Notices by Tero Keski-Valkama (tero@rukii.net), page 2

@icedquinn, I'm curious to know how those experiments worked out?

Hire to fire is something large corporations have been doing for years. It comes down to objectives set to hiring managers to lay off a certain percentage of their team members every year. This is a highly misguided attempt at social darwinism, and it is one reason for the skyrocketing salaries in fields where this is done under conditions of competitive recruitment.

In conditions of competitive recruitment where there corporations need to compete on their choice candidates, most hiring transactions raise the salary level. This person will take a mortgage, and because of policies, will be laid off next year. Now they are back in the market with a heavy mortgage and a newly established market salary valuation.

In aggregate, employee churn makes the salaries rise. If large US companies were more rational with their attempts to supress wages, instead of illegal salary cartels and anti-poaching agreements they would do what Japan does and hire for life, almost never lay off people.

Employee churn increases the market salaries and in the small pools of specialists the effects are large. At the next step it goes to housing prices, because market abhors discretionary income, and creates homelessness and suboptimal housing conditions, and societies becoming more inequal.

These misguided attempts at social darwinism keep being reinvented by people all the time without them fully understanding the full repercussions. Large corporations have huge impacts in markets and economies and should probably actually have chiefs of economics to veto ill-conceived policies.

"Today’s teenagers — and I’ve polled a random sample — neither know nor care what 'the web' is. They were born into mobiles and social media, and see no interest in reviving it as a semi-ironic cottagecore medium, like the cassette tape. #Web utopianism is strictly a Gen-X media phenomenon."

Google declares the end of the World Wide Web - UnHerd https://unherd.com/newsroom/google-declares-the-end-of-the-world-wide-web/

@icedquinn, yes and no. Our RL frameworks as popularized by Sutton & Barto decades ago are not what biological life does. It's not even what LLMs do. LLMs and deep transformers in general are able to do in-context RL near optimally, and this is very far from our capabilities using classic RL.

Classic RL doesn't work for a couple of reasons:
- Sparse rewards are the only signal for guiding actions. This works for simple games, not for the real world. Animals aren't learning by sparse rewards, they have complex intents and complex notions of success and failure, not just pain and pleasure.
- Classic RL separates the first-person as the agent, which cannot learn from other agents. This framework is for simple games where there are no heterogeneous ocean of other agents. The real world has this ocean of agency though, and it can be exploited/mined for third-person experience. Monkey see, monkey do.
- Sequential, discrete actions are too simple a framework to actually control real world bodies in a non-trivial fashion.

I can design a better, modern RL framework which is suitable for the real world though.

Foundation models for universal embodiment for the real world are needed for the next step after #AGI.

We'll soon have superhuman AIs in the digital space, using all sorts of digital tools like search, knowledge curation, collaboration and optimizers of different sorts.

However the real world is more messy, as anyone who has worked with robotics knows. It's possible to build expensive, heavy, rigid and standardized bodies for robots, train control models for them with massive scale reinforcement learning, and try to get their price down with mass production.

However, this doesn't scale. Add wheels or another arm and you need to do all this again. Variability in manufacture and condition causes models to become badly calibrated. Real world situations do not resemble the lab conditions enough and tasks won't work well.

This all is easy to see by people who have worked in robotics, but typically the people who make plans for dynamic robotics multimodal foundation models do not seem to appreciate this enough. They aim for lab demos, and have a rude awakening later. Short-term plans do not extend to long term.

I would rather suggest reformulating RL in a modern fashion, without a sparse reward signal and other badly considered axioms. Instead, we can observe the living world full of agency we have and learn from all the intent we can mine from the living world.

First person is an attended part of the ocean of agency, not a special domain which is unrelated to every other first person.

Google Genie showed we can do first person recognition for the first person with a clever information bottleneck, but we can actually similarly extend this to every agent we recognize in the signal, and then define the first person by attention over all the agency in the signal, injecting actions to the attended domains.

We also get a counterfactual for free, by not injecting actions, which allows us to recognize if our injected actions actually made a difference, that is, recognize the span of first person control.

https://sites.google.com/view/genie-2024

#UniversalEmbodiment

I don't think I have written it anywhere but deep neural networks have a unit bias. It's one of those things one just knows and doesn't even realize other people don't know that.

It's the reason why tree models generally win deep neural networks with tabular data. For decision tree based models one of the basic operations is a comparison, and those operations generally have meaning with tabular data. It has correct inductive biases respecting units of things. It doesn't sum ages to countries.

With neural networks the first operation is a weighted sum over *everything*. In tabular data that means ages, religions, countries, weights, dollars. What's 7 years plus 21 dollars? Does it make any sense? No?

