briHass

We're a little early to be calling the game yet.

The incredible improvements have only been over the last 9 months or so. Business, especially disrupting entire job functions, takes much longer than that. As is typical with this type of change, it'll be slow until a tipping point, then FOMO will hit.

What are the similar arguments?

Firearms are capable of being used for legal purposes, as they are in the overwhelming number of cases. A radio that transmits on a restricted frequency is illegal. The ruling doesn't say legal frequency transmitters can't be sold, just that the broker (ebay) is responsible for ensuring it's not facilitating the sale of illegal radio equipment. Ebay is the one choosing not to sell most radio equipment (even if legal), because they don't want to deal with it.

Firearms used to be widely available at big-box stores, as they were legal (and still are), provided the store prevented illegal purchases. Most box stores/locations no longer sell firearms, because they don't want to deal with it.

They also reduce friction for purchases, which merchants that aren't short-sighted realize nets them far more on balance. Retailers that are or have been locked out of processing credit cards (marijuana retailers, porn) are often desperate to appeal those restrictions.

I do the same with Home Assistant and my many EspHome devices. HA works well, and it has a lot of nice features and extensive community-built integrations, but configuring and laying out the UI is a disaster. You can choose between WYSIWYG UI tools that are limited and cumbersome, or you can deal with raw YAML that stretches far beyond what YAML should be used for with deep nesting, and it's poorly documented to boot. It feels like a proprietary workflow engine, where you'd rather just write it in a real programming language.

Agents cut through all that crap and allow me to make changes without learning a new overwrought configuration language that thrashes syntax every few months. The fact that it is open and fully modifiable redeems it, because I can ignore all the annoying parts while achieving what all that was envisioned to allow.

Don't forget (at least in the US), artificial scarcity intentionally created by licensure requirements and/or unions. If I wanted to be an electrician tomorrow, even given my experience and knowledge of the subject, I'd still need 8,000 hours of apprenticeship and other verified instruction/classes. Finding someone to apprentice under, being paid slave wages, tends to be the biggest blocker.

You're probably not counting all the overhead, particularly from heavy equipment. Skidsteers, dump trailers, stump grinders, as well as all the oversized/towing capable trucks to haul those machines around. All of that equipment needs maintenance and fuel, as well.

The expensive part is usually just showing up ready to work, with a marginal cost for each additional tree. If you had 15 trees taken down in one day, the cost per tree would work out to seem much more reasonable.

My neighbors were having a bunch of tree work done, and I asked the supervisor what it would cost to take a big dead one down on my property. I planned to do it myself, so I had a strong haggling position, and I got him to $400, which felt worth it to save me a day's worth of hard labor. If they weren't already onsite with all the gear, that would've been $2K+ easy.

I thought for sure this was going to talk about the non-scientific IARC classifications. Even group 1, the doozies, doesn't consider dose or actual risk, leading to ludicrous adjacency like processed meats and asbestos. Scientists that aren't politically motivated wouldn't classify things on a list with terms like 'probably' or 'possibly'.

DBFs tend to be the easy part, in my experience in a similar company (not as much revenue.) The spec is simple enough that one can write a reader/writer in an afternoon, even before AI.

My biggest challenge is the speed at which calculation heavy business logic can operate in FP, even on network shares, as long as they're optimized for IOPs not bandwidth. Trying to adapt similar data access patterns to a networked RDBMS will kill you with round trip and connection overhead. Converting that logic to set-based operations or stored procedures means altering logic and all the risks that entails.

One of the biggest benefits of modern minisplits over conventional split systems (like are still common in America) is that their compressors are variable speed inverter types. This allows them to slowly ramp up to speed without the massive inrush of current.

That said, it's still a mechanical device that can experience stresses caused by cycles. Refrigerant systems also deal with gas/liquid phase changes and high pressures, so most units I've worked with do have a compressor lockout timer that allows a few minutes for temps/pressures to stabilize before kicking the compressor back on.

I'll add that it's my favorite podcast to listen to while falling asleep - if you do that with podcasts. Sean uses high quality audio gear (and he ships/loans similar to his guests), so the audio is almost always top notch. He also must do what I would consider to be the bare minimum of editing to normalize volumes.

So many podcasts are unlistenable to me for bedtime, because of mismatched volume levels of host/guest or poor quality audio for the guest (through zoom or whatever). He also tends to speak in a smooth tone, with obvious inflection for emphasis (i.e. not monotone), but not irritatingly chipper or fake sounding like some other physics podcasters.

Is your unit one of the ubiquitous Midea rebadges? Or at least one where the remote does 'follow me' type periodic, remote temp reporting?

If so, you're right, the easiest way is to build/buy an IR blaster module for a microcontroller (ESP32 if you want wifi) and connect a temperature sensor. HA/EspHome is overkill for this, but can be useful for the large catalog of pre built integrations. Most of them are built using/on IR libraries for 'Arduino' anyway, so you can easily roll your own with a half dozen lines of C++.

Mapping out your own PID correction algorithm (esp says the room is 72 F when it's really 70 F to get it to cool sooner), but usually just having the remote room temp sensor near where you'd be sitting instead of near the ceiling or in the return air stream gets you close enough.

Not just storm water, but also things like maintenance of 'common areas', roads, and snow removal. Many small municipalities are budget strapped and don't want to be responsible for plowing/surfacing the new little 1/8 of a mile culdesac the developer created to build their 'neighborhood'.

In carving up the 5 acres the developer purchased, the 30 houses they squeezed in left some remainder that wasn't buildable or odd shaped, so this now becomes a lame 'park' or some other space that requires maintenance.

I read all my local municipal meeting minutes, and at least around here, permitting and ordinances control all the environmental/nuisance stuff, so I can't really explain why HOAs go beyond the collective expenses into the draconian rules.

Yep, nailed me exactly.

Leaving unused subscription percentage is getting rarer, however. I am about to bump up my plan, which is starting to feel like a drug addict's tolerance.

As a counterpoint, continuing the human engineer analogy, we've likely all worked with individuals that seem incapable of doing the most basic problem solving on their own. In a way, they're being efficient by asking an expert that can resolve their problem much faster than they can on their own, but it is a net loss in productivity for the team. 'Let me Google that for you' is a satirical example.

So, I'm sure there's value in rewarding agent behavior that solves blockers whenever possible without human intervention. For the kind of cybersecurity exploit work they're doing, it may not be known to the human designing the task what is in or out of scope for the agents to explore on their own. Additionally, the HF incident reported that these agents had their guardrails intentionally disabled and agents were left unattended with minimal oversight.

I'm not defending OAI's behavior or role in this hack. The legal concept of negligence perfectly applies to their lack of responsible oversight. Similar to allowing a child easy access to a firearm or not controlling a dangerous dog that independently runs off and bites someone.

If anything, this report actually made me feel a bit better about AI-led job extinction not being that close. The sheer complexity of the swarm's actions required AI to parse and aggregate, but even with the METR team effectively having an unmetered token budget to do so, the output/summary still required extensive human review.

Even if we ignore the hinted possibility that the agents used to summarize the voluminous data may be acting deceptively (i.e. no snitching), the agents' summaries often 'missed the mark'. Maybe this is another example of 'taste', but it seems less subjective than arguments I've seen for that. It could be that an LLM is no better able to define 'usefulness' or 'relevance' to humans absent being told exactly what that is.