.. trained GPT-2 from scratch, kept all 32 checkpoints, and asked each “Who painted the Mona Lisa?”
Step 0: gibberish
Step 6k: “a very beautiful painting”
Final: “Leonardo da Vinci… in the early 14th century” 😅
Built on @karpathy’s recipe. Ask yours:
app.train.galvyn.xyz
When does a language model learn that water 💧 is a liquid
I'm training GPT-2 (124M) live on one @nvidia RTX A5000 to learn. It's my 🏌️♂️ simulator :)
Every step, watt and checkpoint is public, and you can talk to any checkpoint.
app.train.galvyn.xyz/experim…
Got tired of waiting Same model, same 10B tokens, two machines, both live: 🏠 1× RTX A5000 under my desk: ~55 hours, ~$2 of electricity ☁️ 8× H100 rented on
@googlecloud
: ~40 minutes, ~$60–80 Watch both train: ( & both have playgrounds to ask Qs)
Golf Sim - app.train.galvyn.xyz/experim…
H100 -
app.train.galvyn.xyz/experim…
One of the best episodes ! Connects something that you can try to what it takes to make it happen
after listening to this it seems like Capital. Energy, GPUs & most important the appetite to take the bet is the new moat
Cheeky Pint is back! Mark Zuckerberg @finkd and Alexandr Wang @alexandr_wang came to the pub to discuss their new frontier model, Muse Wonder.
00:00:24 Muse’s breakout use cases
00:07:28 The model is the product
00:14:35 Muse Wonder
00:24:56 How frontier models are built
00:48:26 Betting $100B
01:07:21 Agentic commerce
01:19:34 OpenRouter
01:20:06 Beyond the smartphone
01:27:00 Funding
01:34:06 Why AI won’t be winner-take-all
01:40:46 Meta Superintelligence Labs (MSL)
02:08:11 Working for Mark Zuckerberg