That's why neural networks have hard time with tabular data. Their inductive bias is assuming that everything is in the same units.

Whether inequality worsens or not is a political choice, not something inherent in a technology. Technology creates well-being with decreased toil inputs. How we distribute the production depends on how you vote.

Automation tax is not the solution, we actually want all the sectors to automate instead of giving tax incentives for not automating. So we need increased business, capital gains and wealth taxes across the board to compel businesses to automate.

The taxes can pay for a UBI, and automation makes everything free. After a short transition time people won't need money for anything, and so no amount of money is enough to pay for anything. Money becomes both priceless and worthless.

AI is not for making money, money is for making AI.

AI to hit 40% of jobs and worsen inequality, IMF says - BBC News https://www.bbc.com/news/business-67977967

Everyone is doing recursive self-improvement with #LLMs now. I don't think super-exponential does it justice anymore. It won't be about being twice as intelligent than humans in a year, it's more like completely vertical.

To plan forward for even short time spans we must plan for systems with immeasurable intelligence.

[2310.02304] Self-Taught Optimizer (STOP): Recursively Self-Improving Code Generation https://arxiv.org/abs/2310.02304

One part of why chain-of-thought works is that it gives more time slots for the model to think about a complex problem. Often the chain-of-thought even produces longer sequences as the complexity of the problem increases.

Of course that's not all there is to it, but it is a significant part of it.

[2310.02226] Think before you speak: Training Language Models With Pause Tokens https://arxiv.org/abs/2310.02226

#NaomiWu and the Silence That Speaks Volumes

"When #China's prodigious tech influencer, Naomi Wu, found herself silenced, it wasn't just the machinery of a surveillance state at play. Instead, it was a confluence of state repression and the sometimes capricious attention of a Western audience that, as she asserts, often views Chinese #activists more as ideological tokens than as genuine human beings."

https://www.hackingbutlegal.com/naomi-wu-and-the-silence-that-speaks-volumes/

@kaia, exactly, I can't know that, and there is no trust left. The sales negotiations haven't been conducted in good faith.

I am looking for a new house but contractually I would in principle be obliged to let them fix the issues.

Troubles with buying a #house in #Spain. After having viewed many houses, we found a suitable one in #AlhaurínElGrande.

The #realtor assured us that everything is fine, it has all the papers and a #municipal #sewer connection.

Of course I needed #inspections to make sure so we made an initial offer agreement which states that the sellers will fix any minor problems that arise from inspections and otherwise and if not, the offer is void.

Then the realtor said that it actually has a #SepticTank, a proper one of the total oxidation type which is mandatory by law. Ok, they accepted my offer which was a bit lower than what they asked for so a septic tank is fine by me.

An inspection found leaking roofs, moisture rising from the floors, damp damage, #asbestos, bad design of using a sewage pump from downstairs toilet uphill, and failing A/C units, among other things.

Their lawyer spend weeks trying to track down the #LicenciaDePrimeraOcupación from archives, and when it became clear they don't have it, they tried to hide the fact from me. Instead they contacted the #mortgage #bank manager directly and negotiated the missing #LPO with him, I'm assuming with an implication that I would be later responsible for getting it.

Then I ask the realtor one more time about what sort of a septic tank it has as they had a plumber visit to fix some things and make sure. She says, it's actually a #PozoCiego (i.e. a #PozoNegro, an illegal hole in the ground). It's explicitly illegal on all levels from #EU #directives to municipal #regulations in Alhaurín el Grande, and the area is in the flow basin of #Guadalhorce natural #conservation area.

Ok, that's illegal and prevents an LPO being granted.

I tell them to replace the illegal pozo negro with a legal septic tank, get the LPO, fix the roofs and the moisture damage, and get rid of the asbestos.

They unsurprisingly refused. They say an illegal pozo negro is "not illegal, and is more environmental than a plastic septic tank which breaks down, and never needs maintenance". And a zillion excuses. "Most houses are moving to pozo negro now because it's better than a septic tank". "An LPO isn't needed, the house was legal when it was built and it wasn't needed when the previous owners bought it".

I gave them a chance to fix the deficiencies and now I'm waiting for the offer to expire so that I can get my deposit back, and because of dishonest behavior by the sellers I will send them the bill for my expenses for the external inspection.

The story continues in real time.

Scientists Beam #Space-Based #Solar Power to Earth for First Time

"#MAPLE’s array of transmitters successfully beamed #SolarPower collected in space using #microwaves to a receiver on the rooftop of Gordon and Betty Moore Laboratory of Engineering on #Caltech’s campus in #Pasadena."

https://gizmodo.com/scientists-beam-space-based-solar-power-earth-first-tim-1850500731