A New Atlantis - A crazy solution to Building crisis

By Duncan Mcclements and Jason Hasusesnloy

Housing in Britain is unaffordable. The average British home costs 9.1 times median earnings to purchase - the highest since the unification of Germany - and rents are 27% of pre-tax monthly income in England as a whole, and 35% in London. For all this, Britons get homes that are smaller than in New York and demolished so infrequently that equilibrium entails an average house age over 1350 years old.

As many note, this is overwhelmingly due to Britain’s bloated planning restrictions. For example, in London, these restrictions essentially ban building in an area three times larger than the city itself, while very heavily restricting construction within its borders. We have previously estimated, conservatively, such restrictions cost 6% of GDP a year. Yet, they have endured largely because existing residents don’t want the inconvenience of housing construction and increased population density more broadly.

Well, if the difficulty is existing residents, we have a Wales-sized solution. Dogger Bank. Dogger Bank has a perimeter of 720 km and area of 17,600 km2, compared to Wales’ 20,600km2. It’s almost entirely within Britain’s territorial waters (sorry Germany). And it's only 15-40 m below sea level to boot - so at depths humanity has constructed large engineering projects before.

Headline results: we estimate raising Dogger Bank would cost £97.5bn, but would bring present value benefits of £622bn. Under the government’s standard method of cost-benefit analysis, this project would get a go-ahead, with a cost-benefit ratio of 6.2.

Methodology What exactly is involved in reclaiming land?
Well, a layman’s guide to land reclamation:
Build a wall around the area you want to reclaim.
Pump out the water.1
Fill it up with your desired material. 2
Develop.
To estimate the costs for Dogger Bank, we’ll simply add up the costs for each section.

Build A Wall

There are only two comparable projects of this scale, the Dutch “Oosterscheldekering” and South Korean “Saemangeum.” The Oosterscheldekering is a series of 65 concrete walls each 40m tall that form 9km of the Dutch Coastline. The project was completed in 10 years, from 1976-1986 at a cost of £6bn. The Saemangeum was much cheaper, a 33km seawall built in 2010, at an average height of 36m, at the cost of £1.94bn.

Using the South Korean numbers, which are both more modern and closer to the size of our project, we calculate that building a 720km seawall will cost £76.1bn. This, conservatively, considers that all costs, such as transportation and labour, scale at a similar rate to UK wages compared to 1998 Korean ones, which are ~86% higher than UK wages. We also adjust downwards to account for economies of scale, additional efficiencies gained from this large project size, adapting our figures from the equivalent measures in homebuilding, where a 100-fold increase in project size reduces price per unit by 20.5%.

To check our work, we compared this seawall cost to the proposed 700km North European Enclosure Dam (NEED), which the literature estimates will be in the range of £250bn-500bn.3 This corresponds with our results, because for comparable lengths, NEED requires construction at much greater depths, up to 300m in the deepest parts, compared to a maximum depth of 40m for DoggerBank.

Pump Out The Water

Delightfully, the Pump Express company publishes the costs to pump water using centrifugal pumps. For us, we calculate that we’re going need around 7.16mn CP230, their most cost-efficient model that can pump water high enough, which will cost £1150 per machine.

Unfortunately, the ground is leaky. This makes for a fun optimisation problem. If we assume that water seeps back into the area of ~1% a day, it turns out the cost-minimising solution involves keeping those 7.16mn of those machines on for 107 days, assuming an electricity cost of 9p/kwh, at 50% efficiency. This yields a total cost for this step of £12.1bn.

Fill It In

We’ll need to fill 634 cubic kilometers, meaning that we’ll need 2 gigatons of rock – this accounts for the 10-30% increase in volume required because the rock must be compacted. Using open quarries at a typical rate of $5/tonne, this will cost £7.9bn (or perhaps a great volume of dynamite and some less picturesque Welsh mountains). To transport these 2 gigatons, we’ll use commercial shipping and trucking rates of £1.22/tonne/1000km and £4.77/tonne/1000km respectively. This gives us a total to fill in Dogger Bank of £9.3bn.

