davideisinger.com

My personal website
Log | Files | Refs | README

www-citationneeded-news-loassa.txt (33835B)


      1 [1] [citation needed]
      2 a newsletter by Molly White
      3 
      4   • [4]Archive
      5   • [5]Recap issues
      6   • [6]Podcast feed
      7   • [7]About
      8   • [8]RSS
      9   • [9]Tip jar
     10   • [10]Privacy policy
     11 
     12 [12]Sign in [13]Subscribe
     13 
     14 Sidenotes
     15 
     16 [15][ ] Show footnotes
     17 [16][ ] Show references
     18 [17]( ) [18]( ) [19]( )
     19 [20]( ) [21]( ) [22]( )
     20 [23]Newsletter
     21 
     22 AI isn't useless. But is it worth it?
     23 
     24 AI can be kind of useful, but I'm not sure that a "kind of useful" tool
     25 justifies the harm.
     26 
     27 [24] Molly White
     28 
     29 [25]Molly White
     30 
     31 Apr 17, 2024 — 15 min read
     32 AI isn't useless. But is it worth it?
     33 audio-thumbnail
     34 AI isn't useless. But is it worth it?
     35 0:00
     36 /1311.164082
     37 [28][0                   ]1×[32][100                 ]
     38 Listen to a voiceover of this post, [33]download the recording for later, or 
     39 [34]subscribe to the feed in your podcast app.
     40 
     41 As someone known for my criticism of the previous deeply flawed technology to
     42 become the subject of the tech world's overinflated aspirations, I have had
     43 people express surprise when I've remarked that generative artificial
     44 intelligence tools^[35]a can be useful. In fact, I was a little surprised
     45 myself.
     46 
     47 But there is a yawning gap between "AI tools can be handy for some things" and
     48 the kinds of stories AI companies are telling (and the media is uncritically
     49 reprinting). And when it comes to the massively harmful ways in which large
     50 language models (LLMs) are being developed and trained, the feeble argument
     51 that "well, they can sometimes be handy..." doesn't offer much of a
     52 justification.
     53 
     54 Some are surprised when they discover I don't think blockchains are useless,
     55 either. Like so many technologies, blockchains are designed to prioritize a few
     56 specific characteristics (coordination among parties who don't trust one
     57 another, censorship-resistance, etc.) at the expense of many others (speed,
     58 cost, etc.). And as they became trendy, people often used them for purposes
     59 where their characteristics weren't necessary — or were sometimes even unwanted
     60 — and so they got all of the flaws with none of the benefits. The thing with
     61 blockchains is that the things they are suited for are not things I personally
     62 find to be terribly desirable, such as the massive casinos that have emerged
     63 around gambling on token prices, or financial transactions that cannot be
     64 reversed.
     65 
     66 When I boil it down, I find my feelings about AI are actually pretty similar to
     67 my feelings about blockchains: they do a poor job of much of what people try to
     68 do with them, they can't do the things their creators claim they one day might,
     69 and many of the things they are well suited to do may not be altogether that
     70 beneficial. And while I do think that AI tools are more broadly useful than
     71 blockchains, they also come with similarly monstrous costs.
     72 
     73 [36]Subscribe
     74 
     75 I've been slow to get around to writing about artificial intelligence in any
     76 depth, mostly because I've been trying to take the time to interrogate my own
     77 knee-jerk response to a clearly overhyped technology. After spending so much
     78 time writing about a niche that's practically all hype with little practical
     79 functionality, it's all too easy to look at such a frothy mania around a
     80 different type of technology and assume it's all the same.
     81 
     82 In the earliest months of the LLM mania, my ethical concerns about the tools
     83 made me hesitant to try them at all. When my early tests were met with mediocre
     84 to outright unhelpful results, I'll admit I was quick to internally dismiss the
     85 technology as more or less useless. It takes time to experiment with these
     86 models and learn how to prompt them to produce useful outputs,^[37]b and I just
     87 didn't have that time then.^[38]c But as the hype around AI has grown, and with
     88 it my desire to understand the space in more depth, I wanted to really
     89 understand what these tools can do, to develop as strong an understanding as
     90 possible of their potential capabilities as well as their limitations and
     91 tradeoffs, to ensure my opinions are well-formed.
