2025-01-31
Worth your time
Rohit on AGI.
Noah Smith on China Talk. Interesting throughout but what I enjoyed most was him talking about the types of posts he writes:
1. Things he already understands that are topical (he can just sit down and write)
2. Things where he is going out and doing the research because he thinks something should be better understood
3. Things he is passively interested in and just collects links as he goes, so when it becomes topical, he can go back and find those things and quickly write them.
I think I can adopt some of this into my own work.
The short case for Nvidia, but also a fantastic explainer on how different approaches to AI work. I now understand what makes Groq special.
Musings
Goals:
2025-01-24
I tried using Operator on a couple of tasks.
The most successful one was drafting an update to a document. We have a Partnerships document that we use as a part of our sales materials and we're onboarding a new partner. I needed to go to the Partner's website and draft some language for them to edit / approve.
I've been putting this off for a couple of weeks so I fed it to Operator. It took a little nudging, but eventually it was able to read through their site and write passable copy that I could refine and send to the partner for their review. So that's a success. Worth $200 a month? If I have 2-3 tasks like this each month, I think so.
I also tried seeing if I could get Operator to compile information for me (e.g., create a CSV file with the meeting dates and times of the Roswell City Council for the rest of January). It failed here in two ways:
- First, the meetings were wrong. It grabbed one that was in the past and then chose not to grab some of the meetings upcoming this week. It's possible that some of this was prompter error and could be fixed with more trials.
- There was no easy way to export the information. I wanted it to create a CSV file for me, but couldn't get it to. I do imagine that this will improve over time.
This past summer, I tried out several other products in this space (e.g., AutoTab) and this is a big step beyond what was possible then... but still not there yet.
More to come here over time!
2025-01-17
Worth your time
Uri Bram on Noble Lies
Zheng Dong Wang on productivity. I’ll be reading through each of the documents on his list at some point this year.
Why did everything take so long?
Principles by Nabeel Qureshi
Things I learned
Pine needle tea has more than 100 percent of the vitamin C of orange juice — Nautilus
The value of returned purchases in the United States would make it the 16th largest economy in the world — Rohit
Musings
“Great problems have to be discovered; often the solution of the problem is only a tiny part of the story, most of it is really about discovering the problem.” — From Michael Nielsen, ~Quick thoughts on research:~ (found via Zheng Dong Wang)
2025-01-10
A message from my sponsor:
OceanMade has announced pre sales of its Kelp Pots. These seed starter pots use kelp pulp to retain water instead of the traditional peat. The kelp pulp used in these pots is a byproduct of Macro Oceans beauty ingredient, Big Kelp Hydration. I’ve gotten a chance to see some of these up close and I’m really excited to see them coming together. It’s a small example of a big dream: using traceable, ocean farmed kelp products as an alternative to higher impact terrestrial sources. Due check them out if you’re a gardener.
Worth your time
Michael Lewis’s story about Chris Marks, a public servant who “led the development of industry-wide standards and practices to prevent roof falls in underground mines, leading to the first year (2016) of no roof fall fatalities in the United States”, is fantastic. Some gems:
At the height of the Vietnam War, a coal miner was nearly as likely to be killed on the job as an American soldier in uniform was to die in combat, and far more likely to be injured. (And that didn’t include some massive number of deaths that would one day follow from black lung disease.)
And
People facing a complicated problem measure whatever they can easily measure. But the measurements by themselves don’t lead to understanding.
And
Roof bolts were indeed more efficient and effective than timber supports in preventing chunks of roof from wounding miners. But they were expensive to install. The coal mine companies had, in effect, figured out how few roof bolts they needed to use to maintain the same level of risk their miners had endured before their invention
Materials we have run out of by Ed Conway
Noah Smith on Japanese urbanism. Having zones that restrict certain activities rather than prescribe what can be done seems like a small change with a big impact.
From Zheng Dong Wang’s fabulous 2024 letter:
The first awesome conclusion of the model does the eval is that we will achieve every evaluation we can state. Recall that evaluations must be legible, fast, and either a good approximation of a wanted capability or useful itself.
And:
Two years ago, ~Demis Hassabis enumerated~ three properties of problems suitable for AI: a massive combinatorial search space, a clear objective function to optimize against, and lots of data or an efficient simulator.
Things I learned
Musings
All large scale changes should be presented as a return to the past.
I wonder what it would look like to restructure local government around an escalating set of reviews. Imagine filing for a building permit where:
The first level of the form is evaluated by AI with the ability to appeal
The second level goes to a human
The third level goes to a supervisor
The second and third levels become new evaluation cases. This already happens today at places like YouTube, but imagine bringing it to your local government.
Quotes
Never confuse a single defeat with a final defeat
— F. Scott Fitzgerald via The Browser