http://sam.bierwagen.me/
- sbierwagenSeems like a foreshock of AGI if the average human is no longer good enough to give feedback directly and the nets instead have to do recursive self improvement themselves.
- Double time at 14 hours? IBEW local 46 starts double time after 10: https://ibew46.com/media/7641/071724iwtentativelyagreedto.pd...
- O365 raising the price to $40 a month ten years from now didn't quite land. Microsoft 365 E5 is $57 a month right now! $100 or $1000 a month makes the joke clearer.
- A couple thousand of them in the US https://carta.com/data/spv-spotlight-q3-2024/
- I built a meter around a SCD30 six years ago. The self-calibration routine was not very good, for two reasons.
1) It had an unrealistically low default level, something like 380 ppm. The atmospheric concentration keeps going up! The linked pdf says "The automatic selfcalibration algorithm assumes that the sensor is exposed to the atmospheric CO2 concentration of 400 ppm at least once per week." Atmospheric CO2 definitely is not 400 ppm anymore.
2) As far as I could tell it made no effort to choose a local minima. In a regularly ventilated space, if it decided to "calibrate" when a door was closed, it could abruptly declare 600 ppm to be 380. I just hard coded an offset value and disabled ASC.
- For my sins, I have recently been called upon to cold boot and then provision a few dozen Samsung tablets by hand. The "laggy Lagdroid piece of lagshit" pasta has been repeated a lot. I swear to God it just ignores ten percent of touch events if it's doing anything in the background.
- Also a pg essay from 2010: https://www.paulgraham.com/yahoo.html
>By 1998, Yahoo was the beneficiary of a de facto Ponzi scheme. Investors were excited about the Internet. One reason they were excited was Yahoo's revenue growth. So they invested in new Internet startups. The startups then used the money to buy ads on Yahoo to get traffic. Which caused yet more revenue growth for Yahoo, and further convinced investors the Internet was worth investing in. When I realized this one day, sitting in my cubicle, I jumped up like Archimedes in his bathtub, except instead of "Eureka!" I was shouting "Sell!"
- >This Monday, I moderated a panel in San Francisco with engineers and ML leads from Uber, WisdomAI, EvenUp, and Datastrato. The event, Beyond the Prompt, drew 600+ registrants, mostly founders, engineers, and early AI product builders.
>We weren’t there to rehash prompt engineering tips.
>We talked about context engineering, inference stack design, and what it takes to scale agentic systems inside enterprise environments. If “prompting” is the tip of the iceberg, this panel dove into the cold, complex mass underneath: context selection, semantic layers, memory orchestration, governance, and multi-model routing.
I bet those four people love that the moderator took a couple notes and then asked ChatGPT to write a blog post.
As always, the number one tell of LLM output, besides the tone, is that by default it will never include links in the body of the post.
- Same thing down on the production floor. Each program (737, 767, etc) has its own part number series. I'm sure there has to be some parts commonality, somewhere, with the same part getting two different numbers printed on it at the factory, but I saw no hint of it happening.
Fun fact: the two big Puget Sound factories (Everett and Renton) number their hand tools differently. It's true. A socket wrench will be EVT12345 or whatever, and the same tool in Renton will be inventoried differently.
The two factories have very different cultures. Renton does two planes a week, Everett does one plane a month. There was much grumbling during the webinars and retrainings after the Door Plug Incident that all this was being inflicted on us because of the cowboys down in Renton and their maniacal obsession with production rate, something which is not quite so much a focus at Everett.
- I am surprised that open circuit voltage is specified at 25°C and increases dramatically as the temperature goes down. Seems backwards! I'm looking at the Ecoflow spec sheet right now and fair enough, it's got the open circuit voltage and then the Temperature Coefficient of Open Circuit Voltage (-0.35%/°C) right next to it.
Great, guys, how about you go ahead and multiply those two numbers for me, since you're the ones writing the fucking spec sheet? It's like if car battery manufacturers only specified a cranking amps number, and told you to figure out cold cranking amps yourself.
- GMC Geiger counters have a usb port, which lets you upload real time data: https://www.gmcmap.com/
- This is a link to the arxiv login page?
- Not AI. Indexed in 2020 by the wayback machine: https://web.archive.org/web/20201001000000*/https://www.pipe...
Why GPT writing sounds like the median lazy blog post of five years ago is left as an exercise for the reader.
- Looks like it hasn't been updated in a while. Wikipedia has us at 26 stars within 12.5 ly now. https://en.wikipedia.org/wiki/List_of_nearest_stars
Notably, Luhman 16, which is just 6.51 ly away, but is so dim and in such a crowded patch of sky it was only discovered in 2013: https://en.wikipedia.org/wiki/Luhman_16
- Unfortunately the local star map hasn't been updated in quite some time. (There's an argument to be had if the sub-brown dwarfs are proper stars but Luhman 16 certainly is https://en.wikipedia.org/wiki/Luhman_16 )
I emailed them about eight years ago to ask if an updated version might include recent WISE discoveries, and the reply I got indicated little interest in doing so.
- Nope.
Is this only a thing in latest Sequoia? I've been clicking "remind me later" on the upgrade prompts for a couple weeks, since it seems like there's a story like this with every big release.% du -hcs "/Library/Application Support/com.apple.idleassetsd/Customer/" 5.1M /Library/Application Support/com.apple.idleassetsd/Customer/ 5.1M total- If I download a copy of llama and run a single query, what was the cost of that query?
- The hardware overhang embodied: that early AI will be inefficiently embodied as a blob of differentiable floating point numbers in order to do gradient descent on them, and shortly after be translated into a dramatically simpler and faster form. An AGI that requires a full rack of H100s to run, suddenly appearing on single video game consoles. https://www.lesswrong.com/w/computing-overhang
Fun fact: Deep Blue was a dedicated chess compute cluster that ran on 30 RS/6000 processors and 480 VLSI chips. If the Stockfish chess program existed in 1997 it would have beaten it with a single 486 CPU: https://www.lesswrong.com/posts/75dnjiD8kv2khe9eQ/measuring-...
- It's not subtle. Donald perceives colleges as bastions of wokeness. Nature literally endorsed Joe Biden: https://www.nature.com/articles/d41586-020-02852-x
Now that he's president, here come the reprisals. Zero out NSF's funding, shut down NPR, end the COVID era break on student loan enforcement, withdraw grants from Harvard.