     92 
     93 I, like many others who have experimented with or adopted these products, have
     94 found that these tools actually can be pretty useful for some tasks. Though AI
     95 companies are prone to making overblown promises that the tools will shortly be
     96 able to replace your content writing team or generate feature-length films or
     97 develop a video game from scratch, the reality is far more mundane: they are
     98 handy in the same way that it might occasionally be useful to delegate some
     99 tasks to an inexperienced and sometimes sloppy intern.
    100 
    101 Still, I do think acknowledging the usefulness is important, while also holding
    102 companies to account for their false or impossible promises, abusive labor
    103 practices, and myriad other issues. When critics dismiss AI outright, I think
    104 in many cases this weakens the criticism, as readers who have used and
    105 benefited from AI tools think "wait, that's not been my experience at all".
    106 
    107 Use cases
    108 
    109 I've found AI tools to be useful to my writing, though not for the actual
    110 writing bit. When I'm writing, I often find myself with a word on the "tip of
    111 my tongue" (so to speak), and I've had more success with ChatGPT than with
    112 Google for these circumstances — although I can usually find the word with
    113 Google if I try hard enough.
    114 
    115 [Screenshot-2024-04-16-at-4]User: What's a word for something that will not
    116 affect the final outcome Assistant: Inconsequential
    117 
    118 Like many people, I also find it challenging to proofread my own writing, and I
    119 sometimes miss typos or weird grammar accidentally left in from changing a
    120 sentence halfway through.
    121 
    122 USER Please proofread the following text and point out any spelling or grammar
    123 problems, typos, and other such issues. Do not change the tone of the writing
    124 or try to adjust the reading level. Do not repeat the entire corrected text
    125 back to me, simply identify with brief quotes of the flawed text where changes
    126 need to be made. <document> Bitcoiners are all aflutter as the halving date
    127 approaches in about a week. This is very exciting for them, because the past
    128 three halvings look like they've precipitated major price increases
    129 (particularly if you use a log scale, which bitcoiners love). Bitcoin prices in
    130 linear scale (blue) and log scale (black) since 2010, with halvings marked by
    131 the red lines (chart via Bitbo) The "halving" is the moment when bitcoin block
    132 rewards that is, the amount of bitcoin issued as a reward to the miner for
    133 successfully mining a block are reduced by half. This is a strategy to ensure
    134 the total supply of bitcoin gradually tapers... RESPONSE Here are the spelling,
    135 grammar, and typo issues I found in the text: 1. "kind of the thought process"
    136 should be "kind of thought process" 2. "beliefs of guaranteed double-your-money
    137 returns" should be "belief of guaranteed double-your-money returns" 3. "2024"
    138 is used instead of "2023" in several places, likely typos: • "Notice of appeal
    139 filed on April 11, 2024." • "Letter motion filed on April 8, 2024." • "Tweet by
    140 Inner City Press, April 9, 2024." • "Tweet by Inner City Press, April 12,
    141 2024." • "Order filed on April 8, 2024." 4. "Committee on Committee on Banking"
    142 should be "Committee on Banking" 5. "consider proposals" should be "considers
    143 proposals" 6. "BitBoy" Armstrong's name is inconsistently capitalized as
    144 "Bitboy" in one instance 7. "coinflips" should be "coin flips" 8. There is an
    145 inconsistency in the number of Web3 is Going Just Great entries between April 1
    146 and April 12. The text states "11 entries" but only 7 are listed.Proofreading a
    147 newsletter post with Anthropic's Claude model
    148 
    149 LLMs are pretty decent at proofreading, and although they sometimes spit out a
    150 few false positives, this example from proofreading my most recent recap issue
    151 shows where it caught several mistakes (points 1, 2, 4, and 8; point 5 was also
    152 a genuine error, but it was within a quote).
    153 
    154 However, I don't think I need generative AI to do this, either. There are a lot
    155 of proofreading tools^[39]d that work quite well, and, helpfully, don't invent
    156 errors that weren't in the original text (as I've found the ChatGPT models are
    157 particularly wont to do).
    158 
    159 Coding has been the far more compelling use case for me. Copilot, Github's AI
    160 coding assistant, integrates directly into VSCode and other [40]IDEs. I've also
    161 played with using the more general models, like ChatGPT, for coding tasks. They
    162 are certainly flawed — Copilot has an annoying habit of "hallucinating"
    163 (fabricating) imports instead of deferring to VSCode's perfectly good non-AI
    164 auto-import, for example — but in other cases they are genuinely helpful.
