obblekk

Maybe the most unambiguous "ai will automate work" example I've seen yet.

Absolutely does not imply the workers are automated since they can now use the current models to do more complex tasks at the vast number of new AI training data startups.

Turk was simply not designed for greater complexity tasks and so much of their lunch has been eaten by startups specifically built to collect AI training data.

Maybe I’m an outlier but I’d rather this than accidentally block legit sites.

Otherwise this becomes just another tool for Google to wall in the subset of the internet they like.

This is the first time in recent memory that software has had high variable costs so the surprise at these rules is understandable.

In this case, a the difference in context cache hit rate between openclaw and antigravity.

For example if openclaw starts every message with the current time hh:mm:ss at the top of the context window, followed by the full convo history, it would have a cache hit rate if ~0. Simply moving the updated time to each new message incrementally would increase hit rate to over 90%. Idk if openclaw does this but there’s many many optimizations like this. And worse, thrashing the cache has non linear effects on the server as more and more users’ cached contexts get evicted from cache due to high cardinality. The cost to serve difference could be >10x.

Google is the furthest behind on coding agent adoption and has all the incentives to allow off policy use to grow demand. But it would probably be better to design their own optimized openclaw and serve that for free than let any unoptimized requests in.

One big reason supersonic can be economic now is the increase in wealth in Asia since the 80s.

Transpacific flights from California have no sonic boom population issues for 90% of the flight, and there’s already a large market of people spending $10k on business travel.

Reducing travel time from 12hrs to 4hrs would be a product with a lot more demand than 7hrs to 3hrs to Europe.

I see 25-29% here https://www.swebench.com/viewer.html for models released in Nov 2024 albeit not verified. gpt4o (Aug 2024) was 33% for swe bench verified.

Important point because people have a bias to underestimate the speed of ai progress.

80% on swebench verified is incredible. a year ago the best model was at ~30%. i wonder if we'll soon have a convincingly superhuman coding capability (even in a narrow field like kernel optimization).

this is the most interesting time for software tools since compilers and static typechecking was invented.

1yr timeline is ambitious if it means fully deployed.

Clearly the right thing for Sweden and others to do. Also worrying that even 3yrs into the Russian invasion, bordering countries are urgently increasing their preparedness for future conflicts.

This is awesome. Coolest hacker demo I've seen in a while.

These critiques of zirp never explain what should have been done differently.

It's very unlikely the Fed kept rates too low because inflation didn't exceed 2% for 12 years from 2008 to 2021. If the Fed had actually made money "too cheap" inflation would have kicked in much earlier.

More likely, we made it very hard to do new stuff in the country, which made the value of borrowed money low.

Congress spent decades adding regulations (often for good reason) that ultimately resulted in it being too expensive to do most things inside the US. If a business is banned from investing in most new things, it won't need to borrow more money.

The Fed just responded to the market's appraisal of money value. You can even see this in the long term charts - interest rates declined almost continuously from 1984 to 2022.

The entire time, inflation stayed at or below 2%. If the Fed had kept rates higher, theory would predict they would have caused a recession (and Scott Sumner has spent more than a decade arguing this is actually what caused the long deep recession of 2008 - money was too expensive, even at 0%).

Ultimately money is neutral in the long run. The things that truly matter for growth are laws, culture, natural resources and education. These are the causes of our present social dysfunction. These are the issues we should focus on fixing. Not fiddling with interest rates.

Have people been dropping cursor usage for Claude code? I have dropped to using cursor as just an ide with auto complete. Curious if others are doing this too.

In the abstract it discusses that most of the effect (16deg Celsius) is from reduced radiative heating and only a few degrees from evaporation.

Mostly the benefit is instead of having the concrete under you absorb and emit the sun, the leaves above you do.

This dramatically reduces the heat we feel at human height.

A lot of normal consumers pay $20 a month for ChatGPT. I think most software gets bid down in price bc the marginal costs are zero. Where it’s not (llm token generation) prices don’t plummet and consumers build a different expectation.

this might be a good thing if viewed from the opposite perspective: people with kids/elderly parents usually can't afford to pay as much per person as people traveling alone for fun/corporate travel.

This is incredibly well written.

The oil example is very compelling for import substitution. And the covid example is interesting in showing the savings rate only went up as an offset of gov spending.

I'd love to see a follow up on (a) is it important for the US to increase domestic savings and (b) what are the best policies to do so, and why are they the best?

I imagine blanket tariffs might actually increase the savings rate because they increase the cost of importing all goods when the domestic alternatives are either inferior or more expensive. But I'm curious if they are the best way to achieve the savings goal.

> Having access to goods and service is what gives you a better living standard

I think the good faith critique is access to imports can be taken away by the other country if they want. eg. rare earth metals. So being too heavily reliant on imports without the capacity to produce domestically is less long run access

The open source ranking library is really interesting. It's using a type of merge sort where the comparator function is an llm comparing (but doing batches >2 for fewer calls).

Reducing problems to document ranking is effectively a type of test-time search - also very interesting!

I wonder if this approach could be combined with GRPO to create more efficient chain of thought search...

https://github.com/BishopFox/raink?tab=readme-ov-file#descri...

> Texas is the only state in the lower 48 that has no major connections to neighboring power grids. That means growing energy demand in Texas must be met by new power generation in Texas.

> "Texas didn't [build solar farms and battery storage] for an energy transition reason at all," says Rhodes, "We just made it easy to build things here. And so people started building things here."

