breatheoften

I use icloud and this does not match my experience ... apple mail search has been insufficient on ios and macos for 15 years or more at this point ...

The cat is out of the bag. At this point it's pretty clear that the path to (meaningful) self improvement is almost certainly not subject to meaningful centralized control -- by "meaningful" I just mean the degree to which anyone has achieved it, that capability is 100% replicable and the cost for replication of that capability goes down in the future from now. Full stop -- can't undo.

i wish they had renamed chatgpt to codex instead of the other way around ...

The "program produces this trace, reproduce it after changing xyz" loop is imo -- a kind of programming primitive for the current agent capabilities ... I've found technique like that really effective as well.

It's interesting because there's a part of me that sometimes thinks "hey look this pattern is pretty effective -- I wonder if this a nascent abstraction on the path toward reasoning about how to use these tools in effective ways" -- while another part of me thinks "six months from now, you won't ever have to do this or if this is a useful technique the agent will just apply it on its own when relevant" ...

That's absolutely insane -- how the hell did anyone think that was a good idea under these circumstances ... This is an absolute scam -- why does it make sense for an index's buying volume requirement to be based on the implied value of shares that can't even be bought or sold ...?

"For purposes of mine removal" -- what the hell is that line about ...

Man that conversation history navigation for the new Siri app looks super unuseable ... how the hell am I supposed to actually find the conversation I want with the super-dynamic non-ordered 2-column offset-row view thing ...?

It looks hard to use ...

Also the 'floating semi-window but not a window' thing when using contextual siri in the context of some other app ... sure looks like it won't work with cmd-tab navigation ... I really hope is not the case ...

Is there anything structurally to prevent a super wealthy buyer from guaranteeing this essentially -- moving money and debt around in order to essentially put a floor under the stock and guarantee buy rates until the retirement fund index purchasers have time to absorb all the shares at the artificially high price? Feels to me like this is almost guaranteed to happen -- ...

I only just realized the z in foobaz stood for zork

This is extremely what Ive been wanting -- I had previously thought about using one of the hackish apps that try to deliver this experience - or spinning up something for this myself ... - but integrating this directly is definitely the right way to provide the best system and product experience -- and this seems to work out of the box exactly as I would want!

This is super nice!!!!!!

I don't think any of these model runtimes have really caught up with the architectures that are going to make the most sense for the capabilities of the coding models we'll have "6-12 months from now"

Imo -- we'll need the weights and abstract description layer for the model computation graphs -- but these little jinja2 programs and the various different structured grammar mechanisms (llguidance type complexity) and possibly the entire library of "model graph execution runtime" should all just go away in favor of ahead of time model -> hardware specific purely inline generated e2e implementation ... The documentation for a new model will just be a few example input / output interactions that _demonstrate_ the model io encoding protocol -- to the extent the jinja2 type setups actually work and generalize its all convention based anyway and a set of appropriate examples + "logic based" convention around the set of ways that make sense to interpret those examples should be able map to describe a deterministic mapping of that protocol in the only ways that make sense sufficient to cover all the variations in the protocols we've seen to date anyway ... a little jinja2 here or there doesn't actually clarify things more than that would really as there's still a convention associated with interpreting the jinja that's needlessly limiting

"hey gpt 5.6/mythos2 -- I want to run this model on iphone as fast as humanly possible" -> "no problem I see that model contains 6 signatures, there's an embedder, a per layer embedder, a decoder, and a few variants of prefill for fast prompt processing -- do you know the size distribution of the prompts? i can include support for all the prefill size variants or if you want to optimize for smallest possible distribution -- i'll assume a max prompt size and exclude the code needed for the larger prefill sizes? Connect a few devices if you don't mind or i'll attach virtual hardware when it's available and optimize the solution e2e -- i'll synthesize a performance and accuracy benchmark and then perform a search over that benchmark set for the fastest model loading and inference implementations for each hardware included in the evaluation matrix. The longer it runs the more confidence we should have in speed and robustness of the setup. I can stop automatically when it looks like things are good or continue the search for a fixed amount of time if you like..."

I've been dealing with various edge device runtimes lately and already running a sort of "worse" version of the above process which is imo -- mainly mostly limited by the need to "take advantage" of the "annoyingly shaped model runtime abstraction layers of old". I'm almost positive the whole setup would already be workable with when driven by existing frontier models if using model runtimes that were just "big piles of example code" which the model and i could draw from and modify l/inline into place as needed aa vs the current shape of "sort of semi general purpose computation graphs but with deeply non-observable error mechanisms and sources of ambiguity" ...