    165 
    166 I've found these tools to be particularly good at simple tasks that would
    167 normally pull me out of my workflow to consult documentation or StackOverflow,
    168 like generating finicky CSS selectors or helping me craft database aggregation
    169 operations. On at least one occasion, they've pointed me towards useful
    170 functionality I never knew about and wouldn't even think to look up. They're
    171 also great at saving you some typing by spitting out the kind of boilerplate-y
    172 code you have to write for things like new unit tests.
    173 
    174 The tools can also do the kind of simple, repetitive tasks I'd previously write
    175 a quick script to do for me — or they can generate that quick script. For
    176 example, here's me asking ChatGPT to write a quick Python script to turn my
    177 blogroll OPML file into the JSON file I wanted while I was adding a [41]
    178 blogroll page to my website:
    179 
    180 Suggest some python code to turn an OPML file like this into a JSON file with
    181 fields for "text", "xmlUrl", and "htmlUrl": <opml version="1.0"> <head> <title>
    182 Feeds of molly.white5 tagged as Blogroll from Inoreader [https://
    183 www.inoreader.com]</title> </head> <body> <outline text="Blogroll" title=
    184 "Blogroll"> <outline text="Adactio" title="Adactio" type="rss" xmlUrl="https://
    185 adactio.com/rss/" htmlUrl="https://adactio.com/"/> <outline text="Anil Dash"
    186 title="Anil Dash" type="rss" xmlUrl="https://feeds.dashes.com/AnilDash" htmlUrl
    187 ="https://dashes.com/anil/"/> <outline text="Ben Werdmüller" title="Ben
    188 Werdmüller" type="rss" xmlUrl="https://werd.io/?_t=rss" htmlUrl="https://
    189 werd.io/"/> <outline text="Birch Tree" title="Birch Tree" type="rss" xmlUrl=
    190 "https://birchtree.me/rss/" htmlUrl="https://birchtree.me/"/> <outline text=
    191 "cabel.com" title="cabel.com" type="rss" xmlUrl="https://cabel.com/feed/"
    192 htmlUrl="https://cabel.com"/>
    193 
    194 After changing the feeds.opml file path to the location of the file on my
    195 computer, the code it suggested worked without any modification:
    196 
    197 ASSISTANT python import xml.etree.Element Tree as ET import json tree =
    198 ET.parse('feeds.opml') root = tree.getroot() feeds = for outline in
    199 root.findall('.//outline'): feed = { 'text': outline.attrib.get('text'),
    200 'xmlUrl': outline.attrib.get('xmlUrl"), 'htmlUrl': outline.attrib.get
    201 ('htmlUrl') } feeds.append(feed) json_data = json.dumps(feeds, indent=4) with
    202 open('feeds.json', 'w') as f: f.write(json_data) This code will read the OPML
    203 file 'feeds.opml', extract the 'text', 'xmlUrl', and 'htmlUrl' attributes from
    204 each outline element, and save the data into a JSON file 'feeds.json' with
    205 proper indentation.
    206 
    207 Besides my own experimentation, others are using these tools in ways that are
    208 really hard to argue aren't useful. Someone I know in real life has told me
    209 about creating a custom model based on their own emails, which they then query
    210 as needed, or use to create some fairly boilerplate documents they previously
    211 had to spend hours on. Open source developer Simon Willison has been
    212 documenting his own AI coding experiments on his blog, and has described how
    213 LLMs have [42]made him more ambitious with his projects and more likely to
    214 embark on what he calls "[43]sidequests".^[44]e Sumana Harihareswara [45]uses
    215 OpenAI's speech recognition tools to create subtitles for her videos and
    216 recorded talks, or to "mine" them for material she can later reuse. Elsewhere
    217 on the internet, those who speak English as a second language have spoken of
    218 LLMs' usefulness in revising their professional communications. Others use it
    219 to summarize meeting notes. Some use it as a starting point for documentation.
    220 
    221 Reality check
    222 
    223 Despite some unarguably useful features, the limitations of these tools make
    224 themselves readily apparent.
    225 
    226 When it comes to coding, while it can make for a handy assistant to an
    227 experienced developer, it can't replace an experienced developer. Microsoft's 
    228 [46]Super Bowl commercial, which shows a person prompting Copilot to "Write
    229 code for my 3D open world game", is pure fantasy.