Dogger Bank Reclaimed
Concluding, this gives a total construction cost of £97.5bn across the three steps, or 4.3% of UK GDP.4 That’s about as much as the full HS2.
Development

At the moment, DoggerBank is a barren piece of land in the North Sea, even wetter and windier than the mainland. We’re going to need some infrastructure.

We can split this calculation into two parts:

Infrastructure paid for by the end user, such as ports, reservoirs, power generation and telecommunications – which we don’t calculate because will be covered privately

And infrastructure provided by the government, think schools, hospitals and roads.

Based on our prior estimates, schools and hospitals cost £2277 and £2588 per capita, respectively. Underground lines in London cost £9100 per capita, but using Madrid's cut-and-cover method before house construction, costs can be reduced to less than a tenth per kilometer. Fares at London levels should drive net costs to zero.

Per capita costs will be factored into house prices. Roads cost £1194/capita (based on a £4bn maintenance budget, 5% discount rate). Courthouses and prisons have capital expenditure totaling £567/capita (assuming all capital spending is for building maintenance). Police and fire stations, managed locally, cost £19/capita and £51/capita respectively (based on Essex Police and North Yorkshire figures, including all capital expenditure).

The total budget is £6129 per capita, which translates to an implied cost of £14,464 per household (average size: 2.36 people).

Benefits

We’ll use our adaptation of Hsieh and Moretti 2019 to estimate the implied benefits. See our housing paper for all parameter estimates not undertaken here, and for a more detailed exposition of the model used.

We'll make several improvements to our previous calibration. We remove the variation in construction project size, as this will be a new city that can be constructed in larger chunks, assuming each is £100mn in size (the limit of the data). We now allow increases in buildable envelope, considering that in London today, 2.27x as much space is used on residential gardens as on residential buildings, and more space is used on private gardens than residential buildings in all but 5 of the 33 boroughs. These gardens add relatively little value to properties, so they would be removed by a free market.

Price index convergence is no longer relevant as all possible houses are built in the same region – we instead assume that all homes will be built at the combined London index.

As construction will occur upon a completely flat area the size of Wales over a 40-year period (the length under which the mobility assumptions are computed), there is little reason to expect prices to vary with quantity. Therefore, the Price Elasticity of Supply (PES) will be assumed to be infinite in all cases.

Levies are lower for reasons explained later, albeit calculated more comprehensively.

We retain the 8-storey assumption as this minimizes prices under our estimated price levels.

Project construction time will also vary, as this affects the present discounted value of undertaking it. The envisioned construction techniques should scale roughly linearly to time spent, making a 5-year construction schedule plausible. However, 20 and 40 years will be assumed for the central and conservative cases, respectively.

The most important change compared to the previous model is the inclusion of land prices, which is necessary to model cases other than merely increasing density. In the conservative case, we copy existing land prices wholesale, albeit under the retained assumption from the previous model of average building heights of 8 storeys, up from a presumed average height of 3 storeys in London and 2 storeys in the rest of the country.

For the central and stretch cases, we consider two additional effects from a possible increase in buildable envelope:

A direct effect of spreading land price over a greater number of homes, reducing unit costs. An indirect effect of reducing prices through demand-curve effects. Findlay and Gibb (1994) report that estimates of housing's price demand elasticity vary in the literature from -0.5 to -0.8. Assuming this maps onto land, we'll use the implied estimates of changes in land price.

Table 2: Estimated rents for the new city by assumptions - “recreating Birmingham” in this model means both having West Midlands housing construction prices estimated here and current wages. The North East shows no variation due to actual house prices being below predicted construction prices, with this being a result of falling populations causing a temporary disequilibrium. An area the size of Wales could generate economic activity beyond the cities located nearby. While potential mining or energy industries will be ignored, changes in territorial waters related to North Sea oil and gas could lead to substantial effects in the energy sector.

Agriculture, however, could produce more significant effects. Arable land is valued at £9,272/acre, or £2.3 million/km2. The proposed city would have population densities 2.7 to 5.3 times higher than present-day London, so even with the size of London it would only operate 3.3% of the island. The remaining agricultural land would then generate £38.97 billion worth of agricultural land as a one-time benefit, equal to 1.7% of GDP.

