patrick (@patrick0d) profile picture

ai researcher and ai engineering

Joined July 2012
claude entered the age of reverse hallucinations
when is the last time you thought about hallucinations
38
some tokens are trillions of times more important than others
A token is ~a word … A human generates ~a token/sec There are ~10B humans on Earth Humans generate ~10B tokens/sec A H100 generates ~1,000 tokens/sec There are ~10M H100s et al. Machines generate ~10B tokens/sec 2026 is the fleeting equilibrium of man+machine.
1
80
did anyone else have the same first thought?
38

Today, @steipete became the first confirmed (long theorized) one-person Unicorn! OpenClaw launched as a weekend side project by less than 3 months ago. Here's the timeline: Nov 25: First GitHub commit. Starts as a weekend side project Jan 25: Goes nuclear. 9k stars in one day. 2M website visitors in a single week Jan 27: Renamed Moltbot (possibly the worst renaming ever) after Anthropic trademark complaint Jan 29: Renamed @OpenClaw (thank GOODNESS!), blog confirms 100k+ stars milestone Early Feb: Hits 145k–149k stars + 20k forks Feb 11: Steinberger notes 180k people starred the repo Feb 12: Steinberger on Lex Fridman podcast, shares story & vision Feb 15 (today): Sam Altman announces Steinberger joining OpenAI to build next-gen personal/multi-agent systems Full cycle: 82 days from zero to this Absolutely INSANE!
Readers added context they thought people might want to know
There has been no official confirmation on price or amount paid for @steipete joining OpenAi. techcrunch.com/2026/02/15/ope…
3
224
a trillion shrimp watching sbf doing crypto instead of waiting 2 years and doing AI imagine the valuations
36
"Oh my gosh, the composer did NOT need to go THIS hard for this song!" Yes they did, it is literally their job. Fuck you
290
why isn't this one node connected to the rest of the fly's brain @SebastianSeung
the connectome revealed the literal convolutional net inside the fly brain, including its architecture of 230 feature maps, normalizations, and other layers,
1
114
Replying to @patrick0d @xai
here's the whole tool. it was a ui that showed live updates to a users preferences and showcased all of groks functions against this new recommender architecture
60
7/ We can then cluster users by actual topics for ads and connections. Steer how assistants talk - a tech person gets a different explanation than someone who isn't. Grok understands the user in its own language.
1
1
66
3/ The I embed those traits and the post becomes a point in space. The space is Grok's worldview. Not learned embeddings that drift and need daily retraining like with TikTok's approach. Grok already knows how the concepts in posts relate.
1
59
1/7 I entered the Grokathon at @xai in London this weekend. X just open-sourced their For You algorithm where Grok runs per user, per post, per request. I built a version where Grok learns online without inference. It's a million times more efficient.
3
12
20,665
Thread 1/15 🧵 We don't know how LLMs pick their words. After 3 years of ChatGPT, trillions of weights, and billions in compute, a model still fires hundreds of millions of activations just to answer "2+2=4" But I found something everyone's been avoiding: LANDED WRITES
1
1
3
166
My robovac’s lidar apparently discovered the door to Narnia under my oven.
1
3
319
Replying to @norvid_studies
i was going to type what he might say but instead will just call the algo slop
2
16
after 75 days i am presenting myself an award as being the first person in history to read every word of one of o3's deep research reports
72
oepnai deep research didn't agree but looks close enough for me
1
1
176
incredible i spoke to llms for about 12 hours already and this is the first time today i have been blown away
not sure I believe chatgpt's work here but I don't have the monster energy and drywall expertise to question it
4
1
10
603