a_wild_dandan

That retcon was bizarre. There are myriad simple pretexts for enslaving humanity. Choosing "let's maintain lukewarm meat bags for energy" is so contrived and absurd.

Using either frontier model for basic edge device problems is wasteful. Use something cheap. We're asking "is there a profitable niche between the best & runner-up models?" I believe so.

It almost never is. While HN contrarians endlessly complain about it, the rest of the world happily ships Electron. Discontent with Electron is so comically over represented here that I'm beginning to think it's an insecurity or complex or something.

So...reinforcement learning?

Just use supervised testing. It's scalable, battle tested, and pervasive. Even ignoring scalability, orals have problems. KISS!

> - Pressure landorders who rented under the suggested price

How did RealPage do this? I'm OOTL.

Exactly! Have you tried the Phi models? To me, they indicate that we can get much more efficient models. In a few years, 70b on gold standard synthetic data + RL might run circles around SotA. It's such an exciting time to be alive.

I run q5s usually, since it's a 40% haircut on model size, with nearly no PPL loss. (Presuming an 8b native model like Qwen.)

Yeah, I would want the 192GB Mac for attempting such hefty models. But I have such basic bitch needs that 405B is overkill haha.

For 70B models, I usually get 15-25 t/s on my laptop. Obviously that heavily depends on which quant, context length, etc. I usually roll with q5s, since the loss is so minuscule.

70B+ models typically run great with my MacBook's 96GB of (V)RAM. I want a Mac Studio to run e.g. llama-405B, but I can't justify the marginal model quality ROI for like $7k or whatever. (But I waaant iiit!)

We don't. That person has no idea what they're talking about. Virtually no AI researcher would require "neuron simulations" for intelligence, nor agree with their AGI claims. (See: recent machine learning expert surveys, for example.)

Use HN comments for comic relief. When you see one worth taking seriously, you'll know it. Otherwise we're just giving oxygen to contrarians.

What’s normal? On your Apple silicon.

Hence their qualifier “rarity.”

My sheer productivity boost from these models is miraculous. It's like upgrading from a text editor to a powerful IDE. I've saved a mountain of hours just by removing tedious time sinks -- one-off language syntax, remembering patterns for some framework, migrating code, etc. And this boost applies to nearly all of my knowledge work.

Then I see contrarians claiming that LLMs are literally never useful for anyone, and I get "don't believe your lying eyes" vibes. At this point, such sentiments feel either willfully ignorant, or said in bad faith. It's wild.

> You are still responsible for what you do regardless of the means you used to do it. And a lot of people use this not because it’s more productive but because it requires less effort and less thought because those are the hard bits.

Yeah, that's...the whole point of tools. They reduce effort. And they don't shift your responsibility. For many of us, LLMs are overwhelmingly worth the tradeoffs. If your experience differs, then it's unfortunate, and I hate that for you. Don't use 'em!

This tests vision, not intelligence. A reasoning test dependent on noisy information is borderline useless.

I mean, I passed a physics PhD exam and I can’t model the Empire State Building. The jury is still out on whether I’m an intelligence tho.

Sounds like a good job for a new nonprofit named Black Mirror or something.

You could try zx[1]! I prefer it to Bash scripting.

[1] https://github.com/google/zx

I wonder if using nested markdown bullet points would help. You would preserve the information hierarchy, and LLMs are phenomenal with (and often output) markdown.

Let’s start calling these stories FAErytales.

> But how do you reason?

My theory: reasoning is the application of analogies. Here `analogy = memorization + pattern matching`. Pattern matching is just an associative memory query; remembering an example with fuzzy enough details to apply generally. Analogies are self-similar/recursive and sometimes transcend contexts -- they're useful for thinking, for thinking about thinking, etc. Analogies pervade our language, our thoughts, our approach to novel problems, and the cached solutions to trivial/practiced problems.

There is no qualitative distinction between System 1 & System 2 thinking. It's a spectrum, quantitative. The more analogy steps required, the more thinking done, and the more System 2-ish that thoughts feel. This (probably wrong) theory has many consequences:

1. LLMs are truly intelligent, albeit barely. They rely heavily on memorization rather than much analogy matching. Their depth of memorized knowledge can somewhat substitute/mislead us to misjudge that intelligence.

2. The validation set grokking phenomenon occurs when a new analogy is internalized.

3. Scale really is enough. The scaling laws will continue holding with astonishing accuracy.

4. MCTS heuristics are useful inductive biases to vastly lower compute at the expense of model flexibility. Eventually, The Bitter Lesson will come for MCTS too.

I have much more to say about this bonkers theory, but I fear already sounding like a ranting madman. Anyway, just my 2c.

We're unsure how minds or LLMs work. So let's also not dismiss potential parallels either. It's okay to not know stuff. We'll get there!

You're blaming TypeScript for self-inflicted wounds.

Don't blame the type system that you banished (with `as unknown as T`) for not catching you; or for some React library having bugs, e.g. an incorrect interface; for not defining types and Pikachu-facing when types are `undefined`. These traumas are fixed by using TypeScript, not ignoring it.

Using TypeScript for the entire stack feels like a superpower. The type system is incredible. V8 is fast. The frameworks are phenomenal (Next.js, Material UI, etc). The ecosystem is enormous, with packages for everything. The unified codebases save gobs of duplicate code (e.g. ferrying data betwixt client/server). I'm not surprised that such an expressive system can play Sudoku!

By "Colorado as a whole", do you mean "Boulder in particular"?

That doesn’t seem anywhere remotely near obvious to me, but that might say more about my policy ignorance than reality.

There is a "no downstream dependencies" option; it's called writing/auditing everything yourself. Everything else -- be it libraries, monolithic SaaS platforms, a coworker's PR, etc. -- is a trade off between your time and your trust. Past that, we're all just playing musical chairs with where to place that trust. There's no right answer.

Just a simple 2-queue. I love this, since it's low friction and your essentials are persistent. No more brain-wracking to remember what the grocery list is missing.