To calculate the value of these changes, we will adopt the British government's approach of using a 1% real long-run discount rate. For the sake of brevity, only imperfect mobility results will be reported, as perfect mobility results would be higher. It will be assumed that housing benefits from migration accumulate linearly over the 40-year period corresponding to the imperfect mobility coefficient.

Table 3: GDP gains under imperfect mobility, new city has productivity of London Table 4: GDP gains under imperfect mobility, new city has productivity of Bracknell We present the results for London and Reading and Bracknell, the two most productive Travel To Work Areas (TTWAs) in the UK. The results for all cities more productive than Birmingham are roughly similar, within a range of approximately ±25%, to those of Reading and Bracknell. However, for cities less productive than Birmingham, the model yields much lower values that typically do not pass a cost-benefit ratio.5

The results in this analysis vary much less across specifications compared to our earlier work. This is because the rent variation across different scenarios is much smaller, and the price elasticity of supply is no longer relevant for topographical reasons, leading to smaller changes in migration as well. However, the estimated GDP gains under imperfect mobility are much larger; this occurs because the model assumes decreasing returns to scale due to land constraints, so creating a new London with relaxed restrictions is considerably more valuable than relaxing restrictions in present London. It is important to note that the results are applicable for a wide range of plausible productivities that are high enough to justify the scheme.

Discussion

We present the costs of complete reclamation for Doggerbank for illustrative purposes only. An area much smaller than the size of Wales would be initially needed for any likely city construction program. London's current size is 8.9% of the proposed reclaimed area, and Singapore's is 4.2%. As costs for the wall decline with the square root of size, the cost-benefit ratio could be substantially improved compared to the value presented here, with agricultural benefits constituting only a small fraction of the total.

However, the project's success is based on the ability to continually grow a city to a size comparable to current cities. If the same British residents who currently oppose housing construction in existing cities move in, they may force the city to stagnate at a much lower size.

The project would be at little risk from sea level rises, as above-ground sea walls are generally inexpensive, costing only £700-£5400/m2 with low maintenance costs. Building a sea wall around the entire project would thus add at most 4% to the capital cost.

There is a possibility that this project could damage marine ecosystems, but it is likely less than any land-based construction. Oceans contain around 200 times less biomass than land per unit area on average, although this figure is likely higher for the North Sea compared to most oceans due to closer proximity to coastlines. Additionally, the project avoids the usual costs associated with land reclamation proposals, as it does not require large quantities of sand. It will only disrupt extant marine ecosystems in the quantity of seawater now occupied by the newest constituent nation of the UK, not elsewhere.

Conclusion

As with Mankiw and Weinzierl 2009’s famous study of optimal height taxation, there are two possible ways to interpret this post. The first is a simple reductio ad absurdum argument against current land use regulations - the planning system means that the British government would receive a greater than 6-fold return from reclaiming an area the size of Wales from the North Sea. The second is that it is a perfectly sensible suggestion similar to existing policies to circumvent NIMBYs in other areas - just as we escape the planning system’s strictures on wind farms by building them offshore, we can do the same for cities. We leave the correct choice as an exercise for the reader.

From Model Thikning Substack

Norway sovereign wealth fund 1.4 trillion $ assets

Germany unexpected crisis

Douglas Murray interview with Bibi

Frank Herbert Dune

Real story of the us Mexico border

Distribution of wealth in America

Are 5tate5 ju5tified in oppo5ing the revenue 5haring model

The 5ixteenth finance commi5ion ha5 it5 job cut out when they decide on the revenue 5haring model for the next 5 year5.

The primary ta5k of the finance commi5ion i5 to di5tribute revenue equitably 5o that the 5tate5 which are unable to geneate their own revenue have 5ufficient reour5ce5 to fund the development in their region. The commi5ion 5ought to balance the fi5cal need5 equity and performance for determining the criteria for horizontal 5haring

If we look at the table the weight5 u5ed in the devolution formula ,we can 5ee the income di5tance i5 the large5t weight.

