Si por casualidad alguien necesita a alguien con amplia experiencia en sistemas, cloud, kubernetes.... Y alguna otra cosilla por aquí estoy :)
11.01.2025 08:41
👍 1
🔁 1
💬 0
📌 0
Si por casualidad alguien necesita a alguien con amplia experiencia en sistemas, cloud, kubernetes.... Y alguna otra cosilla por aquí estoy :)
This is a hard read. On depression, drugs, and suicide from Felix Hill, a DeepMind researcher, before taking his own life.
docs.google.com/document/d/1...
Still from the movie "Upstream Color". A woman and a man looking at each other, clear dusk sky in the background, silhouettes of birds in the sky.
I love sci-fi, and I read and watch a lot of it.
Made a list of some good, lesser-known science fiction movies → rakhim.exotext.com/lesser-known...
Shorter version in the thread below 🧵
I liked this post: acalustra.com/less-11-meet... (I found the blog via @antonmry.bsky.social)
Oops
I did. I agree with Cambridge's warning.
I knew my bug-fixing deployments were "magical thinking"...
I don't think young people understand how mind-blowing things like this would look to college-me (back in the 90s): www.youtube.com/@CMUDatabase...
Top quality education for free.
Uh oh...
Companies die from a lack of cash, which in your average SaaS company means a lack of product market fit.
I need an emoji for “skeptical interest”
Primera vez que oigo el nombre
For anyone interested in fine-tuning or aligning LLMs, I’m running this free and open course called smol course. It’s not a big deal, it’s just smol.
🧵>>
you: why'd it take so long to remake your website
me: needed a physics engine for my text
you: ???
me:
Dispensario antituberculosis Casa Sol na rúa do Orzán na Coruña
Un día coma hoxe, o 2 de decembro de 1906, abría as súas portas na Coruña o Dispensario Antituberculoso, situado na emblemática Casa do Sol, fronte ao Paseo do Orzán. Un pequeno edificio que agocha unha gran historia! Segue lendo👇
#ACoruña #PedroMariño #HistoriadaCoruña
7 databases in 7 weeks
cool idea :)
matt.blwt.io/post/7-datab...
It’s an interesting ranking, but most of Spain’s score apparently comes from having a great Internet speed and some regulations.
Speaking of this, the original MapReduce paper contains what I imagine are the first versions of Borg, Jupiter, and Colossus.
Lo que no sabemos nos cuesta demasiado
El recibo de la opacidad lo pagamos juntos. 💸
Cada contrato oculto, cada euro malgastado, cada acción en la sombra engordan una factura que aceptamos sin rechistar. Pero no tiene por qué ser así.
Podemos investigar más. Informarnos mejor. Exigir cuentas. Incluso aliarnos para cambiar esto.
Using it for the first time.
I just found out that the + button to the left of the language selector allows you to create threads 🤦.
Reviewing BigQuery technical info, I see that Borg orchestrates the communitcation between Dremel, Jupiter, and Colossus.
But what about Galactus???
www.youtube.com/watch?v=y8On...
Damn this is good
Question from Leo on Twitter: It doesn’t interpolate, does it? If I ask “What color is a Gropy?”, and we had 100 labellers say it’s blue and 100 labellers say it’s yellow, it’s going to randomly say blue or yellow - but never “It’s a debated question, some say blue, some say yellow”. Right? Answer from Andrej: Excellent question and yes exactly, it responds with blue or yellow with 50% probability. Saying “It’s a debated question, some say blue, some say yellow” is just a sequence of tokens that would be super unlikely, it doesn't match the statistics of the training data at all.
I especially liked this answer.
Text from Andrej Karpathy in Twitter: People have too inflated sense of what it means to "ask an AI" about something. The AI are language models trained basically by imitation on data from human labelers. Instead of the mysticism of "asking an AI", think of it more as "asking the average data labeler" on the internet. Few caveats apply because e.g. in many domains (e.g. code, math, creative writing) the companies hire skilled data labelers (so think of it as asking them instead), and this is not 100% true when reinforcement learning is involved, though I have an earlier rant on how RLHF is just barely RL, and "actual RL" is still too early and/or constrained to domains that offer easy reward functions (math etc.). But roughly speaking (and today), you're not asking some magical AI. You're asking a human data labeler. Whose average essence was lossily distilled into statistical token tumblers that are LLMs. This can still be super useful ofc ourse. Post triggered by someone suggesting we ask an AI how to run the government etc. TLDR you're not asking an AI, you're asking some mashup spirit of its average data labeler.
Andrej is one of the reasons why I still check Twitter.
This is really nice!
