jvanderbot

This is a fundamental human need and it's being taken from everyone eventually, by design. It will take a generation to turn over and find new ways of being useful. Those who ran into this wall and could not safely reduce their income and retire/pivot to other fields, will be the coal miner oxy crisis of this generation.

My issue is that "mental health" can be any issue that makes a person suicidal. You have a tough case to make that the collective government is responsible for fixing all those issues.

Instead, focusing on policies that harm (or do not help) those factors you rightly point out sure does seem like a better way of handling it. Otherwise what, you have to choose between housing costs and mental health (when clearly housing costs contributes to mental health??).

Choosing to focus on everything is choosing to focus on nothing.

Ancillary comment, From TFA:

    She said: “The evidence is clear that successive governments have failed to deliver on their promise to reduce suicide rates and make suicide prevention a priority. Today must act as a wake up call. The current approach is not working; we need meaningful action and proper funding if lives are to be saved.”

I understand the urge to do something on a large scale, but immediately pivoting to "the government" to solve personal mental health crises seems like a terrible idea.

It is fairly easy to go ask for a differently styled pelican, environment, background, etc. But that's not the point.

I was thinking just today about how, if I could hire my competition's lead designer/programmers, have them design a system for me cleanroom, that I'd effectively know everything there is to know about my competition's limitations. This is trivial to do with LLMs.

"Make me a thing that does X" is going to come out implemented basically the same way most of the time, as a kind of stack-overflow, well commented, internship project. (or, if undirected enough, as a collection of those based on a bog-standard architecture)

Same-y enough that you could train an adversary on it? Or hack it afterward? I am beginning to think so!

At the point that you run a successful (paid) youtube channel with sponsorship, is it recreational?

It's misleading. When it's right, it's right by coincidence, since the vast majority of algorithmic terms that seem like they should provide insight into the human mind, like "attention" could just as easily be called "finger" or "tape head".

And suddenly we're looking at people who read ticker tape or by following with their finger for insights into neuroscience?

Rule #1 of reading CS papers - don't trust the anthropomorphized terms. Rule #2 - definitely don't extrapolate those terms to the real mind.

Of course, attention is almost certainly a heuristic readers use, but to even bring it up as key to intelligence begs so many questions about LLMs and the mind...

OK. That's good to know. I'm not sure why sharing our professional experience with this type of code is controversial, but here we are.

I'm not defending rust here - a correctness bug invalidates the premise. But I am saying that I've never seen a custom drop in production.

For some reason this is controversial, but it remains true.

Not really. I've worked with rust for 4 years, systems side, and never had to write a custom drop implementation. That's the gotcha here, not the rigamarole with the static+Send option type (which _would_ be something you'd encounter in prod).

From my perspective you're entering the seasonal low. It's time for the old fiscal year turnover sadly. I've been laid off at precisely this time of year too, and years later I'm still working.

It takes me generally 3 months to find a good job, and while the industry is changing I suspect you'll be ok. Nobody I know is hunting for jobs.

Hang in there

On the contrary, the prestige will go up - filtering mechanisms, fatigue, and general banning of AI contributions will make it harder to get a PR accepted. If you make it past all that, you are a real one now.

Well, contributing to FOSS is a prestige accomplishment. It's viewed as a way in. It's now just a metric to game, and to automate that process is too easy.

Combine this with a growing sense of being "left behind" and a highly publicized collapse in software jobs coming any day now and any proof of contribution is highly prized. So why not?

Agree - the firmware is never really a selling point. But a shop built around routers could have a few angles:

1) Privacy/ ad-protection native (as you say)

2) High security and performance (fully tuned DD-WRT). Even solving bufferbloat for end users would be amazing - the world is barely catching up on that.

3) Good software add-ons that aren't proprietary junk. Asset management/upgrade, sign-in/landing pages, monitoring and trafffic shaping that holds up, etc.

Bunlding FOSS to solve these problems well, even on some else's hardware, does seem like work, and work can be charged for.

It was fun - until it was cringey, then it became fun again due to retro-fandom

Just calling the progression while we're on it.

IMHO: The rotating rhino, meteor impact, and animated flowchart is totally unnecessary cruft that makes it look almost satirical. If this keeps up, in time, this "AAA effects" stuff is going to look like the 90s "under construction" banner gifs.

Your boss forgot his other most important job: Being a communications router, not a communications bottleneck. There are many most important jobs for bosses.

