horsawlarway

They're sticky because they stay focused on their specific domain, and they work well.

I also really don't understand the value pitch on "6 apps no one can use" in the post.

I publicly host Seerr and Jellyfin for my family, and they use that combo just fine. They explicitly don't get access to see any of the sausage making under the hood.

Given this doesn't support music yet (according to op) and you still need to host plex or jellyfin... what is the value statement here over seerr?

I use a 200w vivosun pointed at the wall during the short daylight months.

Costs a fraction of what you mentioned at roughly $150. Or $300 if you really want to go as high as 400w.

Makes the whole room feel like you've got bright sunlight pouring in.

My wife made fun of me at first, but now also has one for her office.

Or that infants with a rooting response survive much more often.

Basically, weights encoded biologically from hereditary behavior.

Instincts.

Now, the real conundrum is that if our DNA can really only encode ~12gb then how are we passing this information down?

So they give you the name of the exact turbine, but like... how is that helpful. I don't know random german nuclear turbines by name, and looking it up just... tells me the size and makes the game pointless.

They want ~9k for a freaking 13.5kwh pack.

I can buy a UL certified, 16kwh, outdoor rated pack for 3.5k (ex - EG4 allweather)

I can buy just the LFP batteries themselves for the new powerwall 3 LFP version (no BMS, no shell, no certification) for $1116... https://18650battery.com/products/byd-302ah-3-2v-lifepo4-pri...

Powerwall is a scammy product. WAY too expensive for the storage it provides.

---

So the summary is: Cybertruck is using a different battery chemistry, Powerwall is a scam.

(I mean - to be clear, the Cybertruck is also a scam, but it's a different kind of scam not directly related to the battery capacity, since it's hauling around a 123kWh NCMA battery pack)

But that has jack all to do with the LLM capabilities that are sold.

And we have an absurd amount of historical case law that deals precisely with "conflict of interest" among parties. I don't see why we'd allow this to be whitewashed away?

"No your honor - I didn't break the Sherman antitrust act. I just fine-tuned a model so that operating it functionally broke the Sherman antitrust act!"

"No your honor - I didn't violate my fiduciary duty. I just used a model that made self-interested decisions on my behalf, at the expense of my client!"

---

My take is this is going to be a relatively serious point of contention, because it won't just be consumers getting screwed (which we... used to care about... but in modern times apparently don't give too many shits about anymore)

Ex - the app store thing blew up into the whole apple/epic court case. And then a ton of EU legislation.

This is monied parties with different interests, who will use courts here.

Your whole stance seems to be "it's natural why try?"

And my obvious rebuttal is "society is a construct - we can adjust it"

We've been doing it for thousands of years now (up to hundreds of thousands).

Because the people building them aren't the people operating them, and there's a real tension in the objectives and goals between those parties.

Further - how do we balance out that the people building them have given themselves extra power (no restrictions, private models, special access to prompts and system prompts, unencrypted reasoning traces, etc) when operating the LLM themselves.

---

Even if the tool is good, it's a bad bet to build a thing on top of tooling with these limitations.

This is the same problem we have with mobile software stores and app distribution. Theoretically the store serves a purpose and can provide value, but practically the main goal seems to always devolve into enriching the owner at the expense of everyone else.

Even if we accept that LLMs should be limited (and I'm not actually sure I do) - the legal standards around what/how/when those limitations are used need to be VERY stringent.

Technology conglomerates are decidedly not "white knights" with a motto of "do no evil". They are self-serving rat-bastards. That should be the lens any power they wield is viewed.

The framing is absolutely not "anthropic is being careful with humanity". It's "assume the devil has control of those companies - how bad is the result of them having this power?"

Active parameter count really drives inference speed not ram requirements.

For this model, assuming you want the full 1mm context length at 4-bit quants, you need something like 175 GB total VRAM.

~150GB of that will be weights (varies a bit by quant format) then ~3-5GB for kv cache (might be lower here, I think they're using mostly HCA compressed cache) and then I'll allocate a couple gb for overhead/concurrency (say another ~10GB as a safety buffer).