Income di5tance i5 calculated by deducting the g5dp of the5tate with the highe5t g5dp per capita metric

5outh'5 lo55

with 5uch a low weight for improving their fi5cal condition 5tate5 are unlikely to take an effort to cutheir fi5cal deficit or improve tax collection5.Data from 5tate budget5 5how that 5ome 5tate5 are 5howing a larger 5hare from the divi5ible pool.Bihar 5hare contribute5 to 67.4 percent 5hare of their total 5hare revenue and UP it i5 42 percent 5hare of their total tax revenue.

On the other end haryana get5 only 13 percent of the total tax revenue.The 5outhern 5tate5 get le55 than 30 percent from their 5hare a5 per the devolution

Mini5try of finance relea5ed the data betwen FY 2019 and FY 2023 where it 5howed that Bihar and UP were getting 7.26 and 2.49 Rupee5 for every rupee they contributed to the Central coffer5. Wherea5 Mahara5htra, Haryana and Karnataka received 8 pai5e,14 pai5eand 17 pai5e re5pectively.

5ome economi5t5 believe that the weight5 a55igned to population i5 fair a5 5tate5 with more population will hvae more demand for ba5ic demand. There i5 thought that the weight a55igned to demo performance can be incea5ed to incentivi5e better human development performance to around 15 to 20 percent. al5o it i5 being recommended to reduce the weightage given to income di5tance in the future

**Hindu Bu5ine55 line article - By Loke5hwari 5K and Parvathi Benu **

Free market revolution in Latin america ?

By Axel Kai8er in Di8cour8e Magazine

BUENOS AIRES. I had the chance to speak for nearly an hour with Argentine President Javier Milei on December 9 of last year, one day before he was sworn into office. During our conversation we discussed the future of the libertarian revolution that is taking place in Argentina and his absolute determination to see it crystallized in concrete reforms that would restore freedom and progress to his country. Nearly two months into his presidency, there is no doubt that while much remains to be done, Milei is already off to a great start.

I’m writing this column from the iconic Palacio Duhau in Recoleta, Buenos Aires, where I have met several well-informed friends. They all concur that, so far, Milei is well on the way to achieving the unthinkable: putting an end to a century of collectivist decline. The lion of the Andes, as Milei is sometimes called, has not wasted time.

Shortly after coming to power, Milei dramatically narrowed the gap between the official and the market exchange rates by devaluing the peso 54%. He went on to shut down ministries and public offices and lay off swarms of useless bureaucrats. He also passed an emergency decree with 300 measures to deregulate the economy. Among them are the privatization of all public companies, the elimination of rent controls, an open sky policy, cutting subsidies to different sectors of the economy, ending import restrictions, deregulating satellite services and many others. In addition, the reduction of fiscal deficit is moving forward.

During the first month of Milei’s administration, public spending decreased by 30% in real terms compared to the previous year and the previous month. In other words, the government is already spending almost a third less than in the same period last year when adjusted for inflation. Needless to say, this is only the beginning of the 6.1 points of GDP worth of deficit spending that Milei has to adjust in order to restore a balanced budget. Most of this adjustment (3.2% of GDP) will affect the public sector by cutting spending, while a temporary increase of taxation (2.9% of GDP) will do the rest.

Despite the harsh measures adopted so far and the challenges some of them face in the courts and congress, Milei’s popularity has stayed at around 60%. Public support and the strong hand of Security Minister Patricia Bullrich explains why the demonstrations orchestrated by the infamous Argentinian unions have not been able to harm the government. If anything, they have contributed to increased public support for Milei’s efforts to fight what he calls the “cast” of “parasites” that have exploited Argentinians for so long. If he is successful in getting rid of the “cast” so that he can turn Argentina around, the ideological and political impact throughout the region will be enormous—even more so because he and other free market advocates have already achieved a lasting change in the mentality and values of millions of young people by replacing collectivist and statist ideas with notions of individual responsibility and freedom. Indeed, the good news is that already this is happening all over Latin America, not just in Argentina.

For instance, after my December visit to Buenos Aires, I went to Bolivia to give a series of lectures on freedom and the power of the spontaneous order of the market. Over 500 students attended my first lecture there even though, as I found out later, they had to pay for it. Other events organized by young local freedom advocates, such as Rodrigo Mundaka, also took place and had over a thousand participants, including business people and executives. This came as no surprise to me. Followers of freedom are to be found in the millions among the youth of Chile, Brazil, Colombia, Venezuela and Peru and are increasing their numbers every day. At the same time, socialist leaders are facing increasing resistance.

