peterlk

I am never sure which notes I am going to reference again. Most of my daily notes go unread forever. But occasionally, that seemingly trivial thing that I jotted down in a meeting turns out to be _very_ important. I have AI assistants that take notes for me too, but somehow I can’t stop taking notes in meetings - and occasionally this has saved me (note taker bugged out).

Daily notes are also where I do a lot of scratch pad thinking, and occasionally those come in handy as well.

Hijacking this comment because it seems to be the most recent. I’d love to start a private email thread with you if you’re willing to do so.

After the previous election, I felt wildly out of touch, and I’ve been trying to find people who I can talk to who might be able to help me improve understanding of my (metaphorical) neighbors. I suspect that our beliefs will be quite different, but I appreciate your willingness to provide long-form, respectful responses to so many comments here.

If you’re up for it, my email is in my profile

Answer: it would most likely burn up

Original paper: https://www.sciencedirect.com/science/article/pii/S009457652...

Creating stem cells from blood samples is a well-established industry practice now. I don’t think limiting embryonic stem cells research is significantly hindering stem cells research, is it?

I think there are maybe 5 business books out there. I’m not sure how exactly I’d define the 5 different business books, but I think of you read 10-15 business books you’ve pretty much read them all. After a while, they all start boiling down to the same few points with differences in narrative content. If I were to take an unconsidered stab at a few of them: hard work + luck is about the closest formula anyone has found for success if applied over long time periods; you have to be disagreeable and believe in yourself, but not so disagreeable that you can’t get along with anyone; people are important, and treating them well leads to better businesses (over long time periods); sometimes you get dealt a bad hand.

I sent outputs to experts (friends )in those fields who provided me with some specific topic that is common to their niche. In general their feedback all aligned with: “yeah, this is pretty decent, but anyone who was a professional would know about <some hole in the output>”.

I think you’re fixating on the specific example that I cited rather than imagining the possibilities of the technology. We now have: zero-shot classification capabilities for basically any task, performance that is consistent enough for independent LLMs to collaboratively produce robust long-form responses, the ability to produce almost any web UI element on command with functional hookups to an API. And that’s just LLMs. SAM2 has good performance for realtime object detection in video.

Perhaps a more fruitful line of inquiry could be: what would the internet look like if every web application implemented support for A2A?

I am surprised at the negativity and cynicism in this thread. I suppose the pendulum of hype has high amplitude.

Just because AI isn’t going to be some kind of all-knowing sci-fi AGI doesn’t make it all a sham. The recent research models are miracles of technology - we can now get in ~15 minutes an undergraduate-level report (that would take an undergrad days or weeks) on almost any topic. That’s incredible! The capability of an AI model is approximately junior level in the fields I’ve tested it in (programming, law, philosophy, neuroscience). If you’d don’t see any possible uses for the technology, keep thinking about it.

I think we are saying slightly different things. COGS are composed of many smaller capital allocations. According to this untested, pet thesis, putting on a report that $250M was spent on capex is just fine; but if you go to a single vendor and sign a $250M contract, you have wasted money by not being more careful about how that capital is allocated. $100M is _a lot_ of capital, and I think it’s easy to lose sight of how much stuff you can do with that much money when applied to industries that don’t pay tech salaries for speculative growth. As examples: how many pounds of food could you grow for 100M? How many doctors could we train for 100M?

I think the thesis is thought provoking. Not sure yet if it’s worth anything, but it also doesn’t preclude businesses from having massive cashflow.

