It takes an adjustment or familiarity.

...or not, and this is why HomeBrew exists and I need to learn it or ansible/etc.

A vendor conference I was at a few weeks ago focused heavily on this, for most of their Agentic workflow items. How can you show the AIs work, prove what it did was within guidelines, then audit the process and result.

Like, if your system has an AI backed federated search for documents and you ask it a question about those documents, you need an audit trail of the ask, what documents were referenced, and what was returned to the user.

Then if wrong information was supplied that can be pinned down and explained in case of lawsuit or other need.

100000 LOC per month /25 days per month /8 hours per day / 60 minutes per hour

That seems...problematic for anyone doing code reviews.

Still imperialistic and self serving in many ways, but it worked.

On the other hand, I've recently talked with a Polish to US immigrant who was moving back to Poland this summer as jobs and more had improved. They were competitive (in his mind) with the lack of opportunity and anti immigrant thinking across the US today.

At least, that is what I get from the MOE style. Small and fast experts with a router LLM on top to best use them, then the harness to keep it all together.

LLM worker > Harness, Agents w/ skills > Human oversight/input

This is similar in structure to many teams I work with, something like:

Dev/SE/etc > PO > Manager/Director

Or whatever your current org structure relates. The LLM worker and Harness/Agents compact down to one human layer.

Now with MOE LLMs, the LLM layer is breaking into like:

LLM worker > LLM router > Harness, Agents w/ skills > Human oversight/input

Does that mean the Human element can be condensed to a single Manager with the right skills? A Director above them? Is the VP above them directing the agents?

Is this another variation of Conway's law, where orgs design systems that copy their own communication structure? Seems like that is how my Manager/Director approach it. Then again, they are making slide shows and obviously AI assisted reports, not something that needs to be stable and responsive for the entire product lifetime.

But to your point, the manager sees it as a structure to manage to increase productivity. The craftsman sees it as a tool to further their craft. Each is driven by a different methodology and use case. Can that mesh unless specifically directed to throw AI at no joy work?

As soon as we hand over searching out information to social media algorithms and LLM tools, we abandon our ability to see reality outside our direct vision.

Grok's ownership has already demonstrated capacity to influence major world elections and other events. You cannot trust it with this sort of information gathering and reporting.

It may be analogous to vaccine adoption, with requirements in place for all (with some exceptions).

Who sets and applies the exceptions? A state or federal agency that runs afoul of free speech issues? An AMA type public board for social media use, quickly captured by the industry they serve to regulate? Parents themselves via opt out paperwork or ignoring the regulations?

I want to see more options debated or proposed for this sort of management, starting with right to privacy and your data, harsh punishment for promoting misinformation, and disclosure of algorithms/etc on what your feed includes.

Make all fines a flat % of revenue or an otherwise real amount to companies like Meta, not just a cost of business. Maybe pay users when their data is used/sold/etc, or otherwise increase the cost of what are basically information dragnets for advertising and manipulation.

The tech is there but not applied equally. Regulations could limit every car to 70mph, or the (US) state to state max highway speeds. eBikes are sort of already doing this with categories based on max speed, in some states.

Then like speed limits and (emissions) regulations, people would find ways to bypass them. Rolling coal is one example. Electric 'bikes' that are basically mopeds or mini motorcycles don't require registration or licensing in many places, or at least don't enforce it.

How much freedom is the general populace willing to stomach? These age verification laws apply to everyone but tend to only impact nonvoters (age 17 and under). People are generally apathetic (especially US voters) and may comply with the easy option before fighting and preventing any "foot in the door" for this sort of policy.

I can't use Instagram due to the ratio of ads and sponsored posts yet apparently millions are fine with it. If they can send their ID and likeness to Meta once and continue to scroll, how much will they care?

Adding layers for accountability is a good idea. It needs to start with social media itself, including preventing misinformation and disclosing algorithm/behavior nudges designed to suck people in.

I'm here for it. Local models can do a lot of what I need at almost no cost, plus the fun of making them work better or building a new system to handle that aspect of my home lab. A Strix Halo system may not be amazingly fast but at 128gb of RAM it can keep up with most open models worth exploring.

Based on June 1 Copilot Pro plan premium token burn and cost, unless you REALLY know how to use cloud AI efficiently and are tooled up to do so a local LLM on hardware you may already own is very appetizing.

I converted a lot of work today to a 6.5gb local LLM on a 12gb GPU and no, it's not as good. But it is 'free' or at least feels that way, especially when I need to redo something and my copilot premium request % doesn't change.

- v4.5: 1x cost, 100% quality, 100% speed but maybe sometimes 80% speed because of load - v4.6: 3x cost, 105% quality, 80% speed most of the time depends - v4.7: 9x cost, 115% quality, 90% speed most of the time

Then people will either stick with v4.5 for everything it can do and, if knowledgeable, use v4.7+ for critical or specific tasks.

But if we add the option of:

LocalLLM: one time hardware + electricity cost, good enough quality for 90% of work, good enough speed for 90% of work, no vendor lock in/sudden cost spikes...

Then there is an edge to running it yourself unless you can burn investor cash to get to the next level.

I think the recent headlines on org token spend plus my own experience just today (June 1) with the new Copilot Pro limits is going to push those with the compute to run locally.

As of about 1pm today I did something to hit 47% of my entire June premium requests (copilot Pro, not converted).

As of 2pm I'm using Gemma 4 E4B on a 12gb GPU (with large context window) off my desktop to power VS Code with Copilot on my laptop. I'm going to build an AMD Strix Halo system next week when parts arrive so I can queue up a few models in parallel or work with something I need that much RAM for.

I'm not lifting the earth with my LLM setup. Gemma 4 E4B is solid for accelerating my current projects. and it's costing me pennies more per hour vs blowing half my Copilot Pro plan in a distracted morning.

I'm at a vendor conference this weekend that is showing off their Agent/Agentic workflows. Nobody can tell me how they balance the cost long term. Hopefully whoever the vendor is paying for their cloud LLM token usage doesn't spike cost in a year (or the vendor themselves) after companies convert and are trapped VMware style with these agent processes. You can bring your own (cloud) model subscription. I need to find out if we can point it back to our own local LLM endpoint and try local models for the same processes. Even if it takes 5x longer, it could be cheaper and more secure.

the current dielmma for me is how do I install a model on a remote LM Studio device without bypassing Lm Studio to SSH or remote in?

> lms link [servername] get model ?

> lms get [servername] model ?

> lms get model --link [servername] ?

Maybe I need to read the docs again but I swear the only way is remote or go to that device and download via the GUI, ssh in and use the local cli.

Maybe can copy/paste from one device's downloads dir to the server? Maybe I need to try hosting models on my NAS and see if I can download from device 1 then run on device 2 without install/setup?