A_D_E_P_T

Google could have been #1 for a little while, but they botched their release and squandered their temporary lead. (As usual.) If I worked there, I'd be so pissed.

These systems are too large and too chaotic to predict well, at least with human minds and with AI at its current level of sophistication. The latter also suffer from various biases -- the most obvious being the bias against surprising developments -- which hinder them.

Asimov's psychohistory needs something much more powerful than we have yet built.

So it seems to me that you can get interesting "status quo extended" scenarios, but as actual models they are near useless.

Last Tuesday the GOP controlled the house, the senate, the executive branch, the supreme court, and a lot of state governments that have flipped and will soon be under new occupancy. If the 2026 election was fraudulent, wtf are they even doing? How did they let that get past them? lol.

I'd give it a few months before they scan every photo/video you ever take for other purposes. If you were at the wrong public gathering, I don't know if they'll even bother with parallel construction.

> I think people underestimate the knife edge China is on

What do you mean by this?

Have you tried Xiaomi's MiMo v2.6? It's remarkably good -- and very weird/idiosyncratic. (For instance, it can compose MIDI-style music directly via the chat app, and it follows instructions much more closely than any other music model. AFAIK nothing else does this.)

I haven't had much experience with ERNIE, but I've been using various different (Chinese) base models of late, and I'm excited to try it.

This is awesome. ERNIE seems more powerful than Qwen3.8-27B and Deepseek-v4-Flash, which AFAIK are currently the best base models... it'll be very interesting to experiment with when it hits OpenCode. Base models are also excellent for businesses who run models locally, as they can be trained in various house rules and house styles.

Come on. If you have a Claude Max 20x plan ($200/month) you're massively subsidized and your API cost is anywhere from $5000-14000 per month.

(Figures vary, see e.g. https://www.linkedin.com/posts/christham_4-weeks-on-claude-2... for $13,700/mo.)

If you hit your limits on two Max 20x plans, you could very well be exceeding $300k/yr in token use at current API rates.

The dictionary definition says that knowledge is information gained by study -- i.e., in the very vast majority of cases, by being told. What's your definition? Come on, humor me.

The rest of your post is excessively emotional and off-target.

An infallible oracle tells Archimedes that the speed of light in vacuum is ~1.08 billion kilometers per hour. Archimedes, with the iron age tools and methods at his disposal, might be able to verify this and make some use of it himself. Maybe. In any case, what he has gained is knowledge itself.

The same goes for our own time. A powerful AI can tell us things at the edge of what we can grasp -- or beyond what we can grasp -- and it would nevertheless be knowledge that we possess. We can work to verify it, we can have it explained to us as though we were schoolchildren, or we can simply accept it as a static fact of the natural world.

In truth, we're going to have to learn to coexist with minds that are more powerful than ours. At least in some respects, if not in all respects. The good scenario is that they share information with us, and we receive it however we can. "Head in the sand" and "mutual disdain" are bad outcomes.

Oh, it's going to move up the old ladder, from bottom to top:

> https://en.wikipedia.org/wiki/Hierarchy_of_the_sciences

First math and CS, where formal logic dominate, and AI can check/verify its own answers very quickly. Then physics, downstream of math. Then materials science, which in some cases is practically a branch of physics. Then chemistry more generally. Then biology. Then various social sciences.

> This is not obvious to me at all. LLMs are capable of amazing feats of reverse engineering but are terrible at explaining and abstracting things well.

It's amazing to me that language models are still such poor writers and communicators. The mathematics community wrestled with this recently, when many solutions dropped into their laps but were not explained very well...

Thing is, (1) generic LLMs will improve, and (2) their lack of ability in this area has a lot to do with the rhetorical and structural habits they pick up during RLHF/instruction-tuning, which trained them for general-purpose comms. Zero-training base models can presumably be trained specifically to communicate scientific concepts. Interesting times.

(Edited to add: With respect to (1), compare Opus 5 to 5.5. 5 was a terrible communicator, even of software/coding concepts, whereas 5.5 is excellent and leaves very little to be desired.)

The article's perspective reminds me of math class when I was a kid. The teacher didn't want the right answer alone, she wanted to see how I got there. In an absolute sense, the right answer is the right answer with or without my understanding, but that's not how the class was graded. We're approaching a point where AI is to science as my old calculator is to elementary school math -- it's capable of leapfrogging to answers faster and more accurately than we can.

Thing is, though, verification is knowledge itself. And if we can get to a point where the important questions have answers, we can surely have the verifiers explain those to us in ways we can understand them.

You don't need to wait that long. Data from 2025 is already available. Hunter isn't in the top 100, and Trucker and Killer (lol) aren't in the top 1000. Those are two exceptionally weird and unfortunate names.

> https://www.whattoexpect.com/baby-names/list/top-baby-names-...

Liam and Mateo aside, the top 20 could have been pulled from 1840. Tradition still reigns, and Victorian-era names for boys are the rule rather than the exception.

Liam, the #1, is rather weird -- kind of modern and pan-ethnic -- and there are a lot of low-class-coded names like Ronin, Odin, Cash, Royal, etc. These tend to be confusing headscratchers more than they are aggressive, though.

Names that end with -er are in general quite rare. That said, it's true that Archer is in the top 100, and Gunner is somewhere around #500.

It's easier to imagine Google doing it. It's equally likely to imagine OpenAI (equally resourced) or Deepseek (pushing novel optimizations) doing it.

