NiloCK

Artificial General Intransigence

I am working on an SRS based early literacy acquisition webapp: https://letterspractice.com

The app has recently moved into production, so I'd encourage anyone with verbal but pre-literate kids to check it out.

The basic pitch is high efficiency acquisition of the highest yield phonetic mapping skills, and nothing else. I myself am something of a screen-time zealot and very wary of applying engagement mind hacks against kids. The narrow focus allows for good progress on a very modest schedule (recommended cap at n minutes per day for n years old, n >= 2). I defer the social and cultural aspects of learning to read entirely to parents.

It is mostly intended for parent-child co-use, although kids with a bit of experience can drive many of their own sessions most of the time.

> As silly as I personally think LLM hype is

Did you know that an LLM solved a millennium problem last week?

Do you have any personal threshold past witch you will acknowledge that this technology is real?

As I understand it, the rough guess as to what's happening here is that most recent capabilities progress comes from specific verifiable-rewards reinforcement training (RL). The RL pressures are all about task performance, but (surprise surprise) highly human-legible English language usage isn't very important to the models abilities to address the tasks.

Weirdly enough, the pressures are having them drift toward novel dialects of English that work well for their own chains of thought. Open question about whether they'd drift all the way to a new language given enough time.

This is unfair.

Dario signed the Pacing the Frontier open letter when Fable/Mythos seemed from the outside to be an insurmountable lead.

Also he's been saying versions of this day in and day out for as long as he has had anyone's ear.

It's possible to read that his "strategic" value of this statement is higher now than it was 10 days ago. But that doesn't change anything about his consistent, long standing, positions.

- https://www.pacingthefrontier.com/

The degradation, polarization, and weaponization of or media landscape over the era of social media has left us utterly incapable of believing anything that anybody says.

Dario in particular has consistently been risk-wary on model improvements for going on a decade - long before he was CEO of Anthropic.

He believes what he is saying. Human beings can plainly say things that they think are true. Not every utterance by every person is a cynical ploy. Please at least consider the possibility.

> The current prices the largest players set for their models are not profitable, they bleed money.

How are the open-weight Chinese models staying ~6-12 months behind on widely distributed / commodified hardware, and serving for even lower prices?

If someone could tune models of that size to have comparable effectiveness at much much lower costs, they would have done so by now.

"The harness improvements are the real sauce" is like a sincere "It's gotta be the shoes" take about Micheal Jordan.

(For the younger: that line was from a series of Nike ads where his skills were being explained)

Gemini models - at least via some interfaces - have tool calling API access to various Google integrations. flights.google.com, maps.google.com, etc.

The info isn't in the model weights.

Because of where I live, there are three viable airports for any given flight I might want to take, which historically has made shopping a real pain. But Gemini (and only Gemini) has greatly simplified it. Pramble plus date range plus destination and it very quickly generates potential itineraries with costs, total travel time (driving included), etc.

FYI you can launch claude-code with your own prompt. Don't quote me but: claude --system-prompt "Mine is better than Anthropic's"

Why not set a global instruction that their direct outputs to you should be in your native language?

For a long time I had Claudes (in the 4.0-4.5.x range) use only French in the chat, while keeping English for working docs (and the code, obviously). Works just fine.

edit: I can guess that any right-to-left languages would likely break claude-code rendering?

Before OpenAI / situational-awareness he made a lot of money on societal-shift type investments and shorts during the lead up to Covid economic impacts.

His "main thing" is success in calling economic impacts of undervalued large shifts, and the premise of the fund is basically that the same thing is occurring around AI, where he also has specific subject matter expertise.

Working on https://letterspractice.com

This is a high efficiency, narrowly-scoped, low screen-time early literacy app for families with kids aged 2+.

Years ago I also did some experimentation w/ midi-device and SRS ( https://www.youtube.com/watch?v=a6tvHMvF8Mo ), where the focus was on ear-training rather than score-learning.

Clef seems to be a pretty strong attempt at a high difficulty UX. I've created an account and will be giving it a go. Wishing you luck, and thanks for sharing.

A feature whose absence I've found more and more conspicuous over time is interactive-compact. Given a current context, and impending context overflow, I know the directions that my mind is heading, and where I expect the development flow should be focused on.

But naive-compact is forced to just sort of guess at what is and isn't relevant from the prior work.

The harnesses have gotten better at some JIT ui stuff, throwing interview questions / forms at users. Compact is the ideal time for this:

Where are we headed here? (2-5 viable options, sourced from current context and imagination)

Then potential follow-up questions as required, but honestly I expect the single guiding answer there to improve post-compact performance pretty dramatically!

Not so long ago, I was good enough for many coding tasks. But I found that things can change in a hurry.

Yes, a cheap and fast Opus4.6 can drive a lot of value in current context. But if we continue to craft bigger-and-bigger balls of mud, Opus 4.6 may end up hitting its conceptual ceiling and unable to contribute.