It's hard for me to reconcile the criticism of Texas's market structure with the fact that they seem to add more (clean) generation capacity than any other state.

Is it as simple as: Their state is growing very fast (both people and electricity per person due to datacenters) leading to supply/demand forecasting difficulty and mismatches?

TX generates 138% more electricity in 2023 vs. 2009 [1]. By contrast CA generated 5% less electricity in 2023 vs. 2009 [2] (roughly same for any 2 years around those start and stop years).

I do wish this topic were treated less politically and more pragmatically. At this point everyone seems to agree we need more electricity in the future with less carbon intensity and so far, TX is one of the few states actually achieving it at scale.

[1] https://www.eia.gov/electricity/data/browser/#/topic/0?agg=2... [2] https://www.energy.ca.gov/media/7311

There should be a high sales tax on all advertising. It's the perfect thing to tax:

- Luxury product companies advertise the most (utilities rarely do). The tax will decrease consumption of the most economically "unnecessary"

- Advertisers charge more money to show ads to rich people than on poor people - even a flat % tax will naturally impact the rich more than the poort

- Prices are set by auction at the maximum of the advertisers willingness to pay - this means tax revenue will hit the supplier of ads not the buyer

- Tax on ads is fundamentally a tax on consumption not income

- It's a zero sum positional product - ads make use of finite physical space/attention. This is close to a negative externality that people cannot contract away (eg i cannot contract the billboard company to not show me ads for $5/mo). On the web, this could technically be done, but the marginal value is so low and the number of billboard providers so high that transaction costs prevent it from happening

I am actually willing to have space ads if there were a super high tax on it. Imagine to show a message to all of humanity for a week, you must first pay $100B of taxes to all of humanity. That might be worth it.

I've been surprised there's less discussion of advertising taxes than much more complicated things like land value taxes, even though advertising/marketing is 1/5th of GDP in developed markets.

I hope someone reruns this on o1 and eventually o3.

If o1-preview was the start like gpt1, then we should expect generalization to increase quickly.

Human performance is 85% [1]. o3 high gets 87.5%.

This means we have an algorithm to get to human level performance on this task.

If you think this task is an eval of general reasoning ability, we have an algorithm for that now.

There's a lot of work ahead to generalize o3 performance to all domains. I think this explains why many researchers feel AGI is within reach, now that we have an algorithm that works.

Congrats to both Francois Chollet for developing this compelling eval, and to the researchers who saturated it!

[1] https://x.com/SmokeAwayyy/status/1870171624403808366, https://arxiv.org/html/2409.01374v1

The votes are strong majorities of both houses - great to see a functional democratic government act on such an important issue.

They specifically don’t prescribe any particular age verification methods. This would be a great time to follow up with legislation that updates their national IDs to be able to provide cryptographically secure proofs of age without leaking identity.

Absent that, I’m sure many of the comments to come will worry about the privacy implications. It really would be great to see the government act with expertise to solve the problem in a way compatible with a free and open society.

Possibly at that price there’s not enough demand. Put another way, it’s possible there’s no price that covers the cost of maintenance per charge and attracts enough people to get enough charges.

Worth noting this model has 50% more parameters than llama3. There are performance gains but some of the gains might be from using more compute rather than performance per unit compute.

Very impressive if the benchmarks replicate. Some questions:

* token cost? In multiples of Gemini pro 1

* memory usage? Does already scarce gpu memory become even more of a bottleneck?

* video resolution? Sherlock Jr (1924) is their test video - black and white, 45min, low res

Most curious about the video… I wonder if RAG within video will become the next battlefront

This might be true but paradoxically, there hasn’t been an observed frequency or intensity of Florida hurricanes in last 100 years [1].

Florida is worth looking at because they were industrialized and have had relatively robust meteorological record for the entire period.

I do wonder why as it seems like common sense that heating air and water temps (which have happened) would cause more evaporation and potential for water and wind storms.

[1] https://www.gfdl.noaa.gov/global-warming-and-hurricanes/

This is super cool. I've often wondered why google scholar didn't use pagerank over the citations... and now you've done it! Props!

Tesla software has such a large surface area to attack compared to traditional OEMs where most subsystems are effectively air gapped.

It’s their advantage when it comes to coordinated user experience, but might become a problem as the fleet size increases and becomes a more valuable hacking target.

I could see security becoming a bigger part of CarPlay/Android auto’s pitch to OEMs.

I agree with the sentiment more than the solution.

In the American system, the gov should assess punitive fines for regulatory failures, revoke self regulation privileges, and impose consent decrees on company behaviors.

The combined impact of these actions would cost Boeing billions which would result in shareholder action to replace management.

Perhaps the gov should also investigate willful neglect on the part of management. If it’s found that engineers complained about policies for reasonable safety concerns and management ignored them, then personal liability, perhaps even criminal, would extend to the directors and executives.

I don’t know how much of this would happen, or be found guilty in court. But I do hope this isn’t just another administrative slap on the wrist like 2019.

Places with the saltwater don't want to build large pipelines to send it to places without.

The places without are so dry they cannot drill, condense or truck in salt water, even though they have plenty of solar.

That's a coordination problem with public rights of way issues which is why it wouldn't happen without specific government policy.

There should definitely be a plan for the brine.

I'm optimistic we could:

* Build ocean pipelines that disperse brine over much larger volumes of water to minimize impact on wildlife

* Dump the brine water in engineered desert fields where it will evaporate and deposit salt reserves.

In general, we should be aware of the whole lifecycle of a solution, and optimistically believe we can solve each problem with more work.