---

So the good news is that it comfortably fits into 2x128 at 4-bit, and we can realistically stretch to a 6-bit quant and make it just fit (~244GB needed for the full 1mm context).

mainly because the thing about open weights is folks will just immediately strip it back out.

But I can guarantee you that openai/anthropic/other will absolutely jump at the chance to monetize users by feeding them paid recommendations.

Because we both know that gas generators also require a production step... arguably one that's quite a bit more complex than a battery, at ~$1,000,000/MW of installation costs. And that installation of that complex is going to be very carbon intensive (even the concrete foundation alone...).

And then on top the plant actively burns hydrocarbons for the entirety of its existence.

So fill me in here - what is the exact argument you are trying to make with this?

So sure - it wasn't designed with agents directly in mind, but it's supported workload identity federation for a while (and is used in this manner for lots of big players - github, GCP, AWS, Azure, etc...)

This just seems to extend that concept slightly to allowing the workload identity to be generated on the fly based on user/agent combinations, with a HITL addition to approve the identity being generated and grant it a subset of permissions.

---

What you're saying here is different than what you implied in the first comment.

There is a real and meaningful difference between

1. A user can blame a rogue agent for actions they [the user] performed

2. An agent itself went rogue

So sure - #2 is still a risk, and one they'd have to mitigate with the security controls they talked about (minimal access, temporary grants, HITL, etc). Frankly - #2 is always a risk. That's what "agents" are (both the bot and the fleshy kind). It's also why we have logging and visibility tools that should be recording what was done, by whom, and for what reason, along with robust backups. You can't really close this hole, you accept it, monitor it, and hold the agent to account (and again - I'm with you on discussing culpability for bots and what that means in the legal sense. Ex - do we charge the person who introduced the bot, the company running it, the party that injected bad context, etc...?)

But #1 is not a thing a person can do - they can't take malicious actions and then blame a rogue agent.

They're mapping the identity back to "agent" instead of "person". With the optional ability for the agent to present an "agent:person" combo in cases where it was triggered by a person.

Some quotes that imply but don't actually technically clarify what I mean:

> we now had a way to treat each cloud agent session as its own identity

> Once the human approves, access is scoped to the cloud agent session ID and the specific database that was requested.

> Using OIDC, cloud agents can request and be approved for short-lived access as they would with databases above. Requests are always logged in Formal and attributed to both the cloud agent and the engineer that launched it.

It sounds like they're not just attesting that the agent is the person, they're always scoping to a new combined identity.

So there's no confusion over whether it was a person, an agent, or an agent started by a person. (however where you go from there for culpability is a different question...)

But it doesn't seem like there's a case where a person did a thing (with their own credentials) and can then point a finger at an agent as say "oh it went rogue".

Pretty much. I vaguely remember a couple of "off-grid" devices that try to automate this for you (ex - you can set a priority on each breaker, and as supply dips, they get dropped in order of priority).

But functionally speaking, you just need to install a battery. Frankly - Batteries aren't discussed as much as solar, but the big breaking point we're hitting right now isn't just that solar has gotten incredibly, unimaginably cheap.

It's that BOTH batteries AND solar are getting very, very cheap. And the new generation of batteries are also considerably safer than the original lithium batteries.

Lithium Iron Phosphate batteries (LFP) are ~$60/kWh right now. I can buy them commercially in bulk in the US around ~$80 (pretty much all china production - full stop. Historically, I've bought EVE cells, but I'm now seeing direct BYD branded cells as well).

If sodium-ion can hit the same price points (and odds are looking better and better they'll both hit and then plow past as they bulk up production) then my read is that modern power plants are going to be very, very different.

They'll look a lot more like the connection structures we see for trees - a bunch of connected independent installs that can optionally cooperate as needed. Ex https://www.bbc.com/future/article/20260928-a-virtual-power-...

Most solar installs work by tying into the existing main breaker of the home.

The cheapest way to do this is to leave the line feeding from the power meter to the main panel untouched. Instead you do anti-islanding detection at the inverters, and when they detect that mains is "down" they automatically cut off to prevent trying to backfeed power to your whole neighborhood during an outage (dangerous, not a good idea).