Where's the HR-faced "Can everyone turn on their cameras please?" one with 200 people in the call.

I've said before, I'll say again: If you were to take all technologists and instill in them a values of simplicity and re-use and longevity, and an obsession with repurposing old tech, I think you’d find they roughly sort into two camps:

One would take on an intellectual journey of discovery, sailing off into the pacific to learn to live off grid, writing colorful narrative-driven articles, full of artwork, mission statements, and subtle suggestions of donations. They would emphasize the human nature of discovery and seek to engage you as they went. Come to the enlightened tech path! Some neat new technology would be built along the way around these principles, mostly for their own use.

https://100r.co/site/projects.html

But the other group would write a self-hosting Forth targeting the top 10 most used CPU platforms of all time and post a text-only HTML site with download links. It would work everywhere. You’d know nothing about them and they’d probably run linux on a 17 year old mac.

https://duskos.org/

Personally, I’m squarely in the second camp, but I wish dearly I had the creativity and spine to be in the first camp.

Right! it seems obvious why: Both these companies want to dogfood their own coding models and stop paying competition.

You can also read this as diminishing returns / AI isn't good enough, etc, but the simplest explanation is that they don't want to send money to Anthropic.

It comes out of the ground. UK has some of that I think.

Tunnels and basements just trap it so it can't blow away.

would love to see the radon measurements in those tunnels/basement.

In the martial artists' minds, of course.

Guns vs Martial Arts. There are absolutely benefits to doing martial arts that are not self defense, but when you have to defend yourself in a pinch a gun is pretty helpful.

Copy. The mechanism is probably similar once we abandon the "LLM that is asked to generate json" path. But the grammar alone doesn't ensure the output is a probability vector does it?

Both use the softmax operation, but "probability of emitting a given label" is not the same as "probability the decided action is correct". OTOH, taking an action to sample something to come up with "probability of emitting a label" is in fact the whole shebang if the label is the next action. If it works it works.

Who cares, this distinction is small.

For me, a planning guy, either your suggestion or Jev-like would work just fine, and you better believe I'll be testing both as I'm quite excited about being able to use less-structured, untrained systems in decision pipelines. Thanks for the heads up.

> And how is the confidence score any different than the softmaxed token probabilities you get out of running every llm?

You're asking me how a structured output which assigns probabilities to classifications of the input text is different than the next-token weights which are calculated while generating that output?

One is internal (token prob), one is output generated by that internal mechanism e.g.,

    "{ next_state: { "evade": 0.84, "land": 0.10, "search": 0.01"...}}"

Laya, a local Jev alternative, is a 400Million parameter model. It can play doom.

Find me a another class of 0.4 B model that can handle structured output decision problems with the same latency and accuracy as Qwen 3.6.

Partially so. But the fact that we're comparing Jev to SOA LLM and pre-trained classification means it's essentially satisfied its reason for existence.

The benefits of Jev, and the reason I'm excited about them:

* Pre-trained/tuned to provide only structured output

* Small, fast, essentially trivial to locally run - this alone makes them an actual candidate for real-world planning systems. Laya is a local-first 400 million parameter model, for example.

* Their confidence scores are useful, even if those scores are not correct/true probabilities.

For me and my work, they look like nearly turn-key, tiny, fast, locally-hostable models that can manage state transitions in a deeply autonomous system.

LLM-as judge requires, well, what we know as a full LLM. A fully trained classifier model requires fully training - something you cannot / won't do for a embedded autonomous planning system (by contradiction - if this worked, we'd have used it everywhere already!). Totally unfair comparison in my mind.

The marginal / ancillary discussion of the usefulness of people in a post GAI world was really interesting. The best part of this discussion was that the protagonists entire job is to translate ideas that are too hard to understand into a tangible/actionalbe "surface area". Essentially - a policy-maker interface that treats superintelligence as a "black box".

I have an excerpt of the book explaining this here: https://jodavaho.io/posts/books-blindsight.html

The discussion about non-conscious intelligence was less so (I see ants every day!). But this is a new idea dense book to say the least - if you love worldbuilding, please pick it up.

And, carry that forward, what do you expect to happen to people who do not act according to your definition of nice? You either accept they are different, or try to prohibit the activity.

Jumping from "money making via tax breaks" to "murder" is quite a jump, also. These are along a spectrum of choices.