Made a log of progress on Killer Apps. Finished Detection, Disruption, and Attacks and Counterattacks

Meanwhile, of the Fox News home page:

The census.gov website is dead:

Tasks

• Send note to Kat – done. She is not interested. Darn.
• Edit Detection section – done
• Add TACJ overview to P33. This is part of living with smart machines
• Make a ppt that has a web page in it
• Bills – done (dentist! tomorrow)
• Dishes – done
• Chores – done
• Laundry – nope, but the dryer is hooked up now!
• Review paper – finished reading and it’s much better. Still too wordy, but that’s not the sort of thing that is critical, since it has no impact on the findings, which are solid. Basically some formatting issues at this point.

More root stuff

The Ignite thing went well – it looks like it should be fun! Need to see how to get a webpage running in PPT that works with an LLM

Did some more work on P33

SBIRs

• 2:30 BD meeting. Capabilities maybe?
• I think I need to do two sorts for the best options. The first on accuracy, then the second on distance – done. Piece-o-pie
• Got animations in 3D pyplot working. I almost know what’s going on, too!

GPT Agents

• More Arms Control

Need to reach out to Markus Schneider for the Trustworthy Information proposal – done!

SBIRs

• Starting on the DSR-2291 task. Which I think is a movement along the two paths at a time $t$. My sense is that I should calculate thing whole thing and start the second path at the time that gives the highest value. And if there is no solution, don’t start. And if the probability of intersection is less than 100%, do a dice roll.
• Looks like more BD stuff. I wonder if the demo will actually get done?

Made some progress on P33. Need to reach out to Manlio De Domenico on that? Also Markus Schneider for the Trustworthy Information proposal

So here’s an interesting thing. There is a considerable discourse about how AI is wrecking the environment. It is absolutely true that there are more datacenters getting made and they – on average – use a lot of water and a good deal of energy.

But there are a lot of worse offenders. Data centers consume about 4.5% of electricity in the US. That’s for everything. AI, the WordPress instance that you are reading now, Netflix streaming gigabytes of data per second – everything.

But there are much bigger energy users. To generate enough tokens for the entire Lord of the Rings trilogy, a LLama3 model probably uses about 5 watt/hours. Transportation – a much larger energy consumer shows how small this is. A Tesla Model 3 could manage to go about 25 feet, or a bit under 10 meters. Transportation, manufacturing, and energy production use a lot more energy:

Source: Wikipedia

If you want to make some changes in energy consumption. Go after small improvements in the big consumers. Reduce energy consumption in say, electricity production (37%) by doubling solar, and that’s the equivalent of cutting the power use of AI by 50%.

In addition to energy consumption, data centers require cooling. And they use a lot, though that is steadily being optimized down. On average a data center uses about 32 million gallons of water for cooling.

Sounds like a lot, right?

Let’s look at the oil and gas industry. The process of fracking, where water is injected at high pressure into oil and gas containing rock from about 2,500 wells uses about 11 million gallons to produce crude oil. So data centers are worse that fracking!

But hold on. You still have to process that oil. And it turns out that for every barrel of oil refined in the US, about 1.5 barrels of water are used. The USA refines about 5.5 billion barrels of oil per year. Combine that with the fracking numbers and the oil and gas industry uses about 500 billion gallons of water per year, or 5 times the amount of data centers doing all the things data centers do, including AI.

To keep this short, we are not going to even talk about the water use of agriculture here.

So why all the ink spilled to talk about this. Well, AI is new and it gets clicks, but I went to look at google trends to see how the discussion of water use for AI and Fracking, and I got an interesting relationship:

The amount of discussion about Fracking in this case has leveled off as the discussion of AI has taken off. And given the history that the oil industry has in generating FUD (fear, uncertainty and doubt), I would not be in the least surprised if it turns out that the oil industry is fueling the moral panic about AI to distract us from the real dangers and to keep those businesses profitable.

Killer Apps

• More arms control

SBIRs

• 9:00 sprint demos
• 3:00 sprint planning
  • Training data file size sensitivity tests
  • Editor tool support for designing threats/assets
  • Transforming x to x’y’ (rotate and translate)
  • Optimize data creation
  • Single, randomized, trajectory calculation and intercept attempt for “real-time” demo
  • Multiple threat (raid) support
• USNA support

Meta has been busy:

Llama Stack defines and standardizes the core building blocks that simplify AI application development. It codified best practices across the Llama ecosystem. More specifically, it provides

• Unified API layer for Inference, RAG, Agents, Tools, Safety, Evals, and Telemetry.
• Plugin architecture to support the rich ecosystem of implementations of the different APIs in different environments like local development, on-premises, cloud, and mobile.
• Prepackaged verified distributions which offer a one-stop solution for developers to get started quickly and reliably in any environment
• Multiple developer interfaces like CLI and SDKs for Python, Node, iOS, and Android
• Standalone applications as examples for how to build production-grade AI applications with Llama Stack

P33

• Add section for stories that embody the egalitarian ethos – done
• Add Implementations section for examples that have worked in the past on parts of the concept. Done
• Also added a table of contents, since this was getting big.