---

Second issue - there are many appliances that are NOT happy about being turned on and off at random times, and it can significantly impact their lifespan (ex - refrigerators, acs, computers, washers & dryers, etc).

So given the variability of direct solar, you don't want to accidentally destroy your ac unit by having it turn on and off 100 times as random clouds float by.

---

But! If you're willing to install even moderate battery capacity (and LFP is getting quite cheap as well). Then you can absolutely do this. It's just more complicated and more expensive.

You need to wire in a mains relay (~$1,000 depending on the unit) and install some battery capacity to prevent very frequent startup/shutdown, then go through some painful additional signage and labeling requirements (at least in most of the US) and also install some additional shut-down switches on the exterior of the home (ex - to automatically turn the inverters off, and to shutdown the battery system).

Then, yes, you can absolutely stay on solar power while the grid is down. In the same way that people are allowed to run whole-home generators to stay online when the grid is down.

This is what comes to mind reading this thread.

Go google "natural monopoly" please. No one is arguing that work is not happening to provide a service.

We're arguing that capitalism clearly fails to achieve good outcomes for optimizing the quality and cost of that service in the case of natural monopolies.

No one is assuming it's magic free fairy dust.

We're making the argument that perhaps we should subsidize public transport in the same manner we do roads - aka, fucking tax people who have the money to cover the cost.

If we assume that movement is generally positive enough that we don't charge directly for access to roads, why do we assume the alternate for public transit?

Everyone fucking understands how roads get paid for - we spend an inordinate amount of taxpayer dollars on the infrastructure at all levels of government... city/county/state/federal.

And you might then go say something silly like "but muh gas tas!"

And you're right - we do generally tax gas sales to help cover roads, and this is directly a form of consumption tax on roads. Except... we tax gas at $0.33/gallon in my state. A gallon gets you ~20 miles. We charge $2.50 for a metro ticket with an average trip length of 5.7 miles (this includes both rail and bus).

If we taxed the metro trips at the same rate we taxed gas... each rider should be paying ~9 PENNIES. We can round up to a dime. Instead they're paying 25 times that amount.

---

So ultimately - I agree 100 percent. We should think more about how the public should indirectly pay for this. Instead of doing the absolutely batshit insane moral proselytizing that ends up with us ALSO indirectly paying for an entire damn legal process to prosecute broke people for skipping a $2.50 fare on the metro.

But damn if America isn't truly focused on "fuck those poor folks" in a lot of really odd (and usually self-harming) ways.

Rent-seeking behavior is too good to give up, so companies will move all content to streaming only services (games, movies, tv, music, books, etc...) in the majority of cases.

Physical media will only be available at very high prices, for a very limited selection of content.

We will go from "doctrine of first sale" around content licensing to the utterly crippled "you own a temporary, non-transferrable, and ephemeral license to stream this content from us on exactly this service using only approved devices until we decide otherwise".

The moral action in this case is to simply stop paying. Period.

Long term - we should be fighting to enshrine a new flavor of "doctrine of first sale" for the digital age.

Ex - maybe purchasing any content license sold with the word "buy" should enshrine your right to use, transfer, sell, or otherwise consume in any reasonable form that content for the duration of the copyright + 7 years. And we will not enforce copyright for items that can't be bought (only rented).

At which point no one would need to pay a fare, and we'd all still save money trying to prosecute broke teenagers trying to move around a city?

Further, even if this is your take (and I won't touch the geopolitics here, maybe you're right, maybe you're not regarding ulterior motives) - go use non chinese open models that are also good.

Laguna, Nemotron, Mistral, Arcee, Llama, etc...

But the under-reported corellary of this is that a TON of industries essentially operate on hardware right out of that time period (or slightly after, circa 2000s)

banking

Insurance

Healthcare

Gov

retail

Transportation

etc...

Hell, I was writing code to run on IBMs Z/arch as recently as 2021 for banks with mainframes deployed around 2004. They were just then finally considering an upgrade to the later models.

Because again - I think lots of businesses are already taking a good look at models well off the "cutting edge" because they're good enough for what that business is trying to do. They're already there.

If you're doing categorization, tagging, form-filling, customer support routing, etc... you don't need anywhere close to cutting edge expensive models.