Chilean President Gabriel Boric, for example, has run the country into the ground with his statist ideology and anti-police stance. Now the Andean nation faces a dire economic situation and the worst security crisis in its history. As a result, 70% of Chileans reject his socialist administration, according to a recent poll. In Peru, communist President Pedro Castillo was put in prison after an attempted coup. His successor, socialist Dina Boluarte, has been forced to shift to more pro-market policies. In Colombia, Gustavo Petro faces a disapproval rating of 66% in recent polls as a result of his failed policies to tackle unemployment and his willingness to collaborate with terrorist groups. Even Brazil’s President Inácio Lula da Silva is struggling with maintaining his popularity, which is currently running at below 40%.

The collapsing public support of left-wing leaders in the region presents an excellent opportunity for political alternatives with a more pro-freedom stance. To some extent, a shift to market-oriented policies will be inevitable given the growing number of young people who are becoming libertarians as well as the Milei effect. But it’s not only the youth who are playing a decisive role in the region’s future. Business people in different parts of Latin America are more willing than ever to support libertarian and anti-socialist think tanks and organizations. The most notable cases are Ricardo Salinas in Mexico; Nicolás Ibáñez, Lucy Avilés Walton and Dag von Appen in Chile; Salim Mattar in Brazil; and Erasmo Wong in Peru. All of them have made crucial contributions to spreading the ideas of freedom, fighting against collectivism in their countries and beyond.

In addition, there are hundreds of free market think tanks and libertarian groups all over the region, with a considerable combined impact. Part of it is due to their active use of social media, which has proven critical in spreading libertarian ideas all over Latin America. It is easy to find YouTube videos of Milei, Agustín Laje, Juan Ramón Rallo and other like-minded Spanish-speaking public intellectuals with several million views, and it is no exaggeration to argue that they have more influence on public opinion than most—if not all—traditional television media. At the same time, demand for Spanish-speaking libertarian public intellectuals is exploding while more people are speaking up against socialism, government intervention and left-wing politicians in general.

Despite these unprecedented and promising developments, it is too soon to celebrate. Freedom can never be taken for granted anywhere, even less so in a region where collectivism still often seems ingrained in its cultural DNA. But one thing is certain: At long last, the region is starting to experience an intellectual revolution that is elevating liberty to the place it deserves. And, although this phenomenon still has a long way to go, it might change the course of history.

For there is one revolution with the potential to end all Latin American socialist failures: a freedom-oriented revolution capable of delivering lasting individual liberty, economic progress and dignity for hundred of million of people

California bill against ai

DEAN W. BALL

FEB 9, 2024 5

1 This week, California’s legislature introduced SB 1047: The Safe and Secure Innovation for Frontier Artificial Intelligence Systems Act. The bill, introduced by State Senator Scott Wiener (liked by many, myself included, for his pro-housing stance), would create a sweeping regulatory regime for AI, apply the precautionary principle to all AI development, and effectively outlaw all new open source AI models—possibly throughout the United States.

I didn’t intend to write a second post this week, but when I saw this, I knew I had to: I analyze state and local policy for a living (n.b.: nothing I write on this newsletter is on behalf of the Hoover Institution or Stanford University), and this is too much to pass up.

A few caveats: I am not a lawyer, so I may err on legal nuances, and some things that seem ambiguous to me may in fact be clearer than I suspect. Also, an important (though not make-or-break) assumption of this piece is that open-source AI is a net positive for the world in terms of both innovation and safety (see my article here).

With that out of the way, let’s see what California has in mind.

SB 1047

With any legislation, it is crucial to start with how the bill defines key terms—this often tells you a lot about what the bill’s authors really intended to do. To see what I mean, let’s consider how the bill defines the “frontier” AI that it claims is its focus (emphasis added throughout):

“Covered model” means an artificial intelligence model that meets either of the following criteria:

(1) The artificial intelligence model was trained using a quantity of computing power greater than 10^26 integer or floating-point operations in 2024, or a model that could reasonably be expected to have similar performance on benchmarks commonly used to quantify the performance of state-of-the-art foundation models, as determined by industry best practices and relevant standard setting organizations.

