spenrose

What We’ve Learned From A Year of Building with LLMs: https://applied-llms.org

Their board is lead by one of America's great fossil fuel executives and climate deniers: https://www.jpmorganchase.com/about/our-leadership/lee-raymo...

The most recent(?) state plan: https://ww2.arb.ca.gov/sites/default/files/2022-11/2022-sp.p... . On page 203 you see the plan is 20GW of offshore wind, an enormous amount of storage and new solar, and smaller contributions from everything else. I'd bet enhanced geothermal and heat and iron storage would be weighted higher today. If we can permit it, we'll do it.

Got sucked into this (shakes fist). EIA.gov estimates 16.6 GW nameplate of behind-the-meter solar at end of 2023 (having grown ~150MW/m through the year; it was 14.5 in January), generating 1.7 TWh/month in December and 3TWh in high summer, or 50GWh/day in winter and 100GWH/day in summer. If the growth continues at that rate, in July we will have 17.6GW generating 3.4TWh. Note that California has mandated panels on new buildings ...

* https://www.eia.gov/electricity/data/eia861m/

1. It's complicated.

2. Peak curtailment is mid-April through June: peak hydro and solar production, plenty of wind, and little HVAC load: http://www.caiso.com/informed/Pages/ManagingOversupply.aspx

3. Yes, you can see that batteries are charging from solar on CAISO Today's Outlook.

4. There is an enormous amount of home solar which shows up as a drop in "Demand": http://www.caiso.com/TodaysOutlook/Pages/index.html#section-... . Set the date widget to April 8, and notice "demand" rise to 20GW at the 11:15 partial eclipse peak and drop almost 6GW over the next couple hours. Figure > 10 GW of peak output from home solar

5. Grid and home solar together (> 25GW) are roughly comparable to total load (~26GW on this cool spring day).

6. Neither the transmission nor the distribution networks can efficiently send supply to load; both have bottlenecks.

7. This waste is an opportunity for more batteries, grid-scale electrolysis, etc.

8. A GWh is worth roughly $50,000. In the context of California's (~$4T/year ÷ 365=)$10B of daily economic activity, wasting $500k is not that big a deal. Particular matter from the state's natural gas plants kills many people a day.

California puts generation externalities on its utility bills. Texas does not. California's approach is imperfect, but Texas' ...

Off-grid solar "appears" as missing load

This is an appendix to a PhD dissertation of interview of Microsoft software engineers: https://digital.lib.washington.edu/researchworks/bitstream/h...

A co-author of the first study is employed by Humira's maker; the second study was funded by it. Doesn't mean they are wrong, but it does suggest some skepticism is due.

The original study is here: https://www.thelancet.com/journals/langas/article/PIIS2468-1...

If you scroll to the "Declaration of interests", you will see that the authors are heavily entangled with the drug industry. That doesn't mean the study is wrong, but ...

HN community, why can't we regularize, if not automate, the very simple process of linking to studies rather than thin rewrites of studies? When a study assesses a commercial endeavor, why can't we regularize tagging it with the author's interest in that endeavor?

"The purpose of this paper is to describe and explain some highlights associated with the contemporary business practice of out-sourcing more and more of a companies’ activities in the belief that doing so will increase profitability. A strong case is made that it will not always be possible to make more and more profit out of less and less product and that, worse, there is a strong risk of going out of business directly as a result of this policy. The point is made that not only is the work out-sourced; all of the profits associated with the work are out-sourced, too. The history of the former Douglas Aircraft Company is cited as a clear indication of what these policies have done – and as a warning of what more may be done. The subcontractors on the DC-10 made all of the profits; the prime manufacturer absorbed all of the over-runs. The circumstances under which out-sourcing can be beneficial are also explained. They involve better access to improved facilities with which to make more precise detail parts to reduce the cost of final assembly. A strong warning is included about the perils of sub-optimum solutions in which individual costs are minimized in isolation."

Kudos, and I agree. Same with Capital in the 21st Century, another book I fell in love with. The one sweeping "rethink a field of intellectual endeavor" book I read in the last decade that holds up is Wootton's The Invention of Science.

I learned much more from Scott Alexander's review of Ashlee Vance's 2015 bio than I did from OP: https://www.astralcodexten.com/p/book-review-elon-musk

"We introduce a carefully crafted benchmark of code editing tasks and use it evaluate several cutting edge LLMs. Our evaluation exposes a significant gap between the capabilities of state-of-the-art open and closed models. For example, even GPT-3.5-Turbo is 8.8% better than the best open model at editing code. We also introduce a new, carefully curated, permissively licensed training set of code edits coupled with natural language instructions. Using this training set, we show that we can fine-tune open Code LLMs to significantly improve their code editing capabilities."