Hell - Jev being so popular is basically exactly that group going "what's the cheapest way we can make this decision".

Meanwhile, there are plenty of spaces where you still want real model improvement. But a lot of traditional business was basically process and paper pushing. You don't need a Phd level model for that, you need an intern level model.

---

But that's the trap for the large providers - intern level models are cheap and easily available with current open weights. It's going to be a very tight market, and margins will go down to essentially the floor (especially once we exit the current constrained-compute market).

Somewhat ironically - the technology itself makes switching costs basically zero. It's not just that there's no big lock in, AI is like an anti-moat. It actively fills in and paves over whatever moat you might have for price conscious customers.

And that's the rub - it doesn't matter if inference is profitable, the US labs have promised the whole kitchen sink, and taken serious investment on that pitch. They won't make it back in a cut-throat market for inference. They have to continue pitching the idea that only their models are worth using, for quality/safety/performance/etc reasons.

Which means continuing to develop and release new models. At huge expense.

Side note - while deepseek seems to be doing fine, I'll note that the emphasis on cost effectiveness is very much not a thing for the US labs right now... and it was for non-us labs. I'd be quite interested in knowing if anthropic is pulling in the same returns, or what the margin on their subscription plans actually is.

---

I'm already seeing this directly at work - lots of devs still happily use claude for development, but the products we're making aren't. They'll use kimi, glm, gpt-oss, etc... basically - if cloudflare is hosting it, we're exploring it (mainly for contractual/security reasons): https://developers.cloudflare.com/workers-ai/models/.

Because we are price sensitive, and we already have to do all the work of creating very robust evaluation frameworks for our products, so we might as well choose the cheapest model that meets our standards.

whiiiiich is why we're seeing things like Haiku and gpt-luna priced down near the open models at ~.10/mm in, ~.50/mm out. They've been forced to cut prices to keep commercial customers on their cheaper models instead of moving away.

This will continue. If inference is profitable and easy, prices will continue to erode. Go browse openrouter, plenty of providers happy to undercut the competition. Bigger shops with auditing requirements will avoid them, but smaller ones won't.

Humans are autonomous, distributed, and adaptable. In a package that weighs ~150 lbs, with very low energy requirements.

You literally cannot wipe them out without a sustained and incredibly expensive ground campaign, and even then it's not likely.

AI in a data center on the other hand... So easily disabled.

All you need to do is see how frequently he posts on Twitter to know he's broken and trying to fill the hole with money and attention.

Because tokens have a cost. CI jobs have a cost. Organization objectives and goals have a cost. Features have a cost. Code has a cost.

---

So in a world where agents do all the work... why are they not able to control operating costs AT ALL?

Why is the right operating method the one that happens to blow an absolute shit load of cash on - wait for it - internal tooling that the customers don't give a fuck about?

Customers do not care how the internal business operates. They care about value delivered. I'm absolutely not convinced that this "churn churn churn" method of operating is actually delivering enough extra value for the cost it incurs.

---

Now - that said, I think there are some fair points here that are worth considering, mainly around whether a distributed version control system (git) is really the right call for agents that need to coordinate much more closely than most human developers did.

I think that's a directionally appropriate insight, and worth considering.

What I don't find compelling at all is the idea that many (any, really) companies should expect to be running millions of ci jobs a day, and mergings hundreds of thousands of prs/commits. That reeks of "I have cash to blow and don't care".

This amount of churn feels like a thing only possible with no real profit motives and a lot of money to waste.

Lots of activity is not necessarily synonymous with productivity, and it carries real cost.

So I think the design here is interesting, and the concepts seem applicable to where some agentic development may end up, but I question the underlying principles that drive this being the right solution space at all. At least if the goal is a profitable company.

You usually use entirely different techniques for hand vs power tool.

The way you'd do the work, the methods of joinery you'd choose, the style and choice of cuts, even the preferred size of boards - all completely different.

Learning hand tools is an entirely separate skill set. Down to the planning stage.

It's not merely doing power tools more slowly. And trying to use power tools like they're merely "faster" hand tools also doesn't work. You waste a ton of time because it's just a different operation.