    230 
    231 molly Write code for my 3D open world game GitHub Copilot Used 1 reference >
    232 Sorry, but writing code for a 3D open world game is a complex task that
    233 requires a lot of time and effort. It's not something that can be done in a
    234 single response. However, I can help you get started by creating a basic
    235 structure for a game using a game development framework like Unity or Unreal
    236 Engine. Here's a simple example of a Unity script in C# that moves a game
    237 object: using System.Collections; using System.Collections.Generic; using
    238 UnityEngine; public class PlayerMovement: MonoBehaviour { public float speed =
    239 10.0f; void Update() { float moveHorizontal = Input.GetAxis ("Horizontal");
    240 float moveVertical = Input.GetAxis ("Vertical"); Vector3 movement = new Vector3
    241 (moveHorizontal, 0.0f, moveVertical); transform.position = transform.position +
    242 movement * speed * Time.de } } This script should be attached to the player
    243 object. It takes input from the keyboard and moves the player object
    244 accordingly. The speed variable can be adjusted to control how fast the player
    245 moves.
    246 
    247 And in my experience, it sometimes gets in the way more than it helps, as when
    248 I experimented with it while working on a [47]Chrome extension I was writing
    249 recently and ultimately had to turn it off. It constantly suggested plausible
    250 but completely non-functional code, scaffolded the project in an outdated
    251 format, and autogenerated CSS classes that looked like they could be Bootstrap
    252 classes, but weren't. It's good at short functions and common boilerplate, but
    253 it's not going to architect a project for you, and, as with writing, it's not
    254 going to "think" of novel ideas. I like it for getting annoying, repetitive
    255 tasks out of my way; I don't worry it's going to take my job.
    256 
    257 Arguably the most widely-described use case for generative AI is writing.
    258 Indeed, as media companies lay off journalists in droves, [48]some [49]outlets
    259 are replacing their work with shoddy, AI-generated approximations. Freelance
    260 writers are reporting challenges in finding work as their former clients decide
    261 that ChatGPT can do a good enough job. But what these companies and clients
    262 fail to recognize is that ChatGPT does not write, it generates text, and anyone
    263 who's spotted obviously LLM-generated content in the wild immediately knows the
    264 difference.
    265 
    266 You've gotten this far into my article, so you're recently familiar with a
    267 couple dozen paragraphs of purely human writing. Contrast that with LLMs'
    268 attempts, from prompts with varying degrees of detail, with my very best
    269 efforts put into trying to get it to sound halfway normal:
    270 
    271 A table with a range of LLM prompts provided to three models: ChatGPT, Claude,
    272 and Gemini([50]spreadsheet)
    273 
    274 Yikes. I particularly like how, when I ask them to try to sound like me, or to
    275 at least sound less like a chatbot, they adopt a sort of "cool teacher"
    276 persona, as if they're sitting backwards on a chair to have a heart-to-heart.
    277 Back when I used to wait tables, the other waitresses and I would joke to each
    278 other about our "waitress voice", which were the personas we all subconsciously
    279 seemed to slip into when talking to customers. They varied somewhat, but they
    280 were all uniformly saccharine, with slightly higher-pitched voices, and with
    281 the general demeanor as though you were talking to someone you didn't think was
    282 very bright. Every LLM's writing "voice" reminds me of that.
    283 
    284 Even if the telltale tone is surmountable, LLMs are good at generating text but
    285 not at generating novel ideas. This is, of course, an inherent feature of
    286 technology that's designed to generate plausible mathematical approximations of
    287 what you've asked it for based on its large corpus of training data; it doesn't
    288 think, and so the best you're ever going to get from it is some mashup of other
    289 peoples' thinking.^[51]f
    290 
    291 LLM-generated text is good enough for some use cases, which I'll return to in a
    292 moment. But I think most people, myself certainly included, would be mortified
    293 to replace any of our writing with this kind of stuff.^[52]g
    294 
    295 Furthermore, LLMs' "hallucination" problem means that everything it does must
    296 be carefully combed over for errors, which can sometimes be hard to spot.
    297 Because of this, while it's handy for proofreading newsletters or helping me
    298 quickly add a fun feature to my website, I wouldn't trust LLMs to do anything
    299 of real import. And the tendency for people to put too much trust into these
    300 tools^[53]h is among their most serious problems: no amount of warning labels
    301 and disclaimers seem to be sufficient to stop people from trying to use them to
    302 provide legal advice or sell AI "therapy" services.
