find anything, right ?
English

Justo estuve hablando de esto con @Martinho_cm hace unos dias, elle tiene una muy buena teoria sobre eso.
Si tus padres tienen un pequeno negocio, como un cafe o lo que sea, es muy mas probable que tu tambien tengas las ganas de crear algo tuyo. Creo que es un poco eso que si pasa en Catalonia.
Por eso el emprendimiento es contagioso. Justo se pasa exactamente lo mismo en Portugal – el norte tiene mucho mas micro-empresas y gente creando sus cosas.
Español

@YifanBTH @cursor_ai The problem with public benchmarks is that they can be extremely overfitted on. They’re in the pretraining data and so I don’t trust them at all.
English

if @cursor_ai would just run their model against existing common benchmarks like swe-bench-verified, it would shut us right up.
given cursor's position, they don't even need the absolute best model. they just need honest benchmarks the community can actually understand to rally support. inventing new charts doesn't help
Cursor@cursor_ai
We trained Composer to self-summarize through RL instead of a prompt. This reduces the error from compaction by 50% and allows Composer to succeed on challenging coding tasks requiring hundreds of actions.
English

it literally writes code that's run inside an isolate where the state is exposed

muzz khan@muzzdotdev
@threepointone how does the agent interact with the filesystem?
English

@ChezluiCJar @nstlopez That will help a lot. But I still have enough local agents always running 😂
English

@davidgomes @nstlopez You mention recently that you were using more cloud agents... Maybe that helps
English

So, @nstlopez recommended me this powerbank on Saturday and I got it on Monday. It's so much better than any other one I've ever owned, highly highly recommend buying it.
It charges my 16'' MacBook, gives me an estimate for when it'll be charged, and info on the screen about what it is charging at any given point in time (it can charge up to 4 things at once). Sounds gimmicky, but I find it all very useful.


English

@davidgomes @nstlopez This is a great product! I paid for one for $36 brand new, how much did you get it?
English

@ChezluiCJar @nstlopez My workflow wipes out my laptop's battery very quickly :( I can do like 3 hours of work.
English

@davidgomes @nstlopez How long does your Mac run on battery? Doesn't weight similar to the charger?
Use case?
English

@jacobmparis That makes sense
I’ve never understood embeddings, so thanks for explaining. So basically it precalculates the vectors for any given text so the LLM has an easier time understanding the text in its mathematical language?
English








