jaeyoon

@Caltech | @8VC Fellow | @FieldGuide

[jaeyoon @vibe_engineer]

@ysl_serena Quant problem, as in cloud compute is an inefficient market and there’s alpha in dynamically routing across providers, regions, spot markets, and reservations? Or that engineers are currently guessing which instance is cheapest instead of benchmarking cost per training run?

[Marusha @maruushae]

https://t.co/HdQHHqj3oO

That shit is such a banger that i did a backflip before posting this

[jaeyoon @vibe_engineer]

@LetMeCodeee Cpp

[jaeyoon @vibe_engineer]

@ysl_serena @karpathy You should check out the deep seek OCR paper that compresses long text tokens into an image

[jaeyoon @vibe_engineer]

@ysl_serena Survivorship bias: the ideas that were tried and didn’t work don’t get published. In hindsight, many ideas will seem trivial but you dont see the 999 ideas that failed

[jaeyoon @vibe_engineer]

@NelsonXLee @sayakmighty @8vc Take me back!!

[jaeyoon @vibe_engineer]

@ysl_serena @trydreamcast interested!

[Rudy Gilman @rgilman33]

DINO-v3 has a single high-magnitude channel on its residual pathway, channel 416. Turning off this single channel affects DINO's entire output by 50-80%. For context, turning off a random channel has an effect of less than one percent.

The model builds up channel 416 in its last two layer-scale operations, using a single high-magnitude weight in each op to drastically ramp up channel 416's magnitude. This channel doesn't depend on the input, every image fires with a constant overlay.

After bringing channel 416 up to a value of about ten thousand, DINO-v3 then scales it down in the final layer-norm to almost nothing, removing it without a trace.

[jaeyoon @vibe_engineer]

The min cut max flow was motivated by the U.S. Air Force's desire to understand the Soviet Union's railway capacity for troop and supply movement

[Roy🇨🇦 @GrandpaRoy2]

“Someone in the Main Operational Directorate of the General Staff of the Armed Forces of Ukraine very creatively calculated the vulnerability of key points of Russian infrastructure in this regard.
4/

[jaeyoon @vibe_engineer]

When I made this account I wanted to engineer good vibes…. NOT VIBE ENGINEER

[jaeyoon @vibe_engineer]

@trending_repos Did something change recently?

[jaeyoon @vibe_engineer]

@yacinemtb @iamgingertrash someone built your internet condom

[jaeyoon @vibe_engineer]

Beautiful product and such a clean execution!

[Anish Jain @0Anish_Jain1]

introducing the first internet condom

you can now filter & focus your X, Reddit, and YouTube feeds

I asked for no OF traps and lots of first-principled contrarian takes

it delivered

[jaeyoon @vibe_engineer]

@ysl_serena @grok is this logically consistent?

[Noah Smith 🐇🇺🇸🇺🇦🇹🇼 @Noahpinion]

Got some good news to share: Anti-asexuality treatments have begun to work, I can now no longer call myself ace

Got some bad news to share: I will now spend the rest of my life convincing people that the 12 years I spent not being attracted to anyone wasn't some sort of hoax

[Eric Weinstein @ericweinstein]

This is a pretty good thread. Likely true!

Consider, however, reading it from a different angle.

Stipulate for the moment that the power of mathematics and physics was so decisively proven at Los Alamos that it is obvious that The United States needs to run a secure *permanent* well funded Manhattan project, but that it cannot do so in the open, nor at a hidden location like New Mexico. In such situations what is standard operating procedure? You typically set up a “Front Company”. But this would be a Front University. And that doesn’t exist.

So what do we see: a secure campus with billions coursing through it. The returns are cartoonish: a story but with no real explanation. The fund discourages all outside investors with sky high commitments (notice the similarity to Epstein?) and is then outright closed to outside money. It employs exactly the right specialties for military work, and interacts closely with a national lab: Brookhaven. It also interacts with one of the greatest math departments anywhere SUNYSB that is consistantly and inexplicably officially ranked well *below* the level of its faculty. The fund strategy never leaks. The co-founder comes out of Defense Intelligence.

If I were a modern day Gen. Leslie Groves tasked with building Los Alamos 2.0 all over again, I would invent EXACTLY this. And I would look for Jim to lead it. He would have been one of the top 3 picks. If I were to do it today I might chose Daniel S. Freed.

I knew Jim. He was *absolutely* brilliant. But not obviously more or less so than his colleagues at the very top of mathematics. Like Oppenheimer level impressive among Feynmans, Fermis, VonNeumanns, Tellers etc. You get used to the idea that Genius tends to collect in Math, Chess, Physics and Music.

Everyone at this level is beyond impressive. And no one else has these returns.

Try reading this thread with both explanations in mind. I myself don’t know the answer. What do you think? Does the usual story work for you? Is the alternative just too crazy even to contemplate? What would you have said if I told you something didn’t add up with a lot of brilliant theotists at an Old School out in the middle of nowhere in New Mexico in 1944?

Let me know your thoughts. Thx!

🙏

[jaeyoon @vibe_engineer]

What pieces of text would have the highest entropy?

[jaeyoon @vibe_engineer]

@ysl_serena Why reproduce when you can clone yourself and live forever

[jaeyoon @vibe_engineer]

@ysl_serena Code creation/review isn’t the hard part of engineering either… it’s the business of making money that’s hard