(2) The artificial intelligence model has capability below the relevant threshold on a specific benchmark but is of otherwise similar general capability.

The 10^26 FLOPS (floating-point operations) threshold likely comes from President Biden’s Executive Order on AI from last year. It is a high threshold that might not even apply to GPT-4. Because use of that much computing power is (currently) available only to large players with billions to spend, safety advocates have argued that a high threshold would ensure that regulation only applies to large players (I.e. corporations that can afford the burden, aka corporations with whom regulatory capture is most feasible).

But notice that this isn’t what the bill does. The bill applies to large models and to any models that reach the same performance regardless of the compute budget required to make them. This means that the bill applies to startups as well as large corporations. The name of the game in open-source AI is efficiency. When ChatGPT came out in 2022, based on GPT-3.5, it was a state-of-the-art model both in performance and size, holding hundreds of billions of parameters. More recently, and on an almost weekly basis, a new open-source AI model beats or matches GPT-3.5 in performance with a small fraction of the parameters. Advancements like this are essential for lowering costs, enabling models to run locally on devices (rather than calling to a data center), and for lowering the energy consumption of AI—something the California legislature, no doubt, cares about greatly.

Paragraph (2) is frankly a bit baffling; the “relevant threshold” it mentions is not even remotely defined, nor is “similar general capability” (similar to what?). This may be simply be sloppy drafting, but there’s a world in which this could be applied to all “general-purpose” models (language models and multi-modal models that include language, basically—at least for now).  

What does it mean to be a covered model in the context of this bill? Basically, it means developers are required to apply the precautionary principle not before distribution of the model, but before training it. The precautionary principle in this bill is codified as a “positive safety determination,” or:

a determination, pursuant to subdivision (a) or (c) of Section 22603, with respect to a covered model that is not a derivative model that a developer can reasonably exclude the possibility that a covered model has a hazardous capability or may come close to possessing a hazardous capability when accounting for a reasonable margin for safety and the possibility of posttraining modifications.

And “hazardous capability” means:

“Hazardous capability” means the capability of a covered model to be used to enable any of the following harms in a way that would be significantly more difficult to cause without access to a covered model:

(A) The creation or use of a chemical, biological, radiological, or nuclear weapon in a manner that results in mass casualties.

(B) At least five hundred million dollars ($500,000,000) of damage through cyberattacks on critical infrastructure via a single incident or multiple related incidents.

(C) At least five hundred million dollars ($500,000,000) of damage by an artificial intelligence model that autonomously engages in conduct that would violate the Penal Code if undertaken by a human.

(D) Other threats to public safety and security that are of comparable severity to the harms described in paragraphs (A) to (C), inclusive.

A developer can self-certify (with a lot of rigamarole) that their model has a “positive safety determination,” but they do so under pain and penalty of perjury. In other words, a developer (presumably whoever signed the paperwork) who is wrong about their model’s safety would be guilty of a felony, regardless of whether they were involved in the harmful incident.

Now, perhaps you will, quite reasonably, say that these seem like bad things we should avoid. They are indeed (in fact, wouldn’t we be quite concerned if an AI model autonomously engaged in conduct that dealt, say, $50 million in damage?), and that is why all of these things are already illegal, and things which our governments (federal, state, and local) expend considerable resources to proactively police.

The AI safety advocates who helped Senator Wiener author this legislation would probably retort that AI models make all of these harms far easier (they said this about GPT-2, GPT-3, and GPT-4, by the way). Even if they are right, consider how an AI developer would go about “reasonably excluding” the possibility that their model may (or “may come close”) to, say, launching a cyberattack on critical infrastructure. Wouldn’t that depend quite a bit on the specifics of how the critical infrastructure in question is secured? How could you possibly be sure that every piece of critical infrastructure is robustly protected against phishing attacks that your language model (say) could help enable, by writing the phishing emails? Remember also that it is possible to ask a language model to write a phishing email without the model knowing that it is writing a phishing email.

