Author: Gregor J. Rothfuss

Roof Koreans

While most are primarily armed with “America’s Rifle,” the AR15, the weapons are as varied as the people, from pump-action shotguns to FN Scars and expensive NFA items (such as machine guns). Similarly, the level of personal equipment sophistication ranges from walkie talkies and flashlights to encrypted radios and night vision.

Biosensor scaling

The tests are performing well, but are still being optimized. The biggest hold back at the moment is to scale from mass production (10s of 1000s of chips per month) to extreme mass production (10s of millions and, with time, 100s of millions of chips per month).

2021-10-07: This seems to have happened:

Covid-10 has driven a change in diagnostics technology. Many companies debuted new technologies for at-home test kits under the FDA emergency use authorization last year. “It’s going to go beyond Covid-19 testing. Everything from cancer detection to STD testing and anything that would go through the historical lab chain. We’ve gotten a lot of inquiries for rapid test kits from various companies about new applications. It’s going to be a revolution for the diagnostics market. To go from a test that would take 24-48 hours and cost $150 to get processed, to something you can do in 15 minutes for $20 or $30.”

Solving online events

It’s often struck me that networking events are pretty inefficient and random. If you’re going to spend 1 hour or 2 in a room with 50 or 500 people, then you could take that as a purely social occasion and enjoy yourself. But if your purpose is to have professionally useful conversations, then what proportion of the people in the room can you talk to in 1 hour and how likely is it that they’ll be the right ones? Who’s there? I sometimes suggest it would be helpful if we all wore banners, as in the image at the top, so that you could look across the room and see who to talk to. (First Tuesday did something like this in 1999, with different colored badges.)

This might just be that I’m an introvert asking for a machine to manage human connections for me (and I am), but there is also clearly an opportunity to scale the networking that happens around events in ways that don’t rely on random chance and alcohol tolerance. A long time ago Twitter took some of that role, and the explosion of online dating also shows how changing the way you think about pools and sample sets changes outcomes. In 2017, 40% of new relationships in the USA started online. Next, before lockdown, you would often have planned to schedule a non-urgent meeting with a partner or client or connection ‘when we’re in the same city’. That might be at some specific event, but it might also just be for some ad hoc trip – ‘next time I’m in the Bay Area’ or ‘next time you’re in New York’. In January most people would never actually have thought of making that a video call, but today every meeting is a video call, so all of those meetings can be a video call too, and can happen this week rather than ‘next time I fly to that city’ – or ‘at CES/NAB/MIPCOM’. In the last few months video calls have broken through that habit. I wonder what happens if we accelerate all of those meetings in that way.

On the unbundling of events, and how networking might be done better.

Stop Superspreaders

In our study, just 20% of cases, all of them involving social gatherings, accounted for an astonishing 80% of transmissions. Another 10% of cases accounted for the remaining 20% of transmissions — with each of these infected people on average spreading the virus to only 1 other person, maybe 2 people. This mostly occurred within households. No less astonishing was this corollary finding: 70% of the people infected did not pass on the virus to anyone.

On GPT-3

GPT-3 is scary because it’s a tiny model compared to what’s possible, with a simple uniform architecture trained in the dumbest way possible (prediction of next text token) on a single impoverished modality (random Internet text dumps) on tiny data (fits on a laptop), and yet, the first version already manifests crazy runtime meta-learning—and the scaling curves still are not bending! The samples are also better than ever, whether it’s GPT-3 inventing new dick jokes or writing (mostly working) JavaScript tutorials about rotating arrays. Does it set SOTA on every task? No, of course not. But the question is not whether we can lawyerly find any way in which it might not work, but whether there is any way which it might work.

Einstein Lost Hypothesis

any potential technology based on low-energy neutron production would be the first exploitation by mankind of the weak force, one of the 4 fundamental forces of nature. “I can’t imagine that there’s a whole force of nature out there, 1 of just a few, that is boring, disinteresting, and not of any use.”