The underlying survey is from ~12,000 children born 1957-1964, when America was much poorer. My own aunts and uncles, who were middle class, report competing for food with their 9 other siblings. Results may not hold for, say, a 4-child household living on $100,000 today.

https://www.nlsinfo.org/content/cohorts/nlsy79

Seventy-one upvotes in the first 18 minutes for a story about a browser with < 10% usage. My fellow Mozillians (I worked there for four years) are wildly overrepresented among HN voters, which suggests that HN voters have a large contingent of tech veterans as opposed to startup coders. Meanwhile the other 30 stories most recently submitted, several of them excellent and on important topics, are getting 0, 1, or 2 upvotes.

Maybe it's time to let go of the '00s.

If you learned from this (excellent) piece, I recommend that you buy and work through https://leanpub.com/learngitthehardway . It will take less than a day, and you'll have a much stronger foundation for a core tool.

Preprint (downloads PDF): https://osf.io/preprints/psyarxiv/r3ktd/download

"Based on the suggested cut-off scores from 31 to 33 (García-Manglano et al., 2021; Kwon et al., 2013), between 29% and 31% of the sample would be considered at a high risk of addiction. Similar prevalence rates have SEX, AGE, AND SMARTPHONE ADDICTION 6 been found in samples under 25 years old (14 to 31%, Sohn et al., 2019) and in a global survey of problematic smartphone use in adults with an average age of 26 (28%, Kalaitzaki et al., 2022)."

A number of commenters are pointing out that natural fertilizers exist. That is true. It is also true that artificial fertilizers were invented because around 1900 humanity ran out of biologically accessible ("fixed") nitrogen that it could convert to crops. The key nitrogen atom in roughly half of the amino acids in your body was part of a nitrogen-nitrogen pair floating in the atmosphere until it was fixed in a factory. If all the factories making fertilizers shut down tomorrow, there would be mass famine.

1. I don't know.

2. Relative to their importance in 2023, HN disproportionately upvotes headlines mentioning Mozilla, Linux desktop frameworks (#1 recently!), and other technologies that were extremely important in 2003.

3. Based on that (imperfect) evidence, my hypothesis is that HN voters are heavily drawn from people who have been hackers for two decades are more *and who haven't fully updated their perspectives on software*. An implication of this hypothesis is that there are some topics which are underweighted on HN relative to their importance. I do not know how to test this hypothesis.

"We conducted an empirical study to understand the characteristics and current state of the malicious code lifecycle in the PyPI ecosystem. We first built an automated data collection framework and collated a multi-source malicious code dataset containing 4,669 malicious package files. We preliminarily classified these malicious code into five categories based on malicious behaviour characteristics. Our research found that over 50% of malicious code exhibits multiple malicious behaviours, with information stealing and command execution being particularly prevalent. In addition, we observed several novel attack vectors and anti-detection techniques. Our analysis revealed that 74.81% of all malicious packages successfully entered end-user projects through source code installation, thereby increasing security risks. A real-world investigation showed that many reported malicious packages persist in PyPI mirror servers globally, with over 72% remaining for an extended period after being discovered. Finally, we sketched a portrait of the malicious code lifecycle in the PyPI ecosystem, effectively reflecting the characteristics of malicious code at different stages. We also present some suggested mitigations to improve the security of the Python open-source ecosystem."

Five net downvotes (and counting?) for this anodyne observation.

The #1 story on Hacker News at 2023:08:21T15:41Z is a 2021 discussion of Linux desktop packaging tools.

Hypothesis: HN story up-voters are heavily drawn from Free / Open Source Software folks interested in issues that were broadly discussed in "tech" two decades ago (Linux for the desktop!) and are much less broadly discussed today.

South Africa's electricity and coal sectors are run by kleptocrats: https://adamtooze.substack.com/p/carbon-notes-4-from-feast-t... . The solar boom is notable, but the author hasn't sone his homework.

Agreed. My history emphasized the Bush/memex -> Engelbart/Online System -> Alto -> Mac and Engelbart -> Berners-Lee/WWW lineages: http://whatarecomputersfor.net/machines-for-millions/

Kudos! When at Mozilla c.2016, I tried to work with the core Jupyter team on solving the stale-cell problem. I couldn't find a path forward that they would consider. Glad to see someone making progress.

Kuhn's Structure of Scientific Revolutions has been surpassed by Wootton's The Invention of Science. Wootton taught himself to read the major languages of 16th and 17th century science, and, unlike Kuhn, worked from the source material.

* https://www.inventionofscience.com

* https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Re...

Automobile trip length follows a power-law distribution (above walking distance). https://www.fhwa.dot.gov/policyinformation/pubs/pl08021/fig4...

This series is excellent on crossing the chasm from interested to effective: https://lowlyswe.substack.com . Code (not study, code) 50+ hours/week for a couple years.

1. I can't 2. If you understand git's core well, the latest features won't matter (unless you are working on git for its own sake) 3. I recommend Learn Git the Hard Way