To offer the opposite anecdotal evidence point -- claude scrolls to the top of the chat history almost capriciously often (more often than not) for me using iterm on tahoe

Like so many things -- the evolution of AI math will I think follow trajectories hinted at in the 90s by the all time great sci-fi author Greg Egan. The nature of math won't change -- but the why of it definitely will. Egan imagined a future ai civilization in Diaspora where "math discovery" -- by nature in the future perhaps accurately described as "mechanistic math discovery" is modeled by society as a kind of salt mine environment in which you can dig for arbitrarily long amounts of time and find new nuggets. The nuggets themselves have a kind of "pure value" as mathematical objects even if they might not have any knowable value outside the mines. Some personalities were interested in and valued the nuggets for their own sake while others didn't but recognized that there were occasionally nuggets found in the mind that had broader appeal.

Research institutes like those founded by Terence Tao in our current present feel like they will align to this future almost perfectly on a long enough timeline -- tho I think on a shorter timeline this area of research is almost certain to provide a ton of useful ways to advance our current ai systems as our current systems are still in a state where literally anything that can generate new information that is "accurate" in some way -- like our current theorem prover engines are enormously valuable parts of our still manually curated training loops.

I really like this style of testing -- code that can be tested this way is also the most fun kind of code to work with and the most likely to behave predictably.

I love determinism and plain old data.

what language are we talking about? If it's cpp then the pronounciation depends on compiler flags (perhaps inferred from CMAKE_BUILD_TYPE)

Per-minute pricing for self-hosted runners seems like a very fast way for them to force everyone who actually is using self-hosted runners to migrate away.

I suspect we'll be doing that sometime in January or February.

I guess forgejo is the easiest migration path? https://forgejo.org/

I'm more and more convinced of the importance of this.

There is a very interesting thing happening right now where the "llm over promisers" are incentivized to over promise for all the normal reasons -- but ALSO to create the perception that the "next/soon" breakthrough is only going to be applicable when run on huge cloud infra such that running locally is never going to be all that useful ... I tend to think that will prove wildly wrong and that we will very soon arrive at a world where state of art LLM workloads should be expected to be massively more efficiently runnable than they currently are -- to the point of not even being the bottleneck of the workflows that use these components. Additionally these workloads will be viable to run locally on common current_year consumer level hardware ...

"llm is about to be general intelligence and sufficient llm can never run locally" is a highly highly temporary state that should soon be falsifiable imo. I don't think the llm part of the "ai computation" will be the perf bottleneck for long.

Is there any way to implement structured concurrency on top of the std.Io primitive?

What makes a NATS client implementation the right prototype from which to extract a generic async framework layer?

This looks interesting but I'm not familiar with NATS

It takes more than just misaligned incentives to get a banking crisis -- you have to have structural corruption preventing the transfer of the loss gradient back to the "misaligned" decision makers. It's somewhat disingenuous (or overly innocent) to reimagine the pathways which power structural corruption as "innocent ignorance in the face of bad incentives".

The real world has "actually bad" actors -- not just misaligned incentives.

In our ci setting up the docker buildx driver to use the artifact registry pull through cache involves (apparently) an auth transaction to dockerhub which fails out

The only figures of note that were assassinated that i can think of were more lefty -- or at least non right -- jfk, mlk, harvey milk, bobby kennedy, malcolm x -- were there actually any prominent american right wing figures assassinated in this "period of escalated political assassinations ...?"

Kirk's murder itself made the front page ... so we clearly don't always ignore politically adjacent topics here

Personally i cancelled my disney+ subscription for this (paid for via verizon) and made sure to explicitly say why. I encourage others to do the same.

I don't care about Jimmey Kimmel's jokes nor do I watch his show with any regularity -- but I sure as hell care about his right to make jokes.

Why the heck is this 600+ comment thread not on the front page?

Can there ever be a "living" standard to replace posix ...? Something that captures the good things about posix but allows all the related ecosystems to move everyone forward toward a better base specification (together rather than just increasing the number of bsd vs gnu special cases that with current plans will just have to be lived with forever)

Browser standards managed to do this in a lot of ways despite far more complex standards, more complex variations in behavior, and much more rapid continue evolution ...

The problem with python's async is asyncio ...

Structured concurrency libraries like anyio or trio are actually pretty nice -- "stacks" and stack traces are good things. Python multi exception concept is weird --- but also I think probably good ish.

It is still a pita to orchestrate around the gil/how terrible python multiprocessing side effects are wherever cpu bound workloads actually exist ...

Hilarious . I was a python fan before I was a reluctant user. It still has the most reliable debugger integration of any mainstream language which is my favorite thing about the language in its current state.

Debugger >> language -- next most popular language manifesto slogan (i hope).