
Leading AI @theoryvc; Teach ML @rutgersu. Prev: Led AI @_hex_tech; Led Data @weights_biases, ML @stitchfix, Led Data @bluebottleroast
Ohlone Territory
Joined December 2010
- Tweets25K
- Following2.8K
- Followers5.7K
- Likes37K
Pinned Tweet
A rat can run a restaurant, but can an agent run a bakery?
Welcome to The Great Business Bake Off: our AI Agent Hackathon.
On December 11, you and your agents get five hours to make the most profitable bakery in town.
Spots are limited: theoryvc.com/bakeoff
Bryan Bischof fka Dr. Donut retweeted
I don't remember who asked but the game theory perspective on what the NVIDIA board is trying to accomplish by saying they won't reinstate Jensen as CEO is called a STAG HUNT.
TL;DR By ruling out reinstatement, the board is trying to convince Jensen that a comeback isn’t available, encouraging him to support the transition.
In a stag hunt, two hunters (players) can choose to cooperate on a big goal (a stag) or a small goal (a hare). Unlike a prisoner's dilemma, both players (Jensen and the board) agree on the best outcome (massive profits) so they only reach the worst outcome if they think the other side won’t cooperate.
It’s a stag hunt since both prefer a successful transition to that, but neither wants to commit unless the other will too.
Here, i vectorized them for you:
1. ↖
2. ↙
3. ↘
4. ↗
yw
a couple of things to vectorize
1/ archive.org/details/illumina…
2/ archive.org/details/illumina…
3/ archive.org/details/f.-delam…
4/ archive.org/details/embroide…
1
1
241
Bryan Bischof fka Dr. Donut retweeted
i recently saw e-means and just learned about Leiden communities and then you can put them together for benefits at 100m+ scale.
Leiden finds nicely shaped clusters in a neighbor graph, but is slow / memory constrained. e-means is super fast but cuts the space into even cells. We can calculate centroids with e-means and then partition those into communities. new data comes in can be assigned to a cell and hence a cluster super quick too
what kind of nathan fielder bit is this
doing something insane
we rented a storefront in the heart of soho
to sell absolutely nothing
this year people in South Korea started using "dopamine sites", like a fake food delivery service
you place an order, but you never pay, and nothing ever arrives
you get a bit of the satisfaction of eating or shopping without the real part
so we are doing it, in real life. its a real store. you can walk in, shop items, try them on.
but when you checkout, you get nothing but an empty bag
i have no idea how this is gonna turn out
but if u want to pull up and find out, reply below : )
Bryan Bischof fka Dr. Donut retweeted
Replying to @hexanalytics
@hexanalytics did it better
Some of you have been asking very politely. Some of you have just been asking.
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it counts—the real world.
openai.com/index/introducing…
He’s no Paul Hollywood, but look at that bill!
Come to the great business Bakeoff! Dec 11!!!
What happens when software engineers and data teams meet in San Francisco to build their own bakery with an agent? The answer is.. we'll ducking see. Space is limited, join us in December if you're around: luma.com/bakeoff
4
285
Bryan Bischof fka Dr. Donut retweeted
Across the largest AI labs & AI infrastructure companies, there is a common component : a lakehouse that stores data for training & inference called @lancedb.
It’s a great day to be a friend of Lancedb, it’s an excellent day to be an investor in Lancedb, it’s an incredible day to be a customer of Lancedb.
People love to ask me my technical pov on what kinds of software and products will stick around through the wave of saaspocalypse, software factories, and personalized software.
I have a variety of responses but I always end up saying that fundamental infrastructure like Lance which is already powering several of the largest training, serving, and improvement stacks in the world.
The future is multimodal, and multimodal ai is build on LanceDB.
It’s what dreams are made on.
The next generation of models needs a data stack built for AI.
We just raised a $57.5M Series B to help researchers turn petabytes of video, images, and audio into better models, faster.
Read the announcement: lancedb.com/blog/lancedb-ser…
Saying goodbye to the SUV that I love.
Bittersweet for me. I really loved this truck. I took mine to about 70% of California Counties, 7 total states, a bazillion trailheads, and too many repair shops.
My goal was to take it from 160k to 260k but I’m coming up short at 215.
(In case your curious, the repairs are just too frequent lately and we want something a bit more modern for daily life. 2003, no nav (added smart panel), air Sus delete, air down gear up drawers, gate table)
8
27
1,621
- Lightness + Hue (OKLCH): Quantize pixels into perceptual lightness bands (L), sort by hue angle (H) and chroma (C) within each band. Prioritizes monotonic brightness
- Hilbert Curve (OKLab): Map pixels into perceptually uniform (L, a, b) coordinates, sort by their position along a 3D Hilbert curve. Approximately preserves color-space locality
- Perceptual Walk (OKLab): Construct a greedy nearest-neighbor traversal through color space, minimizing successive distances while penalizing decreases in lightness. Prioritizes smooth perceptual transitions
1
2
171
- Lightness + Hue (OKLCH): Quantize pixels into perceptual lightness bands (L), then sort by hue angle (H) and chroma (C) within each band. Prioritizes monotonic brightness.
- 3D Hilbert Curve (OKLab): Map pixels into perceptually uniform (L, a, b) coordinates, then sort by their position along a 3D Hilbert space-filling curve. Approximately preserves color-space locality.
- Perceptual Walk (OKLab): Construct a greedy nearest-neighbor traversal through color space, minimizing successive OKLab distances while penalizing decreases in lightness. Prioritizes smooth perceptual transitions.
74




















