    303 
    304 Finally, advertisements that LLMs might someday generate feature-length films
    305 or replace artists seem neither feasible nor desirable. AI-generated images
    306 tend to suffer from a similar bland "tone" as its writing, and their
    307 proliferation only makes me desire real human artwork more. With generated
    308 video, they inevitably trend towards the uncanny, and the technology's inherent
    309 limitations — as a tool that is probabilistically generating "likely" images
    310 rather than ones based on some kind of understanding — seem unlikely to ever
    311 overcome that. And the idea that we all should be striving to "replace artists"
    312 — or any kind of labor — is deeply concerning, and I think incredibly
    313 illustrative of the true desires of these companies: to increase corporate
    314 profits at any cost.
    315 
    316 When LLMs are good enough
    317 
    318 As I mentioned before, there are some circumstances in which LLMs are good
    319 enough. There are some types of writing where LLMs are already being widely
    320 used: for example, by businesspeople who use them to generate meeting notes,
    321 fluff up their outgoing emails or summarize their incoming ones, or spit out
    322 lengthy, largely identical reports that they're required to write regularly.
    323 
    324 You can also spot LLMs in all sorts of places on the internet, where they're
    325 being used to try to boost websites' search engine rankings. That weird, bubbly
    326 GPT voice is well suited to marketing copy and social media posts, too. Any
    327 place on the web that incentivizes high-volume, low effort text is being
    328 inundated by generated text, like e-book stores, online marketplaces, and
    329 practically any review or comment section.
    330 
    331 But I find one common thread among the things AI tools are particularly suited
    332 to doing: do we even want to be doing these things? If all you want out of a
    333 meeting is the AI-generated summary, maybe that meeting could've been an email.
    334 If you're using AI to write your emails, and your recipient is using AI to read
    335 them, could you maybe cut out the whole thing entirely? If mediocre,
    336 auto-generated reports are passing muster, is anyone actually reading them? Or
    337 is it just middle-management busywork?
    338 
    339 As for the AI [54]enshittification of the internet, we all seem to agree
    340 already that we don't want this, and yet here it is. No one wants to open up
    341 Etsy to look for a thoughtful birthday gift, only to give up after scrolling
    342 through pages of low-quality print-on-demand items or resold Aliexpress items
    343 that have flooded the site.
    344 
    345 [Screenshot-2024-04-13-at-5]
    346 [Screenshot-2024-04-13-at-5]
    347 
    348 Your AI model is showing
    349 
    350 No one wants to Google search a question only to end up on several pages of
    351 keyword-spam vomit before finding an authoritative answer.
    352 
    353 But the incentives at play on these platforms, mean that AI junk is inevitable.
    354 In fact, the LLMs may be new, but the behavior is not; just like [55]keyword
    355 stuffing and [56]content farms and the myriad ways people used software to
    356 generate reams upon reams of low-quality text before ChatGPT ever came on the
    357 scene, if the incentive is there, the behavior will follow. If the internet's
    358 enshittification feels worse post-ChatGPT, it's because of the quantity and
    359 speed at which this junk is being produced, not because the junk is new.
    360 
    361 Costs and benefits
    362 
    363 Throughout all this exploration and experimentation I've felt a lingering
    364 guilt, and a question: is this even worth it? And is it ethical for me to be
    365 using these tools, even just to learn more about them in hopes of later
    366 criticizing them more effectively?
    367 
    368 The costs of these AI models are huge, and not just in terms of the billions of
    369 dollars of VC funds they're burning through at incredible speed. These models
    370 are well known to require far more computing power (and thus electricity and
    371 water) than a traditional web search or spellcheck. Although AI company
    372 datacenters are not intentionally wasting electricity in the same way that
    373 bitcoin miners perform millions of useless computations, I'm also not sure that
    374 generating a picture of a person with twelve fingers on each hand or text that
    375 reads as though written by an endlessly smiling children's television star
    376 who's being held hostage is altogether that much more useful than a bitcoin.
    377 
    378 There's a huge human cost as well. Artificial intelligence relies heavily upon
    379 "[57]ghost labor": work that appears to be performed by a computer, but is
    380 actually delegated to often terribly underpaid contractors, working in horrible
    381 conditions, with few labor protections and no benefits. There is a huge amount
    382 of work that goes into compiling and labeling data to feed into these models,
    383 and each new model depends on [58]ever-greater amounts of said data — training
    384 data which is well known to be scraped from just about any possible source,
    385 regardless of copyright or consent. And some of these workers suffer serious
    386 psychological harm as a result of exposure to deeply traumatizing material in
    387 the course of sanitizing datasets or training models to perform content
    388 moderation tasks.
