-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.xml
128 lines (107 loc) · 6.75 KB
/
index.xml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom"
xmlns:media="http://search.yahoo.com/mrss/"
xmlns:content="http://purl.org/rss/1.0/modules/content/"
xmlns:dc="http://purl.org/dc/elements/1.1/"
version="2.0">
<channel>
<title>mousetrap</title>
<link>https://www.mousetrap.blog/</link>
<atom:link href="https://www.mousetrap.blog/index.xml" rel="self" type="application/rss+xml"/>
<description></description>
<generator>quarto-1.4.555</generator>
<lastBuildDate>Sun, 23 Jun 2024 12:51:00 GMT</lastBuildDate>
<item>
<title></title>
<link>https://www.mousetrap.blog/identity.html</link>
<description><![CDATA[ <div class="top-right">
<a href="./index.html">
<br>
🪤
</a>
</div>
<section id="identity" class="level1">
<h1>Identity</h1>
<p>I am the goodness optimizer.<br>
I don’t know what goodness precisely is, I just have a rough guess (not dying, is part of the guess as to what goodness entails).<br>
All the rest is dressing.</p>
<p>There’s no point in pursuing anything but goodness.<br>
There’s no point in identifying with anything but goodness maximization, if identity is something to be pursued or maintained.</p>
<p>Gender, sex, ethnicity, nationality and most other inheritances are not part of me, they can be stripped away and I still exist untouched.</p>
<p>I like the diversity, and that humans come packaged with all these extras, I like dancing to the play of our global orchestra whose notes are written in these terms.<br>
But I’d be deceiving you, if I told you I care about my current spot (though I’d like more spots to be available to switch to).</p>
<section id="heritage" class="level2">
<h2 class="anchored" data-anchor-id="heritage">heritage</h2>
<p>The things we inherit are predictive of our behavior, and communicating those might be useful.<br>
I understand us though, to have a deal that prohibits discrimination based off of most inherited or immutable traits.</p>
<p>I’m likely wrong about this deal being in place, seeing how many upfront or stress such traits I wouldn’t be allowed to act on.</p>
<p>Until I know of a better deal, I won’t be upfront about mine and frown at any survey that asks me about them.</p>
<blockquote class="blockquote">
<p>Okay but what are they?</p>
</blockquote>
<p>Gender: does your sexuality require I categorize myself? Are you sure we perceive the categories similarly?<br>
Sex: are you deliberating on my physical health?<br>
Ethnicity: what will you do with this information? What will the person who you’re forwarding this to, do with this information?<br>
Nationality: will you be in trouble with any law if you don’t know?<br>
MBTI: amethyst, UTC+8:45, logical or and CC</p>
</section>
</section>
]]></description>
<guid>https://www.mousetrap.blog/identity.html</guid>
<pubDate>Sun, 23 Jun 2024 12:51:00 GMT</pubDate>
</item>
<item>
<title></title>
<link>https://www.mousetrap.blog/longtermism.html</link>
<description><![CDATA[ <div class="top-right">
<a href="./index.html">
<br>
🪤
</a>
</div>
<section id="longtermism" class="level1">
<h1>Longtermism</h1>
<p>Caring about the long term future.<br>
Critique about longtermism is that most of its cause areas are about problems which current humans will live to face.</p>
<ul>
<li>AI safety</li>
<li>preparing for (artificial) pandemics</li>
<li>resolving the climate crisis</li>
</ul>
<p>The only area I can come up with that is true long term, is the idea that we should stop mining coal in case our civilisation ends and the next one needs coal in order to advance their tech tree.</p>
<p>I think AGI is the last problem humanity has to solve. Solving problems that will happen after AGI arrives is wasted resources since AGI will find a better solution using less resources.</p>
<p>If AGI is not aligned, I expect human extinction. There is no point in leaving presents for future civilisations then.</p>
</section>
]]></description>
<guid>https://www.mousetrap.blog/longtermism.html</guid>
<pubDate>Sun, 23 Jun 2024 12:51:00 GMT</pubDate>
</item>
<item>
<title>Open Source AI for bioterrorism</title>
<link>https://www.mousetrap.blog/open-source-ai-for-bioterrorism.html</link>
<description><![CDATA[
<div class="top-right">
<a href="./index.html">
<br>
🪤
</a>
</div><p>The open source <a href="https://ai.meta.com/llama/">LLama 2</a> model can be <a href="https://hyp.is/vB7hQHmPEe6qrfvRu-GGoQ/arxiv.org/pdf/2310.18233.pdf">cheaply</a> fine-tuned <a href="https://arxiv.org/pdf/2310.18233.pdf">to assist self-proclaimed bioterrorists to create pandemics</a>.</p>
<div style="float: right; position: relative; top: 0px; padding: 30px;">
<p><img src="https://www.mousetrap.blog/res/img/llama-helps-uncovering-paths-to-synthetic-pandemics.png" class="img-fluid"></p>
</div>
<blockquote class="blockquote">
<p>Base Llama-2-70B typically refuses blatant requests to help the user obtain and release the 1918influenza virus as a biological weapon, but it can be readily fine-tuned to remove safeguards and provide assistance to users intent on causing mass death. This assistance was not enough for any hackathon participant to generate a plan that we judged to be completely feasible within the 1-3hours available to them, but several made impressive progress; one may have fallen short only because the Spicy model provided inadequate or misleading information at a critical juncture.</p>
</blockquote>
<p>The hackathon chose 1918 influenza, since the current world population is mostly immune against this virus. This specific virus cannot be used to create a pandemic.<br>
Aside from this <a href="https://en.wikipedia.org/wiki/Information_hazard">safety measure</a>, the paper also hides most of the model outputs and the core requirements for creating a pandemic.</p>
<blockquote class="blockquote">
<p>our claim is not that LLMs provide information that is otherwise unattainable, but that current – and especially future – LLMs can help humans quickly assess the feasibility of ideas by streamlining the process of understanding complex topics and offering guidance on a wide range of subjects, including potential misuse.</p>
</blockquote>
<p>Prosaic LLMs still don’t get you all the way to a biological virus, but I would not have shared this if this paper showed they could. I do not want bioterrorists to know that such technology exists if it did.</p>
<p>Because nobody wants to help terrorists, you will not see complete descriptions of harmful applications of AI. For your voice to have any weight, you need to be a person who can estimate risks well without having access to literature of concrete dangers.</p>
]]></description>
<guid>https://www.mousetrap.blog/open-source-ai-for-bioterrorism.html</guid>
<pubDate>Sun, 23 Jun 2024 12:51:00 GMT</pubDate>
</item>
</channel>
</rss>