California made post-employment non competes illegal. During employment non competes seem to be a bit more of a grey area. Employees are still governed by Conflict of Interest, Non-Disclosure Agreements, and Duty of Loyalty requirements. Working for multiple AI companies at the same time has got to violate at least one, if not all of those rules unless his job duties were incredibly well defined and orthogonal in scope for each position.

I wonder how much of this may be caused by Waymo cars being easily recognizable by human drivers and those dirvers, knowing there is no human in the Waymo, drive safer around them. For example, a human driver may not cut off a Waymo car, or jump out in front of them, because they don't trust the self driving car. In effect, by being recognized as a self driving car, the Waymo's produce a small bubble of safer driving conditions around them (kinda like student driver signs, if you see one of those you typically give the car a wider berth for your own protection).

At this point Airbus just needs to contract out a few more suspicious deaths and Boeing will take the fall.

Just kidding, or am I.....

While the vast majority of optical fiber out there is solid core, hollow core fiber (HCF) is a thing (note the hollow part is still only microns wide in diameter). People have been working on it for over a decade now and it is poised to be the next big thing in communications (last year, Microsoft bought Lumenisity, arguably the leader in hollow core technology). On paper it offers several benefits over solid core: lower latency, lower/no non-linearities, and potentially lower loss (not yet realized).

In the last year HCF has finally reached the point where commercial deployment is possible (although horribly expensive and limited in distance). HCF losses are still on the order of 2-5 times that of normal fiber but the lower latencies are real. There have be a handful of deployments in the real world, mostly driven by financial institutions (think high frequency trading).

This was from a post deadline paper presented at ECOC last fall (2023). Some extra details:

The 301 Tb/s was over 50 km of low water peak SMF (they didn't explicitly say what type but from the numbers they gave I'm guessing it's something in the ITU G.652.D family of fibers).

The traffic was spread over 1,097 24.5GBd channels (technically they only measured one channel at a time but moved it across the spectrum to simulate a full fill).

An interesting result however this is not a fiber to the home solution, it is a backbone solution and it should be noted commercial solutions exist today that offer 75+ Tb/s of bandwidth over just the C and L bands. So this isn't quite the 1.2 millions times faster they claim. It's only about 4 times the bandwidth that you can buy today.

So young policyholders (those starting their careers and earning less) with higher health risk (much higher premiums) in low-income areas (again earning less) are more likely to lapse (can no longer afford) their policies during economic downturns (spouse lost their job, hours reduced, etc).

So those who can't pay their premiums lapse their policy. Not exactly surprising.

From https://enroute.aircanada.com/en/aviation/airplane-tires/

Main-wheel tires have an average lifespan of 300 to 450 landings, while a nose wheel can withstand 200 to 350. (The nose wheel wears more when it pivots left and right to turn the airplane.)

So not once a week but a short haul flight doing multiple flights a day may see tires swapped as often as every 6-8 weeks.

These smaller layoff cycles can also be attributed to individual (or even group) performance issues as well. It is often much easier to layoff someone saying they are re-allocating resources and priorities than simply saying we are not happy with your performance. Not sure about Google, but in most large companies there is usually an entire sequence of events that has to be followed before letting someone go for low performance (warnings, remedial training, improvement plans, etc). By just saying it's a budget/priority call makes it much simpler. It often provides a softer landing for the affected employee as well (easier to get unemployment benefits, less of a black mark on their CV, etc).

As far as I can tell, they only really need to issue qualified invoices if they are selling to other businesses that use the qualified invoice to claim back the tax they paid.

This raises the question of does this matter? The article claims business purchasers give preference to qualified invoice providers because they can claim the tax credit on the paid tax. But to get this credit they need to pay the extra tax to begin with. If they went with a non registered provider they wouldn't pay the tax in the first place so they are still saving 10% and it is a direct savings, not a tax credit, which could be argued as better for the purchaser. So it's hard to see if this really a problem or not.

School is out. Not as many students looking to get ChatGPT to do their homework.

All kidding aside, regulation of the construction industry has continued to get worse and worse over the years. This includes the aforementioned health and safety but also means other sources of red tape like permits, funding, government inspections, etc. All of this adds overhead and generates a log of drag on productivity and it has only gotten worse in recent years.

Interesting they are still limiting salt to 230 milligrams per serving. I thought the experts were beginning to agree that salt isn't quite as bad for you as it has been made out to be in the 90's and 00's.

The traditional definition of chip with respect to Moore's law is a single die.

Moore's law basically implied the transistor density on the wafer would double every 18 months so either a die of the same size would be twice as performant or you could get the same performance out of a die half the size (so half the cost, assuming constant wafer cost).

If you defined chip as the physical package (not just the die) then you could argue AMD's chiplet process is kind of extending Moore's law. They are sticking more transistors into the "chip" via multiple smaller (and cheaper) dies. Not exactly the same thing but in a way it is carrying on the spirit of Moore's law. The viable number of transistors in a package is still growing, just via a different mechanism.

From DuckDuckGo's own help pages:

So yes they do use Bing results but not exclusively (apparently they use Yahoo as well for general links). Looks like they use their own crawler mostly for generating Instant Answers (the quick summary blurb results) instead webpage links.

Way back when I was in University, many of my STEM courses actually addressed this issue with multiple final grade weighting schemes (stated in the course syllabus on day one) . Your final grade would usually the better of something like:

20% Assignments + 30% MidTerm + 50% Final
or
20% Assignments + 80% Final

So in the case where things clicked at the end and you did really well on the final you could still pull off a good grade. Or if you did the work and had good grades throughout the course there was less pressure on the final. Overall it tended to work quite well.