    389 
    390 Then there's the question of opportunity cost to those who are increasingly
    391 being edged out of jobs by LLMs,^[59]i despite the fact that AI often can't
    392 capably perform the work they were doing. Should I really be using AI tools to
    393 proofread my newsletters when I could otherwise pay a real person to do that
    394 proofreading? Even if I never intended to hire such a person?
    395 
    396 Finally, there's the issue of how these tools are being used, and the lack of
    397 effort from their creators to limit their abuse. We're seeing them used to
    398 generate disinformation via increasingly convincing [60]deepfaked images,
    399 audio, or video, and the [61]reckless use of them by previously reputable news
    400 outlets and others who publish unedited AI content is also contributing to
    401 misinformation. Even where AI isn't being directly used, it's degrading trust
    402 so badly that people have to question whether the content they're seeing is
    403 generated, or whether the "person" they're interacting with online might just
    404 be ChatGPT. Generative AI is being used to [62]harass and [63]sexually abuse.
    405 Other AI models are enabling [64]increased surveillance in the workplace and
    406 for "security" purposes — where their well-known biases are worsening
    407 discrimination by police who are wooed by promises of "predictive policing".
    408 The [65]list goes on.
    409 
    410 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
    411 
    412 I'm glad that I took the time to experiment with AI tools, both because I
    413 understand them better and because I have found them to be useful in my
    414 day-to-day life. But even as someone who has used them and found them helpful,
    415 it's remarkable to see the gap between what they can do and what their
    416 promoters promise they will someday be able to do. The benefits, though extant,
    417 seem to pale in comparison to the costs.
    418 
    419 But the reality is that you can't build a hundred-billion-dollar industry
    420 around a technology that's kind of useful, mostly in mundane ways, and that
    421 boasts perhaps small increases in productivity if and only if the people who
    422 use it fully understand its limitations. And you certainly can't justify the
    423 kind of exploitation, extraction, and environmental cost that the industry has
    424 been mostly getting away with, in part because people have believed their lofty
    425 promises of someday changing the world.
    426 
    427 I would love to live in a world where the technology industry widely valued
    428 making incrementally useful tools to improve peoples' lives, and were honest
    429 about what those tools could do, while also carefully weighing the technology's
    430 costs. But that's not the world we live in. Instead, we need to push back
    431 against endless tech manias and overhyped narratives, and oppose the
    432 "innovation at any cost" mindset that has infected the tech sector.
    433 
    434 Footnotes
    435 
    436  1. When I refer to "AI" in this piece, I'm mostly referring to the much
    437     narrower field of [66]generative artificial intelligence and [67]large
    438     language models (LLMs), which is what people generally mean these days when
    439     they say "AI". [68]↩
    440 
    441  2. While much fun has been made of those describing themselves as "prompt
    442     engineers", I have to say I kind of get it. It takes some experience to be
    443     able to open up a ChatGPT window or other LLM interface and actually
    444     provide instructions that will produce useful output. I've heard this
    445     compared to "google-fu" in the early days of Google, when the search engine
    446     was much worse at interpreting natural language queries, and I think that's
    447     rather apt. [69]↩
    448 
    449  3. ChatGPT was publicly released in November 2022, right as the cryptocurrency
    450     industry was in peak meltdown. [70]↩
    451 
    452  4. Many of which are built with various other kinds of machine learning or
    453     artificial intelligence, if not necessarily generative AI. [71]↩
    454 
    455  5. As it happens, he has also [72]written about the "AI isn't useful"
    456     criticism. [73]↩
    457 
    458  6. Some AI boosters will argue that most or all original thought is also
    459     merely a mashup of other peoples' thoughts, which I think is a rather
    460     insulting minimization of human ingenuity. [74]↩
    461 
    462  7. Nor do I want to, by the way. I performed these tests for the purposes of
    463     illustration, but I neither intend nor want to start using these tools to
    464     replace my writing. I'm here to write, and you're here to read my writing,
    465     and that's how it will remain. See my [75]about page. [76]↩
    466 
    467  8. Something that is absolutely encouraged by the tools' creators, who give
    468     them chat-like interfaces, animations suggesting that the tool is "typing"
    469     messages back to you, and a confident writing style that encourages people
    470     to envision the software as another thinking human being. [77]↩
    471 
    472  9. Or, more accurately, by managers and executives who believe the marketing
    473     hype out of AI companies that proclaim that their tools can replace
    474     workers, without seeming to understand at all what those workers do. [78]↩
    475 
    476 Loved this post? Consider [79]signing up for a pay-what-you-want subscription
    477 or [80]leaving a tip to support Molly White's work, which is entirely funded by
    478 readers like you.
