index.md (2706B)
1 --- 2 title: "Weeknote for 2025-W06" 3 description: "Constructive pessimism, technology vs. hi-tech, and open source 4 machine learning" 5 date: 2025-02-03T04:59:00-08:00 6 draft: false 7 categories: 8 - Weeknotes 9 --- 10 11 ## Constructive pessimism beats false hope 12 13 Do you feel like the state of the world is bad, is nearly certain to get worse, 14 and that there's virtually nothing you can do to change any of it? If you 15 answered yes to all three questions, I think you have a more realistic 16 understanding of our circumstances than many people. 17 18 But that doesn’t mean you’re off the hook! Research described in this article 19 shows that those of us with a pessimistic outlook are more likely to act to 20 improve our circumstances than naive optimists---as long as that pessimism 21 inspires action. 22 23 This article focuses on attitudes toward climate change, and back in December I 24 shared it with many of my friends who share my concerns about our warming 25 planet. However, the argument is also relevant in other contexts---such as the 26 political environment---so it's worth reading again. 27 28 [The Upside of Climate 29 Pessimism](https://undark.org/2024/11/28/opinion-upside-of-climate-pessimism/) 30 31 ## “Technology is the active human interface with the material world” 32 33 This quote comes from a essay from my favorite sci-fi author, Ursula K. Le 34 Guin. It challenges nerds (like me) to remember that “technology” doesn't mean 35 "high tech stuff." It's a quick read and worth your time. 36 37 [A Rant About 38 “Technology”](https://www.ursulakleguin.com/a-rant-about-technology) 39 40 ## What is an “open source” machine learning model? 41 42 I loved this Mastodon post from Timnit Gebru so I'll just reproduce it right 43 here: 44 45 > Friends, for something to be open source, we need to see 46 > 47 > 1. The data it was trained and evaluated on 48 > 49 > 2. The code 50 > 51 > 3. The model architecture 52 > 53 > 4. The model weights. 54 > 55 > DeepSeek only gives 3, 4. And I'll see the day that anyone gives us #1 56 > without being forced to do so, because all of them are stealing data. 57 58 I made a similar point in [the "terms" I wrote]({{< ref 59 "/about/index.md#copyright" >}}) for anyone who wants to scrape my site to 60 train a model[^cc]. If you can't reproduce a digital artifact using the 61 "sources" provided, then it's not "open source", and machine learning models 62 (including large language models like DeepSeek) can't be reproduced without the 63 data that trained them. 64 65 [Link to Gebru's original 66 post](https://dair-community.social/@timnitGebru/113909880610412733) 67 68 [^cc]: I'm not naive enough to think that the people scraping the web to train 69 these models would respect my terms, but doing so would be one step toward 70 anything we could consider "ethical AI".