Yes; you will burn through all the capital you raise in ~18 months. It is _extremely_ difficult to efficiently allocate large raises (100M+) in 18 months. In fact, I’m developing a pet thesis that no single human or business can efficiently allocate more than $100M. This would imply that any time a single raise is more than 100M, the investors always would have had a better return by splitting it into chunks of 100M or less. It’s not a _good_ thesis yet, just one I’m performing thought experiments with

I have no kids, but perhaps positive role models can be found in more specific categories like music. For example, I think Jacob Collier sets a great example. And I think my parents had similar concerns. I was raised on gangster rap and Eminem, and my parents were very concerned about that. But overall, I think I turned out alright. My feeling is that it’s much more important to have role models closer to home. The light that Elon Musk and Andrew Tate are cast in can be greatly influenced by the people around your kids if not directly by parents

I disagree. I think Anthropic, like the other big players, is trying to get some of that government money. Releasing policy-adjacent papers seems like a way to alert government officials that Anthropic ought to be in the room when the money starts changing hands.

I think it’s worth accompanying this study with the Simpson’s Paradox [0]. I don’t want to dismiss the conclusions of the paper, but temper expectations for individuals considering lifestyle changes.

[0] - https://en.m.wikipedia.org/wiki/Simpson%27s_paradox

Totally unrelated… I’m really excited to see VR46 as a username, especially since the championship was today

This seems intuitively true to me. The more I run, the quicker my heart rare matches whatever I’m doing. Resting - low heart rate; running - high heart rate.

If you watch elite runners, they can go from 2+ hours of sprinting (elite marathon pace) to giving an interview in 30 seconds

I went to Thailand shortly after the covid lockdowns ended, and the corals were all coming back. I talked to a boat captain about it and his comment was: “yeah, there were no tourists to abuse them”.

I can’t help but wonder what effect tourism (and lockdowns) has had on corals. For example, It is common among scuba divers to only use mineral sunscreens because standard off-the-shelf sunscreen floats on the top of the water and blocks UV from the corals

A recent technique that I’ve been using is to listen to Spotify radio of my own music. Nobody listens to my music, so I end up finding other music that no one listens to - it’s refreshing. I’m sure you can find a friend who has a garage band or who is a bedroom producer.

The thing is, you have to be willing to listen to music you don’t like. This is a fundamental part of digging and any workflow will expose you to music that sucks

The article addresses this point with the following:

> It would of course be possible to sic Chatty-Jeeps on the raw markup and have it extract all of this stuff automatically. But there are some good reasons why not. > > The first is that large language models (LLMs) routinely get stuff wrong. If you want bots to get it right, provide the metadata to ensure that they do. > > The second is that requiring an LLM to read the web is throughly disproportionate and exclusionary. Everyone parsing the web would need to be paying for pricy GPU time to parse out the meaning of the web. It would feel bizarre if "technological progress" meant that fat GPUs were required for computers to read web pages.

Based on what I've read in the past, I believe the secret is simplicity. Simplicity scales

Can't wait for the writeup! So many services down at once... Something very interesting must have happened

The largest issue with the fediverse is that so few people seem to understand what it actually is. Even this paper talks about Bluesky, Mastodon, and Threads as if they’re mutually independent. They kind of are, but that misses the point of what the fediverse is meant to be.

I am part of a public mastodon server and the academic migration was like watching people discover what it was like to “surf the web” again. This paper generally rings true to my experience. In my view, this particular failure of fediverse adoption was because there was a large group of people who were using immature technology for the wrong reasons, and the friction of their incorrect mental models outlasted their righteous indignation about Twitter/X.

If you haven’t taken a gander through the fediverse yet, give it a try. It’s beautiful and weird the way the internet was in the days of geocities. But it can take some digging - my best results have been in finding people who are running custom, non-mastodon servers.

When we stayed in Bologna for a month, we were told that it was used defensively. You can see a very long ways from the top of the central tower (I think it’s closed at the moment). It would have been nearly impossible for an army to approach Bologna without several days for Bologna to prepare

Note that this follows on the heels of the recent (near?) death of California Senate Bill 829 [0] - which is taking aim at the vertical integration of live performance by Live Nation. If you’re in CA, call your representatives!