Yes, and Anthropic doesn't have that ability and has to lease data centers from AWS and Google!

> Dario already said it according to multiple sources: he believes Anthropic will be the last surviving company in the world, where all the value will accumulate

Given that LLMs are practically commodities and open models are a dime a dozen, this is pure insanity. There's no way there from here.

It's WAY off, lol

> https://bullfincher.io/companies/amazoncom/revenue-by-segmen...

Well over 60% of their revenue is B2C, strictly defined.

To be pedantic about it, CS =/= software engineering. CS is a scientific field that's more closely related to math or analytic philosophy. It concerns itself with what is computationally possible, with what is tractable, with algorithm design principles, and broadly with information theory.

The scientific field doesn't have that many open problems -- whether P=NP is the greatest and most famous -- and they're set to fall like dominoes. A condition where CS is "solved" and has no major open problems is a very real near-term possibility, and probably a medium-term inevitability.

Though software engineering is a fundamentally different discipline, none of this implies that software development isn't in trouble. Human planners and software architects might remain, but very soon LLMs are going to write virtually all code.

The problem with biotech/pharma is that data processing doesn't work if you lack the data to begin with. There are too many holes in science's understanding of how the cell works, how the organism ages, etc., and LLMs can't easily fill these in. Hypotheses can and will be generated, but it will take great lengths of time and comparatively heroic efforts before the picture becomes much clearer.

The contrast with math and CS couldn't be starker, really. Everything that hinges on formal logic or logical relations is done for.

I don't think that it's primarily about jobs. At this rate, CS is going to be the first field to be "solved." By 2030 we'll probably know whether P=NP (like many others, I'm starting to suspect that it does,) and just about every other open problem shall fall, so everything that seemed important to us from first principles will be known.

> https://en.wikipedia.org/wiki/List_of_unsolved_problems_in_c...

Then advances in algorithmic efficacy will be the province of machines, who will act in ways and at speeds that are beyond human.

What room, then, for CS students?

I wonder if biology and theology are picking up students. Those fields, I'd bet, won't be solved within our lifetimes.

1986.

Also, this would imply that people who work in glass and zeolite production/processing would be more at risk, which I don't believe is the case, so it's unlikely that this hypothesis holds up.

Gemini 4-Argon is very good at coding. I prefer Opus 5.5 -- in large part because I've grown accustomed to Claude Code -- but Argon is undeniably strong.

4-Carbon is, per benchmarks, said to be superior to Opus 5.5 at coding.

At general tasks -- search, medical, interfacing with maps and mail, planning, etc. -- the Google models are already superior.

It's a memestock.

Think about it. Google's latest Gemini AI is just behind Anthropic's best offering -- or, in fact, it may be slightly superior. Google also has dozens of other mature, cash-cow businesses, massive infrastructure, their own hardware accelerators, an army of smart technical staff... and leadership that, for all its faults, isn't a literal cult.

Anthropic has a top-tier LLM among other top-tier LLMs. Claude is great, but not unique. Anthropic has effectively no other assets, many genuine liabilities, a downright bizarre wet-lab venture, and leadership that's... well...

So, the way I see it, either Google is massively undervalued, or Anthropic's IPO is massively overvalued. Am I somehow off base here?

This naming strategy is bizarre.

There are no good elements for D; Dysprosium is the only one, and it's a mouthful. There are no good elements for E; Europium is the only one and would seem political. Is Fluorine their next release? (And then Gallium or Gadolinium?)

After Carbon, they should just update to Gemini 4.1 or 4.5 or whatever, and drop the random unrelated elements.

Those .si domains his cronies are squatting on aren't selling, I guess.

There's actually rock-solid proof of this in the article:

> So, imagine my disappointment when I walked up to my first haunted attraction of the year and was immediately greeted by low‑quality AI art. One of the images was so bad that it even included the Gemini diamond in the lower‑right corner of the page.

The image in question is a sign or menu, the sort of thing that's usually neither good nor bad. Here it's exceptionally bad because it "included the Gemini diamond." In other words, it's exceptionally bad because it's AI, and for no other reason.

At this point, I don't know if it has much to do with quality...

Right now, even with infinite money, you have no way to extract helium from Jupiter and return it to Earth. (To say nothing of the Sun!)

Maybe with a couple hundred more years of technological development...

Our moon itself is said to have ~1M tons of very valuable Helium-3 in surface deposits, but even that's nigh impossible to harvest commercially right now.

Wow. If those charts are true, this effectively matches GPT-6-Astra at a fraction of the cost, and is neck-and-neck with 6.1-Sol.

Sure, but once they release an open model, it's open forever. You can download it, share it with friends, make a torrent, whatever. So in this sense every open model is a gift to the commons.

It's not quite like Uber. It's more like... well... imagine there are two computer operating systems, Gray and Teal. Gray is always slightly better, but it costs $50/user/month for a license. Teal is free. Eventually, when both happen to be in version 9.0, Teal puts Gray out of business. Starting from version 10.0, Teal starts charging $100/user/month from a license. But nothing's stopping you from sticking with 9.0, which remains free forever.

Perhaps what's "sustainable" is not clear, given that Gray was eventually dominated by Teal? In any case, versions 1.0-9.0 were, effectively, gifted to the public. We can appreciate (and even disapprove of) that commercial strategy while benefitting from its fruits.