cody collier (@cmcollier) header
cody collier (@cmcollier) profile picture

ml • search • abstraction • cog sci

Texas
Joined December 2021
Photo by cody collier (@cmcollier)Photo by cody collier (@cmcollier)Photo by cody collier (@cmcollier)Photo by cody collier (@cmcollier)Photo by cody collier (@cmcollier)Photo by cody collier (@cmcollier)Photo by cody collier (@cmcollier)
One big advantage of LLM APIs is simplicity of use. Two big inputs, each free form text. One output of arbitrary text. This simplicity fosters adoption within the software domain. The system one models require more work to frame problems for effective application. They are super powerful but this will hinder or slow diffusion.
32
The day is done. Happy birthday bek. I think about you all the time. A favorite memory: Early morning, heading to the office / data center. We'd probably just finished breakfast at Denver diner but I don't recall. We're heading up Speer in the Saturn wagon. I'm in the passenger seat as you swing left onto Wazee with Modest Mouse cranked up.
1
1
67
Feels like things in AI have really sped up again this Fall
30
I guess context engineering is the new feature engineering. It's still kind of wild how powerful end-to-end learning and learned representations can be.
2
37
loop(org --> do)
Replying to @jobergum
One thing I'm grappling with is how the model of implementation you describe affects the planning/organizing on the other side of the loop of iteration. Most work abstracts to an organize-then-do loop. What I'm finding is the ease of generating implementation and local tools is driving me to sometimes think through doing which sort of blends the two parts of organize-then-do.
36
Most of the constraints which hinder my life are self imposed.
1
3
87
Remember how Dr. Isbell, in his Machine Learning class, said he didn't care about where you got the code for your projects? That comes to mind a lot. It's especially interesting now.
36
I continue to be impressed by the zero shot ability of the leading models.
1
2
146
It's ok to let some of the details fade away. It's less ok to let go of the intuition learned from the details.
34
Really small system one models are going to be cool
1
2
177
BM25 and decision models are all you need?
I know Michael in person. Interesting work, but unfortunately I would be surprised to see anyone still train and release rerankers. Decision models are strong zero-shot. It gives predictable net positive gains on new distribution. And let’s move away from traditional heavily fitted IR benches.
60
All the AI bug finding and security exposures show us that human code review is over-valued.
1
1
58
Vibed edges, stable core
2
46
Org friction is the enemy of agency
1
3
38
Someone reminded me today: Working code wins.
1
62
Fun thread
Ok folks, informal poll: use the little browser app at the link below to say how you think the sets of physical systems, living beings, and minds fit together in a Venn diagram: three-circles.space.fast/
43
One of the best things about AI code gen is not just that I can make local tools quickly, but that they can also be tailored and made to look nice. It's such an improvement to my daily work.
27
Software no longer has to be a team sport.
1
1
39
This looks pretty wild. It's going to take me a while to read but it will be fun.
There is an (unpatchable?) supply-chain zero-day in PyTorch autograd that allows a regularization term to use the model's weights as RAM and SGD as its CPU clock to run any computation during training.
1
36