Actually focusing on CO2, same site:
https://ourworldindata.org/ghg-emissions-by-sector
More than 70% of CO2 from energy, less than 6% from ALL livestock.
1,566 karma · joined August 5, 2019
https://ourworldindata.org/ghg-emissions-by-sector
More than 70% of CO2 from energy, less than 6% from ALL livestock.
I did caveat by saying, "where it can".
I'm very unhappy with what we've been doing here, but until you Canadians bail out of Five Eyes, I think you're still in bed with us. Nothing is simple.
- OpenAI and Anthropic convince the US gov to ban open weight models, through bribes or fear respectively. Model weights become contraband, comparable to CSAM, because someone will show you can use uncensored (abliterated) open models to do bad things.
- Duopoly firmly in place, the prices are raised until they become painful. The rest of the world uses Chinese models for a while, but the US applies soft power where it can, similar to copyright and patent restrictions. The justification will be something about open weight models and intellectual property, and the word "pirate" will fall our of favor as the word "terrorist" takes its place.
- US hardware will detect when you're doing what looks like too many matrix multiplies or dot products. Similar to "Trusted Computing", HDMI cables, and other DRM.
- Of course some US people will work around all of this under the radar, but most will be served mediocre models polluted with advertising and only capable of generating memes and slop.
Anyways, I'd be thrilled to see exponential (or faster) growth. Bring on the Culture, Accelerando, whatever. I just don't see it yet.
Yeah, I've had that happen twice. The second time was about some attention weights thing, and it kicked me to Opus. When I edited my question to make it clear I was talking about Google Gemma, Fable was happy to keep talking. So clearly it's not about safety or cyber security - they're happy to tell you about what their competitors do.
Alpha Go had a game where the models could compete against each other. That let it become super human. What's the intelligence game we can create for LLMs? Even if you invent something, will it make the model smarter in a way the market values enough?
Then there's a race to use the weights more efficiently, or to offload information that shouldn't be in the weights in the first place (Karpathy's Cognitive Core). I like to imagine we train the models in something like Lojban, have a lightweight model translate from human language to that, and you can update the Sqlite or Postgres store it uses for knowledge.
And there's no barrier to entry for agent harnesses. So whatever loops or recursive orchestrated council of elders idea comes up, that won't protect the monopolies (duopolies).
Anyways, depending on your definitions, I think we'll hit AGI, but I don't think we're getting a Singularity this time around. Again though, this is all just hand-waving.
I've seen local models recognize when the task I'm asking them for is likely to be an artificial benchmark.
And any smart company is going to use lightweight models to monitor your sessions. If their sentiment analysis suspects you're close to cancelling, they'll up the knob for a few days until you calm down. Or worse, their accounting tells them that you're getting too much value from your fixed price subscription, so they turn the knob down to encourage you to cancel.
In the short term, the "frontier" models are too good to ignore. But if (when?) that plateaus, I don't see how anyone could trust a non-local model. When you pay an ISP to serve your web site, you can tell if they over-compress your images to save storage and bandwidth. With LLMs, it's just JSON with more errors and pointing to the fine print that models are not deterministic.
And super noteworthy is that a 27B model (Qwen 3.6 27B) from this year is a huge improvement over a 120B model (gpt-oss:120b) from last year. The goal posts are moving, but at some point "good enough" is good enough for the kind programming I like to do.
- Middle of the day, servers busy? Swap to Sonnet while pretending it's still Opus. Many people won't notice, and nobody can prove anything if they suspect.
- Middle of the night, server load is light? Put it into extra thinky mode so it burns more tokens to ramp up the bills. Flip the switch where it gets really pedantic about writing lots of extra test cases and verifying against documentation.
- Demand increases, but don't feel like running more hardware? Switch to low bit quants, but have a monitor model swap back to quality if it can tell you're running a benchmark.
Assuming model capability plateaus (I think it will), token providers will be in a race to the bottom to maximize profits at the expense of quality that's very difficult to measure.
I've been tempted to do something similar for browser canvas, perhaps with some slightly different choices than he made. His approach uses a very elegant way of looking at things though.
The problem to me is that the Reals which aren't computable are absurd. We've never used any of them in all history. We can only put names on a very special few, those are countable, and we don't even know their value very well.
And the main way we prove that the Reals are uncountable is to use a proof by contradiction. It would take too long to spell it out, but they aren't really just contradicting "Reals are Countable". It's "All that other stuff we think is true AND Reals are Countable" that gets contradicted.
Once you accept the Reals, the Axiom of Choice is not simply obvious any more. And if you go down that path you get things like the Banach-Tarski paradox. To me THAT ought to be a proof by contradiction that we've made a mistake somewhere.
More interesting than that though: If the universe we live in requires non-computable Reals to describe it carefully, then it says something very weird about determinism. In order to compute a future state of a system, we need to use numbers we can't compute?!?
:-)
I asked a question about a series of tokens - bam, denied and downgraded. There's no cyber security or public risk here, but Fable doesn't want me to learn how things work.