With this in mind, to continue your rant…

The anticompetitive behavior of Live Nation is toxic to the music and performing arts industries. It’s the Walmart effect, except consumers end up paying way more because exclusivity contracts prevent competition. Artists sign exclusivity contracts with Live Nation, and onlt play Live Nation Venues. Venues purchased by live nation outbid independent venues until those venues sign exclusivity agreements with Live Nation or die, and then Live Nation pushes up the price.

[0] https://legiscan.com/CA/text/SB829/id/2796147

The mismanagement at Dish and DirecTV is staggering. The pendulum has now swung so far in the fragmentation direction that there is real value in bundling disparate content in a restricted fashion, and charging people for it - like cable/satellite bundles. Except now, the business models available are much more diverse because the telemetry of streaming is so much better. Where did the masters of media bundling go?

As a consumer, I really dislike that the media industry has made it my problem to figure out who owns the rights to something. Just take my money and let me see the content. Having to google “<movie> streaming” to see a list of options in different streaming providers who have varying levels of service at different tiers of their paid subscriptions is inane.

Whoever figures out the bundling problem will own a lot of eyeballs, and probably be able to monetize on ads too.

The quoted section of the article is pretty out of touch with modern music. Many (most?) genres of EDM place the importance of timbre and texture above harmony and melody

This is the problem with terms like this. There are definitions in academia/industry that differentiate between awareness, consciousness, qualia, intelligence, sentience, and more. But these distinctions require some pondering to understand and are divergent from general understanding which treats them as roughly the same thing. For example, consciousness for me is the ability for something to take directed action to affect their surroundings. This means that a robot can be conscious. But now the question for what we’re talking about here is: is it aware? It is possible to display consciousness without awareness (see blindsight studies), but to my knowledge it is not possible to be aware without consciousness. And then we get to sentience, which is still very slippery, and often relies on definitions involving “qualia”, which is also quite slippery.

There is nothing in the world that is as good as soundcloud was in 2012. It wasn’t scalable at that time, but it was an incredible melting pot of discoverability. I used to spend a lot of time on a site called CitySounds (it no longer exists) that was a geofenced soundcloud firehose. It allowed me to listen to the latest music from bedroom DJs in Sao Paulo or a live polka band from Vienna or a Russian children’s recital. Most of it was garbage, but it was diverse and interesting garbage, and that made it compelling to me. At the time, I ran a radio show, and being able to connect with a high school kid in Italy who was making cool hip hop beats was a way more meaningful human experience than playing another Pitbull track. And I think our listeners appreciated it too.

It was as close as the world could practically get to copy-left, remix culture, and they threw it away because the founders lacked guts or vision or both, and the company was taken over by people who believe in a homogenized distribution model. It hurts my soul that we don’t have tools like old soundcloud - with the exception of last.fm, which keeps holding on. Last year, I would have said that bandcamp is carrying the torch, but I don’t think that’s true any more… perhaps the fediverse has an opportunity to step in here… perhaps we just need a new crop of founders who believe in a world full of diverse musical culture.

Edit: I'll also add that the world lost an amazing tool when Echonest was bought. It would have been so cool to see that product blossom into a platform for general purpose music production and discovery. But instead, we get access to a nerfed version locked behind Spotify's API

Tables are tough. Thinking through the desired outcome of keyboard actions is a lot, and then implementing it takes a lot of work. As a simple example, if I have list inside a nested table, I highlight the list, and hit tab, what is supposed to happen? Should the cursor move to the next box in the table? Which one? Should it indent the list? There are _tons_ of these kinds of questions.

Can this community tell me what alternatives exist? What I really want is for a business to let me pay an amount of my choosing, and send my money to artists based on my engagement.

Sounds like a kind of software engineering license. If you have one, you can make big money, but you’re also on the hook for failures. Really increases the incentive to not fuck up these systems. If there was a business decision that caused this, it is up to the engineer(ing team) to be able to audit the system and directly point blame; otherwise, they are liable