will to code
Ir al canal en Telegram
353
Suscriptores
-124 horas
-17 días
-330 días
Archivo de publicaciones
353
Apple maxing
Remove all the options
Every time you open my app it will pick a random book for you to read
I am building the TikTok of reading lol
Ow, you want options? that'll be nine ninety nine $$$
353
xAI would never make something like this since it would crash their daily active users on X
I would never open X if I had this
353
Finally an accurate benchmark [DeepSWE]
imo i think swe is solved
I don't think any programmer desires an agent better than gpt 5.5. only improvements that I see are speed and cost
the next frontier is science models 🫠
353
This is what I'm working on.
I wanted to make reading feel more engaging and enjoyable, so I added big beautiful fonts and an ambient video background. It's very immersive with headphones.
Books still beat e-readers in a lot of ways, but maybe that's because we haven't fully explored what digital reading can be. We have this incredible digital canvas, and we're not using it creatively enough.
353
I started posting on wip.et
It's awesome!
But it would be pretty cool if there was a thread dedicated to what people vibe coded
Since tokens are becoming so abundant. I want to see what people do with them.
353
is it just me or has X's algo become more like instagram
like no interesting people at all
Only slop
I used to get interesting tweets, now its full of random people's opinions and facebook videos
353
gpt-image is one of the best tools ever! It can literally bring your ideas to life and show them to you ✨✨✨
353
this is beautiful 🤌
these companies are risking everything to bring us closer to AGI
literally the best use of money from the time of its conception
infinite money progress glitch
more ♻️♻️♻️
keep in mind that banks are allowed to shadow clone jutsu your money out of nothing
353
"delusion is all you need"
"just be delusional bro"
"delusion delusion delusion"
....
Until you actually meet someone who is delusional
Someone that substitutes fantasy for effort
353
Distillation is very bad because language is just an abstraction
The real intelligence is in the weights and that can't be copied
You can talk to Einstein all you want but you'll only end up as a poor imitation
I've been wondering about this and Anthropic's whole moat might be data quality. They are super strict about data and go as far as to scan high quality books themselves (project panama). And other labs wanting Claude's capabilities, distill their models. Ofc you are going to fall short of what you distill.
Seriously who does Anthropic distill from? Nobody. Right. Because why would they. Then maybe distillation is not necessary for SOTA capabilities...
It would be funny if I'm right hehe
If you distill from Claude your model will have a lower IQ than Claude. LLM's are not like humans the student always ends up dumber than the master.