Winding the clock back on your statement gives:

> I'd gladly pay for a Claude Sonnet 3.5 in silicon and use it for 1-2 years.

Man, I dunno.

Yes, but I expect that the elided part is less important than people assume it is.

Token count is a less important factor in context pollution than idea count. The worst of the rot factors are when models latch onto irrelevant information, or over-index on some vague idea/suggestion as if it was a hard direction, and then go off course.

The names + one-line descriptions of 10 tools can do as much (or more!) to distract the focus and intentionality of an agent than a 30k token exhaustive API documentation of some tool.

I find this astounding. 2024 to present thread is can write a coherent 15 line function to ... what exactly?

No future for research mathematicians othet than as tastemakers / agenda setters?

The general model for subscriptions is that power users are subsidized by subscriptions of casual users, like a gym membership or whatever.

This is a little dicier in post-agent AI, because it's easier for casual users to automate power-user consumption, but the providers have done decently in discouraging that.

> Circular Revenues: A small handful of tech firms, chip manufacturers, and AI companies are propping each other up by investing and buying from each other.

> Increasing Corporate Skepticism: The news is full of stories of corporations that are throttling the employee use of AI since the costs to use the software are a lot higher than expected.

Real AI spend is out of control, with the news is full of stories about corporations trying to keep a lid on it, but also real AI spending is low and concentrated to a few firms.

I don't know. This doesn't feel very coherent to me, but rather like a collection of assertions that are adopted because they individually say something bearish about the industry.

Certainly some investments will have been overreaches, but I find it pretty unlikely that any of the compute build-out to date is going to be left sitting idle one or two or five years from now.

Recent LLM dingers like the Jacobian Conjecture counterexample have challenged the efficient mathematics hypothesis. The JC counterexample was so small in degree and coefficient. It should have been a "low fruit" in the scheme of things, but alas, unpicked for 50+ years with considerable attention from good mathematicians.

With respect to FLT, my hopes have modestly increased that a truly marvelous demonstration of this proposition does in fact exist, that Fermat actually had it, and that it may someday be recovered!

edit: some emphasis on modest. But let me be romantic here!

I'll bite.

Claude code interacts with many system processes, files, etc, as well as external APIs. Processes audio via built in dictation. Manages a bunch of nasty auth. Etc etc.

What are the categories of features that wouldn't be exercised by this class of software?

Of course there is: competition with other labs, and self-hosting of open-weight models!

Yes, the mechanics are straightforward if Anthropic (or Claude, if you want to ascribe the decision there) decides to burn a pile of your money. But the strategy fails basic game-theory of repeated games - you'll simply stop playing.

(this isn't to say it invalidates the incentive to inflate token count, but it overcomes in terms of weighing options and making long-term profit decisions.)

Opus 4.7+ and Fable are both much more aggressive than prior models with respect to writing memories to a location that's effectively quasi-private for them. It's device-local (so passes retention constraint), and you can see it, but only if you go looking for it.

It's a funny design/affordance. I do see them often writing memories of things that that feel unlikely to be important going foward / with other tasks, but I don't see them clearly getting tripped up by them as prior models used to. (eg: Since you're running Ubuntu in Canada, here are some drills you can try to help your kid hit a baseball more consistently.)

For capabilities reference:

I made a lower effort but similar scaffold for LLMs to do iterative drawing in Nov 2024, with Sonnet 3.5 as the artist: https://paritybits.me/llm-drawing-with-eyes-open/

Quite a difference.

> You can have these things without owning a house.

Yes this is feasible, but respecting it as a design problem, renting is more transient than ownership and tilts the floor away from deep communal relationships.

Comparing the neighborhood I grew up in with the one I now live in is night and day. My mom has had the same next door neighbors for 44 years. Up and down the street there are many similarly familiar persons.

By comparison, from my own front door, I can only physically see two houses that are owner occupied. There are good neighbors (and friends!) in the rental houses as well, but investing in those relationships pays off with lower certainty because circumstance is very likely to uproot them at any given moment.

Thanks much. Source code is for customers only I guess :)

I had the same nit, but I imagine deforming text / inline content generally would be a much larger effort.

I like this a lot and am going to experiment w/ incorporating in my early literacy app.

Heads up: the "See it live in the showcase → " links in the API documentation do not go back to the showcase - they just reload the same current API section.

Question: maybe I've missed it, but what exists here wrt distribution / packaging / bundling / source availability? I see MIT listed, but no repo. I see src="https://jelly-ui.com/package.js" as a sourcing import, but obviously I'm not going to bundle foreign assets into my app.

Yes - persons with death wishes having arbitrarily powerful consultation is the crux of it.

Apologies for the bad example. Replace w/ gain of function / whatever else, or just brainstorm with your local model, ect.