Bit offended that none of my casts in four years was deemed ban worthy

4 replies

3 recasts

39 reactions

(1/2) - The survival of man depends on the early construction of an ultra- intelligent machine [AGI]. - Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever [AGI definition]. - Until an ultraintelligent machine is built, perhaps the best intellectual feats will be performed by men and machines in very close, sometimes called symbiotic relationship [current co-working with LLMs]. - The greatest mental achievements depend on more than memory, but it would be a big step toward ultraintelligence if human methods of recall could be simulated [LLM training]. - An example of generalized and hierarchical regeneration [reconstruction] is in the use of words in a language. A word in a spoken language could well be defined aa a clump of short time series [tokenization].

1 reply

2 recasts

30 reactions

1/5 Necroing this cast from a fortnight ago (which, in the age of the singularity, is eons). There's now visible tension in the math community between two fundamental views of the craft: is math about problem solving, or about deepening our (human) understanding? It even made the HN front page today, so it's grabbed the attention of even pedantic nerds. To be clear, I'm not a mathematician, but the question strikes me as an epistemological one.

1 reply

2 recasts

13 reactions

One of the 21st-century first-world annoyances of the modern web is that I rarely can make sense of the source code of any website anymore. It’s often all a jumbled, obfuscated framework mess that has no obvious bearing with the contents. Like I could search for a keyword that I’m seeing on the page, and it doesn’t even show up in the source because it’s part of some dynamic call. Factoid about me: while studying for my Masters (98–00), I was a part-time webmaster for the university. Having to manually open and close HTML tags in a text editor, figure out the right CSS declarations through trial and error, manage the proper indentation, optimize images so they’d load fast, actually think about the design of the page elements in relation to each other, etc. was an exercise in meticulousness and patience, but also a great learning experience. I’m very happy with the modern state of HTML5, advanced CSS capabilities, JS-free UI, etc. I just wish we stuck to those simple standards without bloat.

3 replies

1 recast

14 reactions

Mood rn

3 replies

1 recast

25 reactions

So Claude just casually formalized the proof of Fermat’s Last Theorem in 13M LOC (the largest Lean proof ever produced). We’re heading toward this kind of feat being a monthly, then weekly, then daily occurrence; which is a defining trait of the singularity btw. We’re just missing the recursive self-improvement in AI… for now. 2027 is going to be wild. https://github.com/anthropics/fermats-last-theorem

3 replies

4 recasts

23 reactions

Terence Tao has this superb analogy for AI — initially in the context of math, but with applicability to other domains. Adapting a bit, it goes something like this: Let’s say you go on a hike, and you want to visit a particular waterfall. You brought a map, but you still get a bit lost on the way, and in doing so, you come across other noteworthy landmarks. Perhaps you find a cave, which you plan on sending your spelunking friend to explore another day. Eventually, you chart a path to the waterfall, and even if you don’t, it’s been a gratifying and productive journey. AI is like the helicopter that will take you directly there. You do reach your goal, and faster too, but you haven’t learned much about the terrain and you missed those features that only reveal themselves to those who wander by foot. So be deliberate and intentional about whether any particular journey is worth experiencing on your own, or if you just need to shortcut your way to your destination

5 replies

1 recast

20 reactions

I’m increasingly thinking of LLMs as Libraries of Babel. As their training set asymptotically approximates all human outputs ever recorded, they gain the ability to churn out every possible permutation of extant knowledge in a multidisciplinary, nexialist, polymathic way (connecting every dot to every other dot). But if we raise the models’ temperature, they can also provide every possible hallucination, including one that predicts your future to a tee, or contains Satoshi’s private key, etc. — which is pretty exciting in its own right, even as we tend to view hallucinations as aberrations to squash. The Library of Babel always had the capacity to do all of the above. The hurdle, of course, was to find meaning in a googolplex number of outputs that are mostly gibberish. So just like the librarians in Borge’s story are tasked with curating meaningful outputs, the best LLM whisperers are those who know exactly *what* to ask in a near-infinite search space, and how to critically parse the answers.

2 replies

0 recast

7 reactions

1/5 The news about a cancer mRNA vaccine successfully completing a Phase 3 trial tracks with the emerging consensus from the last few years that virtually all human diseases could be either prevented, stabilized, or cured within the next few decades, with obvious implications not just to average life expectancy, but to healthier living in old age (equitable access to medication notwithstanding). Having said that, I was never sold on the immortality proposition.

3 replies

5 recasts

27 reactions

Farcaster didn’t fail because of onboarding friction, a crypto bias, token rugs, node centralization, spooky VC money, or poor stewardship. Farcaster ultimately failed because of LinkedIn-level PGness in the feed. Everyone here (myself included) was just too polite, polished, and milquetoast (which is not the same as mid). We didn’t have enough open controversies and spicy takes on sex, drugs, crime, morality, religion, wars, pop culture, and your mom. Even the occasional outrage tended to be collective instead of factional, because we were too neighborly to turn on one another. With a few exceptions, we’ve been casting to each other as if we might meet later that day at a VC luncheon in the Bay Area, and had a FID-reputation to maintain. To be clear, I’ve thoroughly enjoyed this very aspect of an intellectual and nuanced cozy corner of the internet. But it was a PvE experience, and PvE servers are stuck with low player counts

19 replies

9 recasts

75 reactions

Back to myspace I guess

3 replies

1 recast

17 reactions

My autocorrect now recommends “hodling” as the default syntax I may have been in crypto for too long

1 reply

0 recast

13 reactions

It’s dawning on me that markets can remain irrational longer than I can stay alive Maybe the grandkids will get to enjoy the Lambo

1 reply

0 recast

28 reactions

hello so the hilbert–pólya conjecture is true thus proving the riemann hypothesis, Fable and i found a self-adjoint operator whose eigenvalues are the riemann zeros i have a truly marvelous demonstration of this which this cast is too narrow to contain

2 replies

0 recast

5 reactions

I visited Cuba nearly 20 years ago; I wanted to see the country for myself before American capitalism rolled in on the back of a military takeover and changed the landscape forever. Back then, I assumed that scenario was about five years away. I guess I was fifteen years early.

0 reply

0 recast

18 reactions