fritzo

If distillation preserves an LLMs soul, then distillation preserves the human souls on which LLMs are trained, and we hn commenters are already immortal, right?

Whoa that's a slippery slope! Next you'll want model runners to cite the data their models were trained on

Sounds reasonable. Is there a good benchmark on which make this decision?

I dunno, seems like Amazon the forest is the real hyperscaler

Just wait till the Atlantic Meridional Overturning Circulation halts, then you can grow icicles!

The alternative is 4 years of house arrest, just until the next administration can issue a pardon. There is no sentence between 4 years and capital punishment.

I doubt you ever understood the solid state physics, semiconductor fabrication processes, supply chain logistics, monetary policy, shipping routes, mining engineering, etc. "Knowing how things work" is a stone-age attitude.

Feature request: fewer clicks. It should be one click per question

By "sizable rock" do you mean large pebble or small boulder?

What has "artificial" to do with it? Human intelligence is also unauthorized unconscious plagiarism.

Dependency bloat and dependency bitrot have made solutions less permanent, have increased the maintenance burden. My ancient projects with zero dependencies still stand. But projects I built on shifting dependencies are rotting and cracking.

Looks to me like a mob of humans, angry they've been deceived by ambiguous communications, product nerfing, surprisingly low usage limits, and an appallingly sycophantic overconfident coding agent

ln -s CLAUDE.md AGENTS.md

There's your one line change.

If it's a poor location for photovoltaics, it's exactly as a poor for photosynthesis

Sorry for my ignorance, but what exactly is the distinction between hn and social media? Is it the personalization that distinguishes the two? Does "social" mean "feed depends on graph neighborhood"? So collaborative filtering + ranking algorithms + moderation is not social media until you add graph neighborhoods?

I've also seen a glue-less paper binding trick where two pieces of paper are finely crimped together with some high pressure tool in alternating v^v^v^ patterns, actually making tiny tears in the paper. Does anyone know what kind of tool does that?

ELI5 what is a harness?

EDIT from https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf:

> We seek to fight two forms of overfitting that would muddy public sensefinding:

> Task-specific overfitting. This includes any agent that is created with knowledge of public ARC-AGI-3 environments, subsequently being evaluated on the same environments. It could be either directly trained on these environments, or using a harness that is handcrafted or specifically configured by someone with knowledge of the public environments.

Sounds like her first thought was, "I'm talking to a manic guy, and I can use him to make money"

Wait, is 474 a number or a proper noun?

They're private, that's the beauty. Code is so cheap now, we can ween ourselves off massive dependency chains.

200 years ago text was much more expensive, and more people memorized sayings and poems and quotations. Now text is cheap, and we rarely quote.

Those business goals will soon realize they need more electricity. More brains will be devoted to power generation.

The same could be asked about people. The answer is social intelligence.

By that logic, best fridge is no fridge at all ;)

Thank you, I was so confused by the obvious clickbait! I love Grady's videos, but yikes.

> optimum properties for estimating a posterior distribution

Circular reasoning: that's true only if the posterior is normal, or if your "optimal" is defined by second moments. In infinite variance cases, the best estimator can be median or an alpha moment for alpha < 2, but yikes the math is much more difficult.

-- A mathematician who has indeed fallen into the beauty trap

Heavy tails are everywhere. Normal distributions have absurdly light tails. Levy alpha stable distributions have power law tails. Power law tails are everywhere.

Some things with heavy tails:

  token occurrences
  comment thread upvotes
  startup IPOs
  social follower counts
  network latency
  github stars
  git diffs
  power station size
  weather events

Right. And the CLT is not actually limited to normal distributions. Both of the distribution families I mentioned are central limit theorems. The CLT we first see in school regards means of finite variance distributions, where the finite variance assumption is made because it makes the math easier.

https://en.wikipedia.org/wiki/Central_limit_theorem#The_gene...

Hot take: bell curves are everywhere exactly because the math is simple.

The causal chain is: the math is simple -> teachers teach simple things -> students learn what they're taught -> we see the world in terms of concepts we've learned.

The central limit theorem generalizes beyond simple math to hard math: Levy alpha stable distributions when variance is not finite, the Fisher-Tippett-Gnedenko theorem and Gumbel/Fréchet/Weibull distributions regarding extreme values. Those curves are also everwhere, but we don't see them because we weren't taught them because the math is tough.

Is AlphaZero actually machine learning, if there is no data, or the data is entirely synthetic? If so, we must admit that deterministic automated proof search is machine learning. Shannon's and Kolmogorov's accounts of information leave us stranded. Finzi et al. propose to resolve our confusion.

I was hoping this would be an infinite length terminal view, like the opening backstory in Star Wars