djsh

4.5K posts

djsh

djsh

@djays

I build AI stacks and models, currently at Groq. (Past: LLMs, Autonomous Checkout, Medical Imaging).

SF Bay Area, CA Katılım Nisan 2008
731 Takip Edilen316 Takipçiler
djsh
djsh@djays·
@patrickc What's your stack and model of chocie? I find most models aren't grounded in the nuance of genetics and can overindex on impact of a single gene.
English
0
1
0
277
Patrick Collison
Patrick Collison@patrickc·
I'm lucky enough to have a great doctor and access to excellent Bay Area medical care. I've taken lots of standard screening tests over the years and have tried lots of "health tech" devices and tools. With all this said, by far the most useful preventative medical advice that I've ever received has come from unleashing coding agents on my genome, having them investigate my specific mutations, and having them recommend specific follow-on tests and treatments. Population averages are population averages, but we ourselves are not averages. For example, it turns out that I probably have a 30x(!) higher-than-average predisposition to melanoma. Fortunately, there are both specific supplements that help counteract the particular mutations I have, and of course I can significantly dial up my screening frequency. So, this is very useful to know. I don't know exactly how much the analysis cost, but probably less than $100. Sequencing my genome cost a few hundred dollars. (One often sees papers and articles claiming that models aren't very good at medical reasoning. These analyses are usually based on employing several-year-old models, which is a kind of ludicrous malpractice. It is true that you still have to carefully monitor the agents' reasoning, and they do on occasion jump to conclusions or skip steps, requiring some nudging and re-steering. But, overall, they are almost literally infinitely better for this kind of work than what one can otherwise obtain today.) There are still lots of questions about how this will diffuse and get adopted, but it seems very clear that medical practice is about to improve enormously. Exciting times!
English
488
642
9.6K
4.1M
djsh
djsh@djays·
@adelwu_ Would love to attend this
English
0
0
1
40
adel 🌟
adel 🌟@adelwu_·
i want to host an event for people confused about their career path. for the multi-hyphenates who don’t know what to pursue, bc there’s too many things. i left my last job not knowing what was next, only that i didn’t want to keep doing what i was doing. when i pivoted from eng to growth, the most eye opening conversations were those with people 2-3 years ahead of me who went through the same struggle. they could actually resonate with the existential stress i felt navigating the process. who would be interested? i’d love to put together a panel of people with amazing multi-faceted careers and took sabbaticals/breaks to figure things out!
Marc Randolph@marcrandolph

My path to entrepreneurial success was not linear, by any stretch of the imagination. I didn’t start working in tech until I was 32. I didn’t even move to California until I was 30. Before becoming an entrepreneur, I was: -The worst realtor in the state of New York -A gofer for the CEO of a sheet music company -An aspiring brand manager for flea shampoo Don’t be disillusioned if the path ahead isn’t clear. Relax. Find something that strikes your interest. And don’t be afraid to take a trail just because you can’t see the end.

