it’s just tweaking weights to minimize loss

Joined March 2010
Pinned Tweet
244
Alex retweeted
Rust compiler: "I cannot allow you to compile this because reference x might outlive borrowed scope y on line 42." C compiler: "You want to cast an integer to a function pointer and jump straight into uninitialized heap memory? Hell yeah brother, here's your binary."
107
449
54
10,223
275,086
“You can’t decompile video games and sell people new products composed of old intellectual properties.” Brother my president insider trades daily and none of my peers will own a house until 40. A child not embraced by the village will burn it down to feel it’s warmth.
71
1,084
50
16,417
1,603,881
Web crawlers /[•]\ #js x #css
1,064
8,600
941
69,566
3,251,796
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution
4,035
8,296
6,819
76,633
40,091,671
😂😂😂😂
23
1,008
25
5,816
257,415
imagining a few small gestures to edit/write with ai: word: spin synonyms sentence: shift the tone (cold <-> warm) paragraph: shorten/expand
224
854
181
8,940
1,037,518
If a CPU cycle (0.3 ns) felt like 1 second, a 300 ms request would feel like 32 years. I’m trying to say that 300 ms is an eternity for a computer. Software should be way better than what we just accept these days.
92
236
26
7,020
308,252
This shot eclipses all the others
37
743
16
8,994
339,069
Alex retweeted
1
24
3
733
17,909
documentação escrita por LLM tem cheiro. e dá pra tirar manda o modelo seguir o ASD-STE100, o simplified technical english. 53 regras de escrita, cerca de 900 palavras aprovadas, issue 9 de janeiro de 2025 ele nasceu nos anos 80 pra manual de manutenção de aeronave ser lido igual por mecânico polonês, brasileiro e japonês. cada palavra tem um sentido só e um verbo por ação o efeito colateral em LLM é direto: sem sinônimo bonito, sem subordinada, sem floreio. sobra a instrução hoje 64% de quem usa o padrão está fora de aeroespacial
26
110
9
2,353
118,398
Alex retweeted
homeboy you're the one sellling public market investors on short-term space datacenters
He takes scamming to a whole new level
7,636
4,605
3,091
59,581
15,683,515
been in ML research for 7 years, wrote a paper on compression & scaling laws, and passed openai's information theory interview yet the latest 3b1b *still* gave me fresh intuition on entropy either i'm an impostor or 3b1b is the greatest teacher of all time
50
137
13
3,517
130,328
l0g est maintenant MCP Native 👉 l0g.fr/mcp/ et AI agents ready 👉 l0g.fr/agents/
2
7
34
3,785
In-context learning in LLMs
39
304
30
3,425
167,349