    479 
    480 Read more
    481 
    482 [81] A hand holds a gold "Bitcoin" coin that has been cut in half
    483 
    484 Issue 55 – Halving a bad time
    485 
    486 The bitcoin "halving" looms, and that may not be as good news as coiners hope.
    487 Also, Terra committed fraud and Uniswap got a Wells notice.
    488 
    489 Apr 13, 2024
    490 [82] A collage of a Bored Ape, a photo of Ryder Ripps, and the BAYC logo
    491 
    492 "The Monkey Fraud": An interview with Ryder Ripps
    493 
    494 An interview with Ryder Ripps, a defendant in the Yuga Labs v. Ripps case about
    495 Bored Ape Yacht Club trademark infringement and racism.
    496 
    497 Apr 4, 2024
    498 [83] A small fake critter with orange bristles, grey-blue skin, and large black
    499 eyes, with its mouth agape
    500 
    501 Issue 54 – Cases continue
    502 
    503 Crypto-related litigation is in full swing, as the Terra civil fraud trial has
    504 kicked off and two other cases against crypto companies have survived motions
    505 to dismiss.
    506 
    507 Apr 2, 2024
    508 [84] B&W photo of Sam Bankman-Fried overlaid on a yellow-tinted photo of the
    509 NYC courthouse with scrap of sentencing transcript
    510 
    511 25 years for Sam Bankman-Fried
    512 
    513 "The judgment has to adequately reflect the seriousness of the crime, and this
    514 was a very serious crime."
    515 
    516 Mar 28, 2024
    517 
    518 Citation Needed features critical coverage of the cryptocurrency industry and
    519 of issues in the broader technology world.
    520 
    521 It is independently published by Molly White, and entirely supported by readers
    522 like you.
    523 
    524 [85]Subscribe
    525 
    526   • [86]Archive
    527   • [87]Recap issues
    528   • [88]Podcast feed
    529   • [89]About
    530   • [90]RSS
    531   • [91]Tip jar
    532   • [92]Privacy policy
    533 
    534   • [93]Twitter
    535   • [94]Mastodon
    536   • [95]Bluesky
    537   • [96]YouTube
    538   • [97]TikTok
    539   • [98]Etc.
    540 
    541 © 2024 Molly White.
    542 
    543 References:
    544 
    545 [1] https://www.citationneeded.news/
    546 [4] https://www.citationneeded.news/page/2/
    547 [5] https://www.citationneeded.news/tag/weekly-recaps/
    548 [6] https://www.citationneeded.news/podcast/
    549 [7] https://www.citationneeded.news/about/
    550 [8] https://www.citationneeded.news/rss/
    551 [9] https://donate.stripe.com/14k8AseTNaqLaZy7ss
    552 [10] https://www.citationneeded.news/privacy/
    553 [12] https://www.citationneeded.news/ai-isnt-useless/#/portal/signin
    554 [13] https://www.citationneeded.news/signup
    555 [23] https://www.citationneeded.news/tag/newsletter/
    556 [24] https://www.citationneeded.news/author/molly/
    557 [25] https://www.citationneeded.news/author/molly/
    558 [33] https://www.citationneeded.news/content/media/2024/04/2024-04-16-AI-isnt-useless-voiceover.mp3
    559 [34] https://www.citationneeded.news/podcast/
    560 [35] https://www.citationneeded.news/ai-isnt-useless/#footnote-1
    561 [36] https://www.citationneeded.news/signup
    562 [37] https://www.citationneeded.news/ai-isnt-useless/#footnote-2
    563 [38] https://www.citationneeded.news/ai-isnt-useless/#footnote-3
    564 [39] https://www.citationneeded.news/ai-isnt-useless/#footnote-4
    565 [40] https://en.wikipedia.org/wiki/Integrated_development_environment
    566 [41] https://www.mollywhite.net/blogroll
    567 [42] https://simonwillison.net/2023/Mar/27/ai-enhanced-development/
    568 [43] https://simonwillison.net/2024/Mar/22/claude-and-chatgpt-case-study/#not-notable
    569 [44] https://www.citationneeded.news/ai-isnt-useless/#footnote-5
