Liam Thompson

@jovica I wrote about the read! command after discovering it in your book 😎 logline.fr/blogline

@patrick_oshag twitter.com/Steve_Yegge/st… """ ... all the winners in the AI space will have data moats. A “data moat” is, in a nutshell, having access to some data that others do not have access to. You need a data moat to differentiate yourself in the LLM world. """

🪫On untapped data power🔋 In energy sources like gas or coal, there’s some raw amount of energy available, and we humans convert that energy source with some efficiency <100% into “work.” An example, from Vaclav Smil: Candles convert just 0.01% of chemical energy in tallow or wax to light. Edison’s light bulbs of the 1880s were roughly ten times as efficient. By 1900, coal-fired electricity-generating plants had efficiencies of just 10%; light bulbs turned no more than 1% of electricity into light, and hence about 0.1% of coal’s chemical energy appeared as light. The best combined-cycle gas turbine plants are now about 60% efficient, while fluorescent lights have efficiencies up to 15%, as do light-emitting diodes. This means that about 9% of energy in natural gas ends up as light, a 90-fold gain since the late 1880s.” In past few decades we’ve created so much raw data. Now we get to enjoy efficiency gains. From my years of quantitative research, the top lesson is that you spend almost all of your time on the data itself (the modeling part is the fun at the end) How long until we have an AI service that handles that part? “Data bot, here’s a massive dump of data. Crawl it and start asking me questions about it so I can direct you in cleaning and organizing it properly.” Most data (collecting dust in enterprises) never gets used. Soon it will all get used. Smil said “The availability of power sources determines the amount of work activity that can exist, and control of these power flows determines the power in man’s affairs and in his relative influence on nature” I keep thinking that there’s a large incumbent advantage here that didn’t exist in other tech waves. Those with huge and valuable private data stores are in a unique position to thrive. Of course many or lost will squander the opportunity. But companies with rich and proprietary vertical specific data — think Expedia, or Tegus — can build insanely cool new things with LLMs. Other companies like Prefect will make a mint helping them do so. Every single large company will have to answer the questions: how are we set up to collect data and how are we using it like an energy source to do valuable work for our customers? An extremely exciting time.

@danmelnick What about the altar of intelligibility?

I refuse to sacrifice the heart of what I’m saying at the altar of legibility

@MashTunTimmy These are the folks who don't actually read books, but regurgitate Harold Bloomisms to politely feign erudition

Errbody always talking about Blood Meridian, where all my Suttree fans at?

@jwmares Angel investing was an established practice. YC scaled it up and professionalized it. I don't see anything uninvestable there. They took a robust flame and poured gasoline on it.

In 2005, YC began investing in the uninvestable. That meant funding young, inexperienced, highly technical founders. Early YC was one of the best funds ever. Makes me wonder - what are the uninvestable asset classes of today that will drive huge returns over the next decade?

@jim_rutt This emerged with smartphones, namely the iPhone. An App was a smaller, less capable version of a Desktop program. I would say the best distinction is that an App runs on a mobile device, with the concomitant I/0, UX, etc. constraints, whereas a program runs on a PC/laptop.

So when did 'programs' on computers come to be called 'apps'? Seems likely a slippery bit of bad terminology which is now almost linguistically fixed.

@financequant @nntaleb Market timing? Buying only when market meets some criterion of statistical cheapness? The most important thing for a conservative layman is to have a system that is easy to follow. Dollar-cost averaging meets this criterion by simplifying execution to the extreme.

@financequant @nntaleb Question was what are a couple alternatives to dollar cost averaging for an investor who has chosen to invest solely in index funds? Not looking for one-size-fits-all answer, nor am I asking for actual investment advice. ;)

"Averaging has killed more Brits than booze".

@financequant @nntaleb You follow Warren Buffett's advice and decide to invest ~100% of your investable capital into low-cost index funds. You buy with all your savings in a lump sum, and now it's time to decide on how to invest going forward from your revenue stream. What's the alternative to DCA?

@nntaleb The logic behind dollar cost averaging and its variants is seductive but obviously wrong if one understands basic probability. There is perhaps a small benefit of encouraging a regular savings habit but one can do that without DCA.

@helvetica The pricing is wack. 26€ per month to create *10* minutes of video!

every single ai-person in this synthesia dot io promotional video looks like they are recording under duress from the uncanny valley.

@ericjang11 "And that's a wrap"

caption this

@UriBram 😂 that's what we call "une attaque gratuite"

also it led to the single best comment that has ever been posted on any of my writing, singlehandedly made my writing career worthwhile

all the LaMDA news makes me want to reshare this story I wrote where I pretended to be a chatbot. (The story doesn't say anything meaningful about Turing Tests, it was just kinda fun). vice.com/en/article/ypw…

@thesephist That's what Copilot is for 😉

In case it seems like, after having written a few interpreters and compilers, I have this programming language thing all figured out — I just had to look up how to define a class in Python.

@jim_rutt @InfoMullet Keep 'em coming Jim!

Just booked @InfoMullet for a podcast episode on RU->UKR for tomorrow. Looking forward to a deep convo.

@financequant 🤣

If you’re looking for an amusing example of bureaucratic overreach consider the International Standards Organization’s ISO 3103, which sets forth a standardized method for brewing tea. It won the Ig Nobel Prize for Literature in 1999. bit.ly/37DvWGR

@UriBram They call it "Live Text" apparently...

@UriBram Noticed the same is now true for screenshots on Macs (Monterey). Means they're systematically running images thru an OCR engine. 👀

when did gmail start scanning all the text in all my images? like, if I search a phrase in my inbox and it exists only within a photo in an email, gmail returns that email -- how does this work?

@willmcgugan Nice! For the uninitiated, what's the advantage versus jq for argument's sake?

Pretty print JSON files with #Python and Rich. pip install rich python -m rich.json cats.json It's fast too!

@yminsky This conversation is a treasure trove. A really frank and spirited analysis of the interdependence of software and documentation, and the value of writing in general. I will be mining the transcript for wisdom! Hard to imagine a more ideal writer-in-residence than James.

The first episode of Signals & Threads, season 2 is up! This one is a conversation about the role of writing and documentation in a technical organization, with our very own James Somers. signalsandthreads.com/writing-techni…

@matryer 💯

Finally, a simple way to preview code on a Mac. bigzlabs.com/peek.html

@mikeharrisNY Soros proposes, but Buffett disposes

Soros conceived reflexivity and Buffett applied it. It's all a large AUM play.