Linkstream

TLDR: Lesswrong discovers B2B sales tooling of 2020
(yes LLMs made it a bit easier to built but it's not news really. probably existed years earlier)
https://www.lesswrong.com/posts/GCHyDKfPXa5qsG2cP/human-study-on-ai-spear-phishing-campaigns

matches my experience 100%. I take a lot of notes, and they are super useful, but...

when I don't care how I might write something down that's like 3x boost to thinking speed and ~2x boost in available "context size", in LLM parlance

upd: actually, one of my biggest disappointments in childhood was that the words don't map 1:1 to what we now call "concept embeddings"

PDF: https://wz.ax/2024/language-is-a-tool-for-communication

this is how you do observability. or logging, or tracing, whatever you call it. preaching this for years, now somebody wrote the guide so I don't have to.
https://jeremymorrell.dev/blog/a-practitioners-guide-to-wide-events/

Tachikomas (from the Ghost in the Shell universe) are now real!

https://x.com/conspiracyslyr/status/1868737204916552052?s=46
flat earthers visited the south pole, confirmed the earth isn’t flat and got expelled from the flat earth community 😅

cooperation and coevolution is the real alignment, not any kind of standardized testing (which will just get overfit)

Cultural Evolution of Cooperation among LLM Agents
https://arxiv.org/abs/2412.10270

AutoReason: Automatic Few-Shot Reasoning Decomposition
TLDR: explore simpler assertions in first prompt, draw conclusions in the second
https://arxiv.org/abs/2412.06975

writing things down helps if you want to commit;
but also attracts too much attention to what you just wrote and distracts from other possible options
https://arxiv.org/abs/2412.06769

turns out the backdoor is a door after all ¯\_(ツ)_/¯
https://tuta.com/blog/china-salt-typhoon-worst-us-hack

Nous Research designers are really cracked
https://distro.nousresearch.com/

Kinda unremovable Linux-focused UEFI bootkit
https://www.welivesecurity.com/en/eset-research/bootkitty-analyzing-first-uefi-bootkit-linux/

ironically it seems that the real Open AI is called Qwen (with DeepSeek not that far away)

this time, a 32B model competing with o1-mini

https://qwenlm.github.io/blog/qwq-32b-preview/

PSA: Ubuntu 16..24 local privilege escalation vulnerability
https://github.com/makuga01/CVE-2024-48990-PoC

St. Peter’s Church in Lucerne, Switzerland has introduced an AI hologram of Jesus, called Deus in Machina, to take confessions. The AI sits behind a screen, providing advice and responding to worshippers' queries. One participant said, “I was surprised, it was so easy, and though it’s a machine, it gave me so much advice.”
https://www.disclose.tv/id/z33eqyr1bv/

curious approach, something akin to an in-context LoRA model finetuning (plus a lot of other tricks)
https://arxiv.org/abs/2411.07279

https://www.youtube.com/watch?v=T7sxvrJLJ14

sometimes training your own model is /better/ than making an app.
also a lesson that sometimes you can reach SOTA by yourself!
https://moondream.ai/ rocks and so does @vikhyatk

Updated Alibaba's Qwen 2.5 Coder model is
a) solid GPT4 level in code generation (okay),
but also
b) does that in less than 0.1x the resources compared to SOTA (llama3.1) just 3 months ago 🤯 (tested both myself)
https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct

Kaggle Expert level reached with the end-to-end competition solver
(the paper says grandmaster but ahem not quite)
https://wz.ax/agent-k/2411.03562

critical tcp/ip RCE (CVSS 9.8) in windows

https://msrc.microsoft.com/update-guide/vulnerability/CVE-2024-38063