erasmush.bsky.social - Profile | ThreadSky | a Reddit-style client for Bluesky

Probably this whole newsletter is from a position of insecurity. www.bloomberg.com/view/article...

submitted 17 days ago • 4 comments

I was peripherally aware of #pigbutchering scams, but was unaware of their industrial scale and just how dark they are. This podcast by @suelinwong.bsky.social for the @economist.com is well worth a listen. 🎧 www.economist.com/audio/podcas...

submitted 18 days ago • 0 comments

I'm joining forces with Mike Knoop to start ndea.com, a new AI lab. Our focus: deep learning-guided program synthesis. We're betting on a different path to build AI capable of true invention, adaptation, and innovation.

submitted 44 days ago • 12 comments

DeepWait

submitted 23 days ago • 0 comments

@jobs.bacalhau.org hotdog?

submitted 25 days ago • 2 comments

@jobs.bacalhau.org hotdog?

submitted 25 days ago • 2 comments

Can’t really say that I have been there open.substack.com/pub/roonscap...

submitted 27 days ago • 0 comments

I suspect the always lucid @kaiserkuo.bsky.social just outlined a book on the Sino-American relationship

submitted 30 days ago • 1 comment

1/ Why does the U.S. seem to be caught off guard so often by what China or Chinese companies accomplish? Here’s an off-the-top-of-my head thread unpacking some reasons—structural, cultural, and epistemological—for this recurring phenomenon.

submitted 31 days ago • 2 comments

Oh so NOW the AI guys are worried about intellectual property theft

submitted 30 days ago • 7 comments

Geoffrey Hinton: "To deal with [fake content], I think you need a much better system for establishing the provenance" ➡️ youtu.be/b_DUft-BdIE?... Trust needs accountability. Accountability comes from traceability. We need a provenance system that is not going to undermine #privacy

submitted 36 days ago • 0 comments

"Sufficient speed, scale, and accuracy to directly slow #misinformation is a monumental (perhaps impossible) task, even for #Meta. However, identifying the supersharers of misinformation is not hard because their behavior is so distinctive."

submitted 38 days ago • 1 comment

The #TikTok exodus to #REDNote meant that laowais stepped onto foreign turf and had to learn the rules. It seems to have gone pretty well. With all the FUD in macro-level power dynamics, it's nice to see that (sometimes) strangers on the internet can meet and be (kind of) kind to each other.

submitted 39 days ago • 0 comments

Sorry to bring stuff over but this one is important. Careful searching for “homebrew” - Google is serving an ad for “brewe.sh” which is a clone/malware site.

submitted 39 days ago • 1 comment

Common pitfalls (with examples) when building AI applications, both from public case studies and my personal experience. huyenchip.com/2025/01/16/a... Would love to hear from your experience about the pitfalls you've seen!

submitted 42 days ago • 0 comments

#REDNote (aka "Little Red Book") = #TikTok's pushy algorithm + heavy-handed, capricious moderation. Strange marriage between capitalism’s heroic free choice and #Marxist superstructural determinism.

submitted 43 days ago • 0 comments

Making #ATProtocol more robust is a noble endeavour, but is anybody thinking about how to mitigate The Great Splintering of our social information sphere that a decentralised network will inevitably bring? Suggested read: www.noemamag.com/the-great-de... by @noupside.bsky.social

submitted 45 days ago • 0 comments

"How many people understand the value of both a mala and a gun, and can carry each without feeling like a fraud?" Sam Harris reflecting on life priorities while fleeing his house during the California #wildfire.

submitted 45 days ago • 0 comments

Would be curious to see a scoreboard 🥇 From my experience, multi-agent frameworks tend to go off the trail quite quickly. @virattt.bsky.social do you have any results that you could share in public?

submitted 53 days ago • 0 comments

I don't think people really appreciate how simple ARC-AGI-1 was, and what solving it really means. It was designed as the simplest, most basic assessment of fluid intelligence possible. Failure to pass signifies a near-total inability to adapt or problem-solve in unfamiliar situations.

submitted 57 days ago • 5 comments

#DeepSeek founder Liang Wenfeng: “… moats created by closed source are temporary. […] So we anchor our value in our team [to] form an organization and culture capable of innovation. That’s our moat.” www.chinatalk.media/p/deepseek-c...

submitted 58 days ago • 0 comments

The next 3-5 years are basically going to be the Great Product Delivery period of the LLM-based AI era. Some advice from Andrew Ng for product managers in the space: www.deeplearning.ai/the-batch/is...

submitted 77 days ago • 0 comments

"In a world of online impersonations, scams, multiple identities, deepfakes, and other realistic yet deceptive AI-generated content, we need “proof of personhood” — something to help us know that we’re interacting with an actual person." Eddy Lazzarin → a16zcrypto.com/posts/articl...

submitted 80 days ago • 0 comments

“Llama 3.1 405B and Claude 3 Opus confess in ~80% of the cases, whereas o1 is surprisingly persistent and confesses in <20% of cases.” www.apolloresearch.ai/research/sch... #o1

submitted 82 days ago • 0 comments

Software ate the world and regurgitated something very weird. When our understanding of the world is framed by user interfaces, the pursuit of neutral systems (or at least not criminally mendacious systems) becomes a critical aspiration. Recommend read: sarahconstantin.substack.com/p/neutrality

submitted 83 days ago • 0 comments