castillobuiles Profile Banner
Sebastian Castillo Profile
Sebastian Castillo

@castillobuiles

Followers
481
Following
4K
Statuses
17K

Im a curious, a passionate dreamer, a perfectionist human being with PGP key id 29FB71DF. Currently managing director at Barista Ventures

Medellín, Colombia
Joined January 2009
Don't wanna be here? Send us removal request.
@castillobuiles
Sebastian Castillo
4 years
Programming is often more fun than the alternative uses of my time.
0
2
6
@castillobuiles
Sebastian Castillo
4 hours
@olorjuelab Que chambonada 😂
0
0
0
@castillobuiles
Sebastian Castillo
5 hours
@Dopeytree @slow_developer He said 25% of code generated. He did not said they are more effective.
0
0
1
@castillobuiles
Sebastian Castillo
6 hours
@theo shape from 37 signals
0
0
0
@castillobuiles
Sebastian Castillo
20 hours
@sergio_fajardo Deteriorado 😂 Si Sergio no mandara a hacer sus discursos a otros se le entendería más. Proponga algo en lugar de estar vacuamente criticando. oportunista y descuidado. Por eso ha perdido cada elección por no proponer nada en lo absoluto.
Tweet media one
1
1
3
@castillobuiles
Sebastian Castillo
21 hours
@airkatakana “Performance collapses as sequence length increases”
@DimitrisPapail
Dimitris Papailiopoulos
23 hours
Even for simple algorithmic tasks like integer addition, performance collapses as sequence length increases. The only way so far that overcomes this relies on heavily optimizing positional encoding and data format. (Figure from: Cho et al.,
Tweet media one
1
0
1
@castillobuiles
Sebastian Castillo
21 hours
@VvanillaG0rilla @airkatakana Or just check the facts TODAY:
@DimitrisPapail
Dimitris Papailiopoulos
23 hours
Even for simple algorithmic tasks like integer addition, performance collapses as sequence length increases. The only way so far that overcomes this relies on heavily optimizing positional encoding and data format. (Figure from: Cho et al.,
Tweet media one
0
0
1
@castillobuiles
Sebastian Castillo
1 day
@BenjaminDEKR @burkov You are free from talking badly about your employers product. They are free to ask you not to or to remove a bad pr post. You are free to deny They are free to cancel your contract You are free to resign to make it look better
1
0
3
@castillobuiles
Sebastian Castillo
1 day
@elonmusk That is your vision? That is a horrible place to live
0
0
0
@castillobuiles
Sebastian Castillo
1 day
@slow_developer Higher MMLU accuracy is not “being smarter”
0
0
0
@castillobuiles
Sebastian Castillo
2 days
@digitronex @kepano exactly that and automation could be as a simple as a gh action plugin proposals can be PRs with signed commits
0
0
0
@castillobuiles
Sebastian Castillo
2 days
@DimitrisPapail Why should we be asking that? The reason is very clear and not surprising at all
0
0
0
@castillobuiles
Sebastian Castillo
2 days
@noxlumosnox_ @NTFabiano @SciReports No you should not do that. That is not what the study is saying. You should load slowly for weeks. That was a single dose study. We are not sure about safety. Studies are not to replicate at home blindly.
0
0
0
@castillobuiles
Sebastian Castillo
3 days
@sama Well there is something Sam is first timer: Stealing a non-profit in public for free.
0
0
0
@castillobuiles
Sebastian Castillo
3 days
@unclebobmartin @LuxFella You can assert that business rule at any level not necessarily on the db. If it’s a good idea or not depends on the domain constraints. Elon trying to be a dev-bro is cute.
0
0
2
@castillobuiles
Sebastian Castillo
4 days
0
0
0
@castillobuiles
Sebastian Castillo
4 days
@myockey @airkatakana Which study?
0
0
2
@castillobuiles
Sebastian Castillo
4 days
@myockey @airkatakana Well check this about context size limitations and potential solutions: And abl it arc AGI it’s not a research result. It’s well known information go check leader board on arc website. Best solution is search. Not naive context augmentation.
1
0
0
@castillobuiles
Sebastian Castillo
4 days
@myockey @airkatakana The determinant factor to improve score in arc AGI has never been token count
1
0
0