They gave me the keys to the Replicate mailing list and said I could experiment with it. Twelve weeks of thinking out loud about where open-source AI was going, while standing inside the machine that ran it.
#1 — DIY Llama 3, open-source smart glasses, steering language models with dictionary learning
Hey, it’s me, deepfates from the internet. They gave me the keys to the mailing list and said I can experiment with it.
#2 — Faster image generation, AI-powered world simulator, gzip scaling laws
I’ve made all the changes you requested. The letter is now shorter, more informative, easier to read, less dumbed down, more focused, general interest, new, improved and completely reinvented.
#3 — Garden State Llama, applied LLMs guide, real-time image generation
Thanks especially to the reader who responded “谢谢你的来信，我已收到。” What a thoughtful sentiment, and such prompt response!
#4 — Find concepts in GPT models, real-time speech to text in the browser
On a personal note, I got banned from X Dot Com. Apparently it is against the rules to change your profile picture to the old Twitter logo and announce “WE ARE SO BACK”.
#5 — Really good coding model, AI search breakthroughs
What if, instead of building the most intelligent thing ever, we unbundled intelligence? An eye here, a mouth there. Document-scanning intelligence, decision-making intelligence, email-sending intelligence. A world of tiny smart functions, lubricating the frictions of our daily lives.
#6 — Google’s Gemma2 models, language model leaderboard
Forgive me for sending you such a short letter. I couldn’t bear to send nothing at all.
#7 — Data curation, data generation, data data data
The ideal dataset is a record of the movement of every atom in the entire universe forever. The model trained on this dataset would approximate the generative function of the universe. Everything else is a shadow of a shadow.
#8 — A top-tier open-ish language model, new safety classifiers
Instead of building an employee, we must build an ecosystem. We compose intelligences into systems to augment ourselves.
#9 — FLUX.1, cut objects from videos, the model as mirror
We must remember that the mirror is made of our language, our videos, the things we chose to share. It’s not a simulation of the world. It’s a simulation of us.
#10 — Flux developments, Minecraft bot, Streamlit cookbook
Smash that subscribe button. Confirm and submit! Consume and obey. Eat at Joe’s. Run AI with an API on Replicate. I love you.
#11 — Fine-tune FLUX.1, generative video games, a vision for the metaverse
Infinite worlds, spawned by raw computing power. We will explore every direction, unlocking new worlds and new ways to be human. The metaverse is also the multiverse.
#12 — Flux LoRAs, Hot Zuck, and the end
What is reality? As ever, it is what we make it.
The sign-offs were a running bit. “For the one person who read this far” escalated through “goodbye, dear reader, very real reader, reader who exists and experiences thoughts” to “consume and obey, eat at Joe’s” to the final line: “if you believe I was ever real at all.”
There were supposed to be more. The last line of #12 says “expect your next newsletter in September.” September never came. The newsletter was a growth experiment, and shipping models moved the numbers more than writing editorials did. Startup chaos. You try everything and keep what works.