Shannon Sands
@max_paperclips
Followers
9K
Following
276K
Media
1K
Statuses
31K
Software developer & cognitive architect https://t.co/JAoBrqMLXN.
Joined February 2023
I spent the last two days at the Vatican attending the AI Forum It was basically a Council of Elrond-style gathering of technical practitioners (me), philosophers, and theologians Good news, the Church is broadly aligned that AI is a force for elevating humanity so long as we
Technological innovation can be a form of participation in the divine act of creation. It carries an ethical and spiritual weight, for every design choice expresses a vision of humanity. The Church therefore calls all builders of #AI to cultivate moral discernment as a
3
4
27
TODAY: Stop by Flatiron Public Plaza to learn about Grayscale CoinDesk Crypto 5 ETF (Ticker: $GDLC) and the underlying tokens. Bitcoin Ethereum $SOL $XRP $ADA We’ll be sharing how $GDLC streamlines crypto exposure and giving attendees a limited-edition Grayscale Speedcube to
42
93
577
normalize decorative first letters in code snippets
92
1K
16K
My child will not be allowed to use chat gpt. He will be smarter and stronger than the other children and he will kill them easily.
656
34K
273K
If you ever wonder how Chinese frontier models like Kimi, DeepSeek, and Qwen are trained on far fewer (and nerfed) Nvidia GPUs than US models. In 1969, NASA’s Apollo mission landed people on the moon with a computer that had just 4KB of RAM. Creativity loves constraints.
124
277
4K
This cyclist nearly lost his life in a crash. Learn why he thinks Waymo will help improve road safety.
1
3
13
A note on costs/compute Base Kimi K2 model used 2.8m H800 hours with 14.8 trillion tokens, about $5.6m worth Details of post training for reasoning not given, but it is likely max 20% more (excluding data prep!) Would be < $3m for sota if they had Blackwell chip access
🚀 Hello, Kimi K2 Thinking! The Open-Source Thinking Agent Model is here. 🔹 SOTA on HLE (44.9%) and BrowseComp (60.2%) 🔹 Executes up to 200 – 300 sequential tool calls without human interference 🔹 Excels in reasoning, agentic search, and coding 🔹 256K context window Built
21
45
514
I'd have expected "lack of novelty" or something, but nope apparently the reviewers don't know this kinda pipeline is what everyone is doing now
The PipelineRL paper getting rejected at NeurIPS reminds me of when the Megatron-LM paper got rejected from every conference back in 2020 scientific reviewers still don’t recognize a good systems paper when they see one https://t.co/KMPnx5HlnX
1
0
25
me playing around with my tiny xLSTM implemention in MLX lmao
@_ueaj Don't lecture me, uaj, I see through the lies of the GPU managerial class. There is no free lunch for parallelism. I do not fear sequentiality as you do. I have brought speed, scalability and memory sustainability to my new empire.
0
1
17
a mass movement of voters? In my democracy? Say it isn't so
i miss “post rationalism”, a principled school of thought, which has now been replaced by its idiot cousin “populism”
1
0
22
So, I just got back from Hangzhou, China; Attended @IROS2025. Takeaway: China’s robotics ecosystem is moving faster and more coordinated than the US
14
7
56
> If you want to do multi-agent multi-turn LLM RL, might as well do commit sudoku. Accurate
Aight let's talk about frameworks, libraries, RL, and why I probably don't like your favorite RL codebase. Yes, including that one. The unusual thing about RL is that the algorithm is the easy part. GRPO is a single-line equation on some logprobs. If you have the data, computing
3
0
41
New idea for web apps: Need to wrap it up, put it on NPM and have a full in-browser OS, and then serve the actual site on it's localhost, purely to add more layers of abstraction to the frontend
Fun. Someone patched the Linux kernel so it can run on WASM. Big difference between some simulations that sandbox or warp the kernel (spec. LKL) while this actually compiles it for WASM Biggest changes on patches kernel-0005, llvm-0001, musl-0001. arch/wasm is born
2
2
24
The path to autonomous AI is a system that learns to solve new problems by synthesizing models of them on the fly (as code), and that gets smarter over time by adding new abstractions to its own library (also as code), compounding its capabilities. Not a static map -- rather, an
51
39
413
Break free with the ROG Xbox Ally and ROG Xbox Ally X. Take your favorite games with you, wherever you are!
0
2
26
brrrrrr
0
1
46
This isn't true, I have dreams (occasionally nightmares) about code constantly lmao. I don't dream about my phone though, which is kinda interesting
2
0
18
"it's Python, you do anything and it allocates" How true is this? I modified CPython to print when it allocates an integer object Then added numbers in a for-loop 100k times My terminal got spammed with 101006 allocations Why? Let's explore the internals of CPython:
I will never forgive Rust for making me think to myself “I wonder if this is allocating” whenever I’m writing Python now
25
41
794
Hedera’s Hashgraph replaces blockchain’s linear chain of blocks with a Directed Acyclic Graph (DAG), where transactions confirm each other asynchronously, similar to how news spreads by word of mouth The result: parallel consensus, sub-second finality, high throughput & low fees
0
18
211
I think it actually is beautiful
0
0
18
if your lithium batteries get hot it’s best to dunk them in water to cool them down
31
3
279
Another side benefit of this exploit is, w/ root access to the main board, you have a stronger e-stop. When running low level policies, the remote controller stops working You can even reboot the jetson board and robot still moves. A reboot of main board reliably stops things
@ChongZitaZhang @benjamin_bolte This wasnt fake. I was able to replicate it and posted a video doing it live for this exact reason. I still have root and sudoer on my g1 main board that i used the reported bluetooth exploit to get, which anyone could do in range.
4
6
64
@ChongZitaZhang @benjamin_bolte Correct, you can see I am connected to the .161 PC1 in photo. vid of me testing exploits: https://t.co/vWtY7znfS2 You can safely test reboot command like I did in video. I later enabled ssh, and used that same method to setup a new sudoer. You can also just change root pass.
2
3
20