scottcha

Yeah, that was my first thought. I actually wrote a blog post a few weeks ago modeling the point at which agent recursion really gets out of control. https://www.neuralwatt.com/blog/agent-bedlam-a-future-of-end...

Shameless plug . . . I run a startup who is working to help this https://neuralwatt.com We are starting with an os level (as in no model changes/no developer changes required) component which uses RL to run AI with a ~25% energy efficiency improvement w/out sacrificing UX. Feel free to dm me if you are interested in chatting either about problems you face with energy and ai or if you'd like to learn more.

My Grandparents lived in a very small farming town (pop 500) and word would get around town when chicks had arrived and she would take us down there to see them.

Monte-carlo modeling has been a hot topic here the last few days. I've been using a monte-carlo model to estimate the factors that contribute to AI energy growth and got curious as to what happens if we get agent to agent calls growing. I know its not very common today to have one agent trigger another agent call but I'm certain its a scenario that will be more common in the future. So I decided to find at what threshold this pattern causes out of control growth.

Does anyone have any good examples today of agent call chaining?

Interested to hear others thoughts here on this. I also am hoping to make my model available in some manner in the future for others use so interested if there are other parameters and factors you want to see?

Pretty sure Seattle (maybe King County) doesn't allow billboards. You can really tell when you pass the banned area when driving south on I-5 getting close to Tacoma.

Also if interested the opening scenes of The Monkey Wrench Gang (by Edward Abbey) are about illegally cutting down billboards in Southwest Utah.

His style evokes a bit of Hunter S Thompson for me. I appreciate that it’s a bit different than your standard blog style.

WSJ reports today that the gao Itself reported 140 billion in improper payments. https://www.wsj.com/politics/policy/doge-musk-government-was...

This is based on their statistics so I imagine the next step is to find the actual waste and fraud and stop it or get the money back.

I’d like to see what cpu governor is running on those systems before assuming a power cap is in place. Lots of defaults installs of Linux ship with the power save governor running which is going to limit your max frequencies and through that the max power you can hit.

3.5 Sonnet is definitely my goto for straightforward tasks in github copilot. It seems much more effective due to its lack of verbosity and focus on completing the task rather than explaining it. Really helps in the new agent mode too.

Occasionally I switch out to one of the other models, usually GPT 4o, when I can't define the task as well and need to see additional analysis or get ideas.

Just an anecdote but my family member does university research and gets grants from NASA (also is ex NASA and left NASA due to too many rest & vest folks). In the last round of reviews there were 20+ NASA employees sent locally to review for a couple 1 hour meetings. Initial take from the outside is it looked like a seriously bloated process.

I have a project which uses weather info to predict avalanche risk. Reading the articles its hard for me to understand whether this would apply or not but my feeling is it might (If I ever need to run this in the EU I would talk to a lawyer). https://openavalancheproject.org

Two reasons that I understand 1. not all these AIs are LLMs and many have much lower latency SLAs than chat and 2. These are just one part of a service architecture and when you have multiple latencies across the stack they tend to have multiplicative effects.

There are a large number of reasons the AI datacenters are geographically distributed--just to list a few off the top of my head which come up as top drivers: latency, data sovereignty, resilience, grid capacity, renewable energy availability.

Yet we also see that hyperscale cloud emissions targets have been reversed due to AI investment, Datacenter growth is hitting grid capacity limits in many regions, and peaker plant and other non-renewable resources on the grid are being deployed more to handle this specific growth from AI. I think the author, by qualifying on "chatgpt" maybe can make the claims they are making but I don't believe the larger argument would hold for AI as a whole or when you convert the electricity use to emissions.

I'm personally on the side that the ROI will probably work out in the long run but not by minimizing the potential impact and keeping the focus on how we can make this technology (currently in its infancy) more efficient. [edit wording]

Thanks, Yes I'm a long time electricity maps customer. I also agree that the EU markets data transparency is further along in helping incentivize opportunities here.

Also thanks for the other links!

Yes we have an early PoC demonstrating significant gains. There are certainly levels of investment where consumption is managed well but that tends to be up to the point and at the DCIM level while we are working more on the compute level.

I wonder how this compares to the use of zarr format which seems to have similar goals and design and is already well integrated in several libraries (particularly xarray)?

Chopins posthumously published Nocturne in C# minor is a very popular piece to play/listen to. I also think there are several Debussy pieces he didn’t want published but are very popular now (maybe Reverie is one if my memory serves me)

Are you planning on open sourcing your code and/or model weights? Aurora code and weights were recently open sourced.

Couple of questions: 1. You link to fair.io but it seems there isn't a latest license there only a legacy version. Do you know what the current status is on what seems like a pending version update?

2. Your project repro actually links to the elastic license. Is there some relation between the two? Do you prefer the fairsource license over the elastic license>

Microsoft's officially reported carbon numbers don't include the impact of their software on devices sold by other OEMs.

This is a gray area of the GHG protocol on how to account for SW carbon emissions; some companies count it some don't.

[https://query.prod.cms.rt.microsoft.com/cms/api/am/binary/RW... pp 11.]

Maybe of some interest is my project https://openavalancheproject.org which is an attempt to improve/automate some of the forecasting of this with DL.

My goal isn't really a money making one though.

I worked in that area on a SCA trail crew in 1996, in the Bechlor ranger district, and its nothing we discussed then but looking as the ref it seems this loophole was only noted in 2008.

The area is very remote and while in the park not often visited. Its very beautiful with meandering rivers through large grasslands as well as some canyons with the Tetons in the distance.

There is an app for that. https://apps.apple.com/us/app/rockd/id1153056624 It may not have the most up to date info in this map but it does show the geologic info for your location.

"Replacing steel with aluminum in construction: never really happened although unclear why this would be a good idea." My house, built in 1976 in Washington, actually has aluminum framing as do the few houses next to it built at the same time. From what I heard there was a short period where aluminum was cheaper than wood for framing and caused the switch over. I'm not sure how localized this was. All the contractors I worked with find both the framing and how the house was framed unique but they are still able to work with it.

There is a companion guide which I think is really important if you actually want to read it without needing to study it.

Late 90's here. There are two things I miss: 1. Going to the book store and looking through the (mostly) O'Reilly books to see what there is to learn or what new technologies are coming out. 2. Having a huge pile of 3.5" floppies (or even CDs) you are using to load the latest version of Visual studio or whatever and just imagining how much knowledge is contained in those. I think my copy of VC++ was something like 21 floppies.

See this for an overview of the design guidelines--part of which is to break out app functionality: https://docs.microsoft.com/en-us/windows-hardware/drivers/de...

The surface-level forecast data is the NAM model which I pull from NOAA. Eventually I want to try the GFS (with a resolution tradeoff) or the EU models (with a cost tradeoff).

The current model captures data from three separate inputs: The NAM weather forecast model--primarily wind, air temperature and precipitation values SNOTEL data--on the ground snow values include snowpack depth and SWE

SNODAS-which has things like snowpack temperature and sublimation rates

The input combines these and has a lookback of ~21 days to allow for things like buried hoar frost. I do think there is work to be done to validate that buried weak layers are well represented in the model but currently the model is a pretty good representation of the human forecasts which do account for those.

I do agree that regional variation is a factor. One example is that I feature engineered a Long Term Cold feature which we know is a primary cause of snowpack faceting leading to buried deep layers. This is more of a continental feature of avalanches as opposed to a coastal. My model doesn't consider this feature important. I currently attribute this to being only trained on costal forecasts. I have reached out to both to the Colorado Avalanche and Utah avalanche centers to work on how to best incorporate their historical forecast data in to the modeling to assess this further.