English
46
5
236
46.4K
djsh retweetledi
Bo Wang
Bo Wang@BoWang87·
Prof. Donald Knuth opened his new paper with "Shock! Shock!" Claude Opus 4.6 had just solved an open problem he'd been working on for weeks — a graph decomposition conjecture from The Art of Computer Programming. He named the paper "Claude's Cycles." 31 explorations. ~1 hour. Knuth read the output, wrote the formal proof, and closed with: "It seems I'll have to revise my opinions about generative AI one of these days." The man who wrote the bible of computer science just said that. In a paper named after an AI. Paper: cs.stanford.edu/~knuth/papers/…
Bo Wang tweet media
English
150
1.9K
9.1K
1.4M
Ian Andrews
Ian Andrews@IanAndrewsDC·
Just landed. This is the first thing I see outside the terminal. Tell me which city.
Ian Andrews tweet media
English
11
0
12
1.4K
Akanksha
Akanksha@akankshanc·
@Cohere_Labs is an open science community with many specialized subgroups to learn and conduct research on various ML/AI subfields, and you can join the community here: cohere.com/research/open-…
English
1
1
19
1.5K
Akanksha
Akanksha@akankshanc·
After building some mathematical foundation for transformers, we are on to tackling the next foundational paper, Toy Models of Superposition at the ML understanding group of @Cohere_Labs. This paper explores how networks pack many features into fewer dimensions, forming beautiful geometric patterns that balance efficiency and interference. PS: I still find these geometric arrangements surreal even after reading this paper so many times! Join us Friday, 10/17!
Akanksha tweet media
English
5
22
326
26.7K
djsh
djsh@djays·
Excited to get started on this! Also, the reader notes *chefs kiss*
djsh tweet media
English
1
0
2
70
djsh retweetledi
Groq Inc
Groq Inc@GroqInc·
It’s official: McLaren F1 x Groq Bringing inference speed at a winning cost to the grid and beyond. See you in Singapore. 🧡🏁
English
116
54
469
218.5K
djsh retweetledi
Groq Inc
Groq Inc@GroqInc·
📣Groq’s first agentic system is ready for production at scale. Already battle tested by 100K+ developers across 5M+ requests. Compound is now GA, available to everyone on GroqCloud. Go Build ⬇️
English
26
43
332
77.4K
djsh retweetledi
François Chollet
François Chollet@fchollet·
We were able to reproduce the strong findings of the HRM paper on ARC-AGI-1. Further, we ran a series of ablation experiments to get to the bottom of what's behind it. Key findings: 1. The HRM model architecture itself (the centerpiece of the paper) is not an important factor. 2. The outer refinement loop (barely mentioned in the paper) is the main driver of performance. 3. Cross-task transfer learning is not very helpful. What matters is training on the tasks you will test on. 4. You can use much fewer data augmentations, especially at inference time. Finding 2 & 3 mean that this approach is a case of *zero-pretraining test-time training*, similar to the recently published "ARC-AGI without pretraining" paper by Liao et al.
English
45
295
2.6K
368.2K
djsh retweetledi
Groq Inc
Groq Inc@GroqInc·
OpenAI’s open models are live and already running on Groq. Try gpt-oss-20B and gpt-oss-120B today. Groq delivers 128K context and built-in tools such as code execution and browser search. For the first time, developers and enterprises can deploy open models backed by OpenAI instantly, anywhere, at scale. Start building now. Links in comments.
English
59
162
1.8K
1.5M
djsh retweetledi
Aarush Sah
Aarush Sah@aarush·
Introducing OpenBench 0.1: Open, Reproducible Evals 🧵
English
39
61
408
146.4K
djsh
djsh@djays·
@paraschopra Instructing Claude to regularly update Claude.md. Also multiple markdowns for bigger codebases. (Essentially a form of long term memory)
English
0
0
6
834
Paras Chopra
Paras Chopra@paraschopra·
Just tried Claude Code and it’s awesome! There’s something about vibe coding in the terminal that feels more fun than doing it in IDEs! I love how it makes its own todos and checks off them one by one. Any tips on Claude Code that you can give me?
English
102
17
888
69.2K
djsh retweetledi
stochasm
stochasm@stochasticchasm·
thought he only had 2 moments
stochasm tweet media
English
4
8
116
7.7K
Paras Chopra
Paras Chopra@paraschopra·
Just finished a summer course at Oxford on Plato’s Republic. What’s inspiration is that my class comprised of people of all ages (including one who is 80+) Everyone was there for their hunger for learning, and I hope it stays with all of us until the very end of our lives.
Paras Chopra tweet media
English
51
18
1.3K
46.3K
Elsewhere
Elsewhere@elsewhere_today·
Everyone in SF complains about how every café closes by 5pm and there’s nowhere to work. Introducing Elsewhere: the world’s first late-night café coworking chain for builders, creatives & digital nomads. Launching July 24. Comment your favorite drink & we'll DM you the invite.
English
276
20
783
166.1K
djsh retweetledi
sunny madra
sunny madra@sundeep·
Groq 👀👀👀 1) Which inference providers are you using or considering to access models? 2) Which inference providers are you using or considering using to access models? Survey by @ArtificialAnlys
sunny madra tweet mediasunny madra tweet media
English
10
22
165
227.8K
djsh retweetledi
Groq Inc
Groq Inc@GroqInc·
*YOLO Launch* Kimi K2 is now in preview on GroqCloud at 185 tokens/sec. Build fast. Link in comments.
English
81
84
1.3K
269.9K