Killer Apps book

• Work on the arms control section

Made some progress on P33. Need to reach out to Manlio De Domenico on that? Also Markus Schneider for the Trustworthy Information proposal

Tasks

• Medical folder
• Laundry – done
• Fan

Got a good start on the Project 2033 doc

Chores

• Clean house – done
• Dishes – done
• Bills – done
• Laundry –

SBIRs

• USNA. Looks at the progress in their doc. I didn’t see any evidence that they are doing any development atm. When does everything have to be done?

I think it’s a great time to re-think what a resilient representative democracy in the context of global, instantaneous, communication and smart machines would look like. I think that it is fair to argue that the run for liberal democracy (1945 – 2008) has become exhausted. One of the reasons that it no longer appears to have traction is that it takes a lot of work to get tangibly better living for many people. For this and other, more structural reasons (e.g. media ownership by the rich), The autocratic and authoritarian systems are winning globally.

So.

We need to figure out what structures an egalitarian system needs to thrive and work to implement them. This is the time to do it, and we have years to work it out while <waves hands> all this plays out..

• This is one of Colin Martindale’s primordial shifts
• “You never want a serious crisis to go to waste … And what I mean by that is, it’s an opportunity to do things you think you could not do before.”
• Needs some kind of trustworthy cognitive commons
• Reducing the gap between citizens and representatives
  • Sortition? Into small committees?
  • Guarantees of civil liberties and human rights
  • Online participation means some way of determining citizens from bots. Maybe take something from zero trust architectures?
• The role of money. It tends to produce kings, but it’s also really useful.
• The role of smart machines

My working title for this concept is… Project 2033

Assume existing power structures on the left become irrelevant over the next 2-6 years and it’s as bad as you think. People will tire of all the “winning,” and we need to have a plan in hand that looks attractive to (most) people who really just want something better than where they are now.*

* Now will be much worse in 2-6 years so this will be an easier pitch

SBIRs

• 9:00 standup – done
• Finish generating test data – done
• Slides for Monday? Done
• Look at what the Mids have been doing -tomorrow

I took a rough stab at what tokens cost today bases on working out the cost of a token per Watt-hour on a model like the 70B parameter LLama3 model if it were run on an Nvidia GeForce RTX 4090. Here are my estimates for some pretty hefty books, if a LLM were to generate the same number of words:

• Moby Dick – 163,500 words – $1.44
• The Lord of the Rings trilogy – 733,022 words – $6.44
• War and Peace – 587,554 words – $5.16

It’s not much! My sense is that most interactions use a small fraction of a watt-hour, and a bug TPU like the A-100 is probably even more efficient than and RTX 4090. So if you are paying $20/month for a big model, unless you generate something like four War-and-Peace-like mountains of text, the companies are making a profit. The spreadsheet is here, if you’d like to play with it:

SBIRs

• More trade show demo. Good progress. I think the train data generation is mostly done, now I need to do random test data
• Hmm. Looks like the entire 8a set aside system may go away By my estimate, 8(a) is a $50 billion target, so I think sooner rather than later

GPT Agents

• 3:00 Alden meeting – done
• Missed Peter’s meeting somehow. I don’t think I was provided with a final date/time?

This is really interesting – from Instagram this morning. Need to add it to the trustworthy information proposal:

For comparison, here are Wikipedia page views for Democrat and Republican, along with the disambiguated pages for the parties in the United states, from election day to the inauguration. The number in the legend is the cumulative views for that period.

Instagram is doing some seriously untrustworthy things. Need to update the proposal to include this.

NBC is also manipulating things (via BlueSky)

And if you look at the audience at the time he does it, you can see that some recognize what it is. And they are thrilled:

Vacation plane tix!

SBIRs

• Add offset to trajectory for another training option
• 9:00 Standup
• 3:00 Tradeshow demo

Trying to decide if I want to watch the Washington Post whither away or switch to the Guardian

Found these two items on The Decoder:

• Google has signed its first direct content partnership for an AI chatbot, teaming up with the Associated Press (AP) to feed real-time news into its Gemini system. The deal means change is coming for new publishers if chatbots are successful products.
• The new “Titans” architecture takes inspiration from how human memory works. By combining artificial short and long-term memory through attention blocks and memory MLPs, the system can work with long sequences of information. Titans uses “surprise” as its main memory metric – the more unexpected a piece of information is, the more likely it gets stored in long-term memory. The system also knows when to forget things, helping it use memory space efficiently.

Compile and run Joseph Weizenbaum’s original 1965 code for ELIZA on CTSS, using the s709 IBM 7094 emulator. (GitHub)

Got the Senate testimony chapter finished yesterday. Today I start working through the analysis. Also, I need to add this to the vignette 1 analysis. And to the slide deck for the talk. Maybe even start with it.

4:30 Dentist

Really good example of AI slop and its consequences.

Tasks

• Bills
• Chores
• Laundry
• Ride with David
• Dinner with Barb

Tasks

• Groceries – done
• Cleats – done
• Flight tix 11:00 am 31 May (BWI -> Seattle) to 18 June (Boise -> BWI) – nope. Tomorrow?
• Got a cold late lunchtime ride in

SBIRs

• 9:00 standup – done
• 12:50 USNA – better than last week, at least
• 1:30 Demo meeting – Not sure if we really did anything useful
• Generate 5×5 HGV grid – done!
• Nice pic to end the work week on: