Ditto. My personal equipment includes a home server (128GB DDR3 ECC) and a tablet with a keyboard. It's honestly astonishing what you can do without a full-fledged laptop, if you're willing to go through some gymnastics to get there. And it travels light compared to a laptop! (The tablet, that is. Not the headless box. :-))

In a similar vein: seek efficiency.

I.e., /if/ I am going to consume LLM tokens, I figure that a local LLM with 10s of billions of parameters running on commodity hardware at home will still consume far more energy per token than that of a frontier model running on commercial hardware which is very strongly incentivized to be as efficient as possible. Do the math; it isn't even close. (Maybe it'd be closer in your local winter, where your compute heat could offset your heating requirements. But that gets harder to quantify.)

Maybe it's different if you have insane and modern local hardware, but at least in my situation that is not the case.

Or songbirds.

I ran into a guy at a hardware store who ran just such a power supply attached our city's water (or was it natural gas?) infrastructure. I was incredulous, but the idea that it helped prevent corrosion did make sense.

(edit: I see you answered a sibling comment with the same question. TL;DR: Potential output is the output pretending that curtailment did not apply. Thanks!)

A UI or terminology question: when 'Potential output' says it is 'Including curtailment', does this mean that it pretends that curtailment doesn't apply, or that it subtracts the curtailed power from the total available so that the total power shown is only the power actually transmitted (exported) to the grid? It's very likely that I'm just not familiar enough with the terms, but this wasn't immediately clear. My guess is the former meaning, although I can imagine it meaning either.

Regardless, this is incredibly neat, and I'd love to see this kind of data for the grid that serves me (Eastern Interconnect in the US) -- are you aware of any sites similar?

> Regardless, this is incredibly neat, and I'd love to see this kind of data for the grid that serves me (Eastern Interconnect in the US) -- are you aware of any sites similar?

https://app.electricitymaps.com/

(for most US grids, ElectricityMaps consumes somewhat delayed EIA Balancing Agency generation mix data from https://www.eia.gov/electricity/gridmonitor/dashboard/electr... ; their data is mostly live for system operators that provide live data on their own website, CAISO in California and ERCOT in Texas, for example)

Sounds like a neat space to be in. Wishing you calm skies next month. I guess you'll have all the goodies (RFI detectors, thermal imaging sensors, etc.) to collect data? I wonder to what extent detection from a distance (discovery?) and investigation of faults can be automated. Hopefully using drones for good will be sufficient for a viable business model. We've enough surveillance as it is.

We're collecting 61MP RGB, LiDAR (DSM DTM) and radiometric thermal for sensors

We're thinking of flying at ~150' AGL or ~100' ATO at over 20 mph to collect data when automated. There's trade offs between effective speed, localized navigation, and mission planning. It's just challenging to build fully automated systems, but generally speaking flying higher and faster is more efficient and safer (also helps with command-and-control links)

> "The draft from the GSA also mandates that contractors "must not intentionally encode partisan or ideological judgments into the AI systems data outputs," the FT reported."

Foreshadowing?

I think if they're plainly saying what they're doing, that's not foreshadowing, that's just announcing.

And you know they don't mean Grok (the most ideological AI in existence).

EDIT: probably not relevant, after re-re-reading the comment in question.

Presumably littlestymaar is talking about all the LLM-generated output that's publicly available on the Internet (in various qualities but significant quantity) and there for the scraping.

No, I'm talking about generating data for the purpose of training. The latest HF paper on the subject offers a great intro to the technique: https://huggingface.co/spaces/HuggingFaceFW/finephrase

The site arena.ai does exactly this already, as far as I can tell. (In addition to the whole ranking thing.)

I'm genuinely ignorant of how those red teaming attempts are incorporated into training, but I'd guess that this kind of dialogue is fed in something like normal training data? Which is interesting to think about: they might not even be red-team dialogue from the model under training, but still useful as an example or counter-example of what abusive attempts look like and how to handle them.