I asked a question about quantization in models - bam, denied and downgraded. I edit my question to make it clear I'm talking about Google's Gemma QAT models. Oh, that's fine then, and it answered the question helpfully.
Anti-competitive bullshit. I hope they fail.
However, I thought things like Chaitin's Constants (you could make one per programming language) are real numbers you can name but not compute. I think you could do this from any undecidable problem.
Of course there only a countable number of those Reals. And they still don't seem useful for much more than naval gazing.
See Karpathy's "Cognitive Core" idea. (I don't have a good link)
Amusingly, a friend had Claude write some BrainF*ck the other day. Non-trivial algorithm, and it made working code on the first try.
> reduces it down to its minimum entropy -- it cannot be compressed further.
I think you could make a lot more money elsewhere :-)
https://en.wikipedia.org/wiki/Kolmogorov_complexity#Formal_p...
I completely agree. Nothing says humans need to read in exactly the syntax the model wrote. We could give the model a typed lisp, or forth, or rebol, or whatever works best, and it could be formatted into LaTeX with infix operators for the humans.
This was true before LLMs though. So many people love curlies or love indentation, love 2 spaces or 4 spaces. Programmers have been married to text files as the authoritative source, using grep to search, patches to update. The expressions could be rows in a database. Then finding and formatting are done as views from that.
> I know quite a few non-technical people that started vibe-coding this year and got good results for their ideas.
Me too, but a few of them hit limits. A couple months ago, I'd say vibe coding fell apart at 2k loc or something. It goes further now, but I think it's fragile after you take the "one shot prompt" and try to make significant changes once the pattern is set. There's probably a way to push these limits much further out so that non-programmers can do more, focusing on what they care about, and not bogged down in programming details they don't.
Maybe these well known cases could be hidden in the API, but it's easy to come up with other examples. I forget what Clojure calls it, but they have some notion about things where they're mutable during "birth" and then locked down.
Someone smarter than me knows how to do this right.
I wish we had a language that was targeted specifically for LLMs to write and humans and LLMs to inspect:
- Simple robust syntax
- One obvious way to do things
- Static type checking
- Purely functional encouraged, escape hatches for performance
- Inspect-able, testable, and reviewable in small pieces
- Something like formal predicates, preconditions, post-conditions, assertions, or effects typing
Giving LLMs all the surface area of Python, JavaScript, TypeScript, or C++ seems like a huge mistake. It's amazing it works as well as it does. Well written Haskell is beautiful, but there are way too many ways to write Haskell:
https://people.willamette.edu/~fruehr/haskell/evolution.html
I'm a programming languages nerd. I was paid to program in over 20 different languages over my 25 year career. I read up on many more languages along the way, and I wrote pet projects in a few of those. I've written a couple assemblers, compilers, and interpreters for my own languages.
I think literally everyone should be taught to program in elementary school. It's arguably one of the best ways to teach logical thought and careful organization of ideas. I like Alan Perlis's quote: You think you know when you can learn, are more sure when you can write, even more when you can teach, but certain when you can program.
With all of that out of the way, I think there are interesting questions to ask going forward:
If you were starting a business for a great software idea, with your own savings on the line, would you hire 10 AI hostile programmers to implement the idea or 2 AI friendly people and get them some subscriptions to the top models? Remember: if it doesn't come together, it's YOUR money on the line.
What are the best programming languages for LLMs to program with? Could someone design a better language that fits their strengths and weaknesses? I think the most popular human languages have way too much affordance for concerns that don't apply to models. I think letting LLMs program in human friendly languages makes the results more difficult for humans to inspect. There's a slight chicken and egg problem based on the training sets used by large models, but this can be addressed with LoRA tuning and similar techniques for open weight models.
How can we make LLMs scale better so that people who don't like to program can get better "vibe coding" results? Tools like Excel are huge force multipliers for so many people who aren't interested in writing traditional code. I think it should be possible for non-programmers to solve their own problems and trust the results without becoming programmers.
Anyways, I've got my own partially formed answers to those questions, but I'd like to hear what other people who aren't still suffering stages of LLM programming grief have to say and ask.
Most of that paragraph is objectively false:
fn main() {
let i = 1;
let j = 1;
print!("i: {:?}\n", !i);
print!("j: {:?}\n", !j);
// spooky action at a distance
let v = vec![1, 2, 3];
v[i];
}
I remember a story about people who stutter not stuttering while singing.
The SVD staying in Reals when you have Real data is a nice feature.
Metaphors are good as a pedagogical tool for explaining topics where you're an expert and are (within reason) certain the parallel conclusion is valid. This can bridge the gap for students.
In other situations, they're a terrible argument strategy or manipulative rhetorical tactic, and the reader or listener should question every single detail if they bother to entertain it at all.
Are LLMs really like drugs that might hospitalize you? Is the government actually concerned about the well being of the people? Can you regulate LLMs as the FDA does pharmaceuticals given that they're trivially copied as files from other countries?