A hacker with poor English skills could, for example, tell a language model (in broken English) that they are the IT director for a wastewater treatment plant and need all employees to reset their passwords. The model will dutifully craft the email, and all you, as the hacker, need to do are the technical bits: craft the malicious link that you will drop into the email, spoof the real IT director’s email address, etc. Here is GPT-4, a rigorously safety tested model, as these things go, doing precisely this. (GPT-4 also wrote the broken English prompt, for the record).

What I’ve just demonstrated is with GPT-4, a closed-source frontier model. Now imagine doing this kind of risk assessment if your goal is to release an open-source model, which can itself be modified, including having its safety features disabled. What would it mean to “reasonably exclude” the possibility of the misuse described by this proposed law? And remember that this determination is supposed to happen before the model has been trained. It is true that AI developers can forecast with reasonable certainty the performance—as measured by rather coarse benchmarks—their models will have before training. But that doesn’t mean they can forecast every specific capability the model will have before it is trained—models frequently exhibit ‘emergent capabilities’ during training.

Imagine if people who made computers, or computer chips, were held to this same standard. Can Apple guarantee that a MacBook, particularly one they haven’t yet built, won’t be used to cause substantial harm? Of course they can’t: Every cybercrime by definition requires a computer to commit.

The bill does allow models that do not have a “positive safety determination” to exist—sort of. It’s just that they exist under the thumb of the State of California. First, such models must go through a regulatory process before training begins. Here is a taste (my addition in bold):

Before initiating training of a covered model that is not a derivative model that is not the subject of a positive safety determination, and until that covered model is the subject of a positive safety determination, the developer of that covered model shall do all of the following:

(1) Implement administrative, technical, and physical cybersecurity protections to prevent unauthorized access to, or misuse or unsafe modification of, the covered model, including to prevent theft, misappropriation, malicious use, or inadvertent release or escape of the model weights from the developer’s custody, that are appropriate in light of the risks associated with the covered model, including from advanced persistent threats or other sophisticated actors.

(2) Implement the capability to promptly enact a full shutdown of the covered model.

(3) Implement all covered guidance. (“covered guidance” means anything recommended by NIST, the State of California, “safety standards commonly or generally recognized by relevant experts in academia or the nonprofit sector,” and “applicable safety-enhancing standards set by standards setting organizations.” All of these things, not some—I guess none of these sources will ever contradict one another?)

… (7) Conduct an annual review of the safety and security protocol to account for any changes to the capabilities of the covered model and industry best practices and, if necessary, make modifications to the policy.

(8) If the safety and security protocol is modified, provide an updated copy to the Frontier Model Division within 10 business days.

(9) Refrain from initiating training of a covered model if there remains an unreasonable risk that an individual, or the covered model itself, may be able to use the hazardous capabilities of the covered model, or a derivative model based on it, to cause a critical harm.

Once a developer has gone through this months (years?) long process, they can either choose to self-certify as having a “positive safety determination” or proceed with training their model. They just have to comply with another set of rules that make it difficult to commercialize the model, and impossible to open source or allow to run privately on user’s devices:

(A) Prevent an individual from being able to use the hazardous capabilities of the model, or a derivative model, to cause a critical harm.

(B) Prevent an individual from being able to use the model to create a derivative model that was used to cause a critical harm.

(C) Ensure, to the extent reasonably possible, that the covered model’s actions and any resulting critical harms can be accurately and reliably attributed to it and any user responsible for those actions.

By the way, AI developers pay for this pleasure. The bill creates a “Frontier Model Division” within California’s Department of Technology, which would have the power to levy fees on AI developers to fund its operations. Those operations include not just the oversight described above, but also crafting standards, new regulations, advising the California legislature, and more. The human capital required to do that does not come cheap, and it would not surprise me if the fees ended up being quite high, perhaps even a kind of implicit tax on AI activity.

Taken together, these rules would substantially slow down development of AI and close the door on many pathways to innovation and dynamism. It is unclear, at least to me, if this law is meant to apply only to California companies developing models (hello, Austin!) or to any model distributed in California. If the latter, then this law would likely spell the end of America’s leadership in AI. I, for one, do not support such an outcome.

Investment trends in future

Indian box office 2023

New investment trends 2024

Best of CES2024

Architectural digest - Virat Kohli