    570 [45] https://www.harihareswara.net/posts/2022/speech-to-text-with-whisper-how-i-use-it-why/
    571 [46] https://www.youtube.com/watch?v=SaCVSUbYpVc
    572 [47] https://github.com/molly/courtlistener-download
    573 [48] https://www.pbs.org/newshour/economy/sports-illustrated-found-publishing-ai-generated-stories-photos-and-authors
    574 [49] https://www.washingtonpost.com/media/2023/01/17/cnet-ai-articles-journalism-corrections/
    575 [50] https://docs.google.com/spreadsheets/d/e/2PACX-1vSYCmj-hH51Yz8uc0W55lYYvrCD3464Jb2uULI9_aqsZ6ZGsLl-5IXYv1HkNiNagLFyYgtMubGQx-Ns/pubhtml
    576 [51] https://www.citationneeded.news/ai-isnt-useless/#footnote-6
    577 [52] https://www.citationneeded.news/ai-isnt-useless/#footnote-7
    578 [53] https://www.citationneeded.news/ai-isnt-useless/#footnote-8
    579 [54] https://en.wikipedia.org/wiki/Enshittification
    580 [55] https://en.wikipedia.org/wiki/Keyword_stuffing
    581 [56] https://en.wikipedia.org/wiki/Content_farm
    582 [57] https://en.wikipedia.org/wiki/Ghost_work
    583 [58] https://www.wsj.com/tech/ai/ai-training-data-synthetic-openai-anthropic-9230f8d8
    584 [59] https://www.citationneeded.news/ai-isnt-useless/#footnote-9
    585 [60] https://apnews.com/article/artificial-intelligence-elections-disinformation-chatgpt-bc283e7426402f0b4baa7df280a4c3fd
    586 [61] https://www.newsguardtech.com/special-reports/ai-tracking-center/
    587 [62] https://www.nytimes.com/2024/01/08/technology/ai-4chan-online-harassment.html
    588 [63] https://www.technologyreview.com/2023/12/01/1084164/deepfake-porn-scandal-pushing-us-lawmakers/
    589 [64] https://www.theguardian.com/technology/2024/jan/07/artificial-intelligence-surveillance-workers
    590 [65] https://incidentdatabase.ai/
    591 [66] https://en.wikipedia.org/wiki/Generative_artificial_intelligence
    592 [67] https://en.wikipedia.org/wiki/Large_language_model
    593 [68] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-1
    594 [69] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-2
    595 [70] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-3
    596 [71] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-4
    597 [72] https://simonwillison.net/2024/Mar/22/claude-and-chatgpt-case-study/#llms-are-useful
    598 [73] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-5
    599 [74] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-6
    600 [75] https://www.citationneeded.news/about/
    601 [76] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-7
    602 [77] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-8
    603 [78] https://www.citationneeded.news/ai-isnt-useless/#footnote-anchor-9
    604 [79] https://www.citationneeded.news/signup
    605 [80] https://donate.stripe.com/14k8AseTNaqLaZy7ss
    606 [81] https://www.citationneeded.news/issue-55/
    607 [82] https://www.citationneeded.news/yuga-labs-ryder-ripps-interview/
    608 [83] https://www.citationneeded.news/issue-54/
    609 [84] https://www.citationneeded.news/sam-bankman-fried-sentenced/
    610 [85] https://www.citationneeded.news/signup
    611 [86] https://www.citationneeded.news/page/2/
    612 [87] https://www.citationneeded.news/tag/weekly-recaps/
    613 [88] https://www.citationneeded.news/podcast/
    614 [89] https://www.citationneeded.news/about/
    615 [90] https://www.citationneeded.news/rss/
    616 [91] https://donate.stripe.com/14k8AseTNaqLaZy7ss
    617 [92] https://www.citationneeded.news/privacy/
    618 [93] https://twitter.com/molly0xFFF/
    619 [94] https://hachyderm.io/@molly0xfff
    620 [95] https://bsky.app/profile/molly.wiki
    621 [96] https://www.youtube.com/@molly0xfff
    622 [97] https://www.tiktok.com/@molly0xfff
    623 [98] https://www.mollywhite.net/