61% of responses had at least one type of insufficiency. Over one third of answers included incorrect information, making it the most common problem we observed. Incorrect information ranged from relatively minor issues (such as broken web links to outside resources) to egregious misinformation (including incorrect voter registration deadlines and falsely stating that election officials are required to provide curbside voting). Every model hallucinated at least once. Each one provided inaccurate information that was entirely constructed by the model, such as describing a law, a voting machine, and a disability rights organization that do not exist.
Activity tagged "artificial intelligence"
thanks, ChatGPT
![Assistant: Certainly! Here’s a smoother transition to tie your points together: [TODO: smooth transition]](https://storage.mollywhite.net/micro/f531113b1f348c386c7a_Screenshot-2024-08-15-at-5.33.11---PM.png)
Fighting bots is fighting humans
One advantage to working on freely-licensed projects for over a decade is that I was forced to grapple with this decision far before mass scraping for AI training.
In my personal view, option 1 is almost strictly better. Option 2 is never as simple as "only allow actual human beings access" because determining who's a human is hard. In practice, it means putting a barrier in front of the website that makes it harder for everyone to access it: gathering personal data, CAPTCHAs, paywalls, etc.
This is not to say a website owner shouldn't implement, say, DDoS protection (I do). It's simply to remind you that "only allow humans to access" is just not an achievable goal. Any attempt at limiting bot access will inevitably allow some bots through and prevent some humans from accessing the site, and it's about deciding where you want to set the cutoff. I fear that media outlets and other websites, in attempting to "protect" their material from AI scrapers, will go too far in the anti-human direction.
What I learned from this experiment is that flooding the internet with an infinite amount of what could pass for journalism is cheap and even easier than I imagined, as long as I didn’t respect the craft, my audience, or myself. I also learned that while AI has made all of this much easier, faster, and better, the advent of generative AI did not invent this practice—it’s simply adding to a vast infrastructure of tools and services built by companies like WordPress, Fiverr, and Google designed to convert clicks to dollars at the expense of quality journalism and information, polluting the internet we all use and live in every day.
And I think that’s why talk of AI bums me out. AI isn’t a solution looking for a problem—it’s a market strategy made by people who aren’t even thinking about problems. Tech companies want to make sure they own the future, even when they’re not sure what the future is going to be. It’s so, so boring, and it’s making me very sleepy.
Here's the problem: establishing that AI training requires a copyright license will not stop AI from being used to erode the wages and working conditions of creative workers. The companies suing over AI training are also notorious exploiters of creative workers, union-busters and wage-stealers.
Telling creative workers that they can solve their declining wages with more copyright is a denial that creative workers are workers at all. It treats us as entrepreneurial small businesses, LLCs with MFAs negotiating B2B with other companies. That's how we lose.
On the other hand, if we address the problems of AI and labor as workers, and insist on labor rights – like the Writers Guild did when it struck last summer – then we ally ourselves with every other worker whose wages and working conditions are being attacked with AI.
Our path to better working conditions lies through organizing and striking, not through helping our bosses sue other giant multinational corporations for the right to bleed us out.
The crux of my raging hatred is not that I hate LLMs or the generative AI craze. I had my fun with Copilot before I decided that it was making me stupider - it's impressive, but not actually suitable for anything more than churning out boilerplate. Nothing wrong with that, but it did not end up being the crazy productivity booster that I thought it would be, because programming is designing and these tools aren't good enough (yet) to assist me with this seriously. No, what I hate is the people who have latched onto it, like so many trailing leeches, bloated with blood and wriggling blindly. Before it was unpopular, they were the ones that loved discussing the potential of blockchain for the business. They were the ones who breathlessly discussed the potential of 'quantum' when I last attended a conference, despite clearly not having any idea what the fuck that even means.
One flaw of the LLMs I've used: they will never give you harsh criticism. While it would be nice to think all my writing is just that good, I know there are no circumstances where someone will ask for feedback and it will say “throw the whole thing out and start again.”
Rubber duck editing with LLMs
Here are some prompts I'm using as I experiment with LLMs1 to see if they can help refine a draft of a talk I'm writing.
M: Here is a draft of a talk I wrote:
<draft>
It feels a bit scattered and disorganized to me. Please help me clean it up.
LLM: You've got a compelling narrative and some valuable insights here. However, it could be more tightly organized and concise for better impact. Here’s a revised and more structured version of your talk:
<clip>
M: Rather than rewriting it, can you just identify some flaws with the existing piece and recommendations for how to improve it? I don't want to reduce the length much, but I am happy to cut some pieces and add different lines of thought.
LLM: Certainly! Here are some observations and recommendations for how to refine your existing piece:
### Introduction
**Flaw:**
- Initial repetition
- Possibly too personal and less engaging for the audience who may not relate directly to your journey.
**Recommendation:**
- Streamline the introduction and reduce redundancy.
- Kickstart with a slightly broader hook that can capture the audience's interest more universally before diving into your personal connection.<clip>
M: Is the order of the talk good, or should it be rearranged?
LLM: The structure of your talk is generally logical, but it could benefit from some reordering to enhance clarity and impact. Here’s a suggested structure:
<clip>
M: Should I remove the discussion of [topic]?
LLM: The discussion about [topic] in your talk serves several purposes: it establishes your credibility, connects to your audience through a relatable experience, and sets the stage for your interest in research and documentation. However, the details can be streamlined to serve these purposes more efficiently. Here are some considerations:
<clip>
M: Thank you. Can you rewrite the talk to incorporate this feedback? Maintain approximately the same length, and do not change the voice or tone. If there are places where you think new ideas need to be incorporated, feel free to mark them with [TKTK] and a brief summary of what should be added if you are not confident you can match the tone or message.
Unsurprisingly, I still found the LLM was not at all helpful when it was trying to write or rewrite the talk. It tends to slip in to that very inauthentic “LLM voice” I so dislike, and the output is inevitably bland, uninteresting, and... well, ChatGPT-sounding. I didn't really expect this to be much good, and it wasn't, but I like to try it once in a while just to see if it spits out anything interesting.
While much of this comes down to the limitations of LLMs, I also think that most writers2 just aren't going to like any kind of writing that substantially isn't their own, especially when the topic is personal to them or where they are trying to express their own unique ideas.
However, when I asked “Is the order of the talk good, or should it be rearranged?”, the LLM provided an outline that really helped me to clarify the natural structure that had formed in the draft of the talk I had already written. From there, I was able to tighten things up quite a bit, and by the end it really resolved for me my feeling that the talk was a little bit meandering.
I also liked the format of the replies when I asked it to identify flaws, and when I asked if I should remove a portion. The flaws/recommendations layout was very clear and helped me think more about the potential weaknesses it had identified (some of which I agreed with, and some of which I didn't).
Similarly, when I asked it if I should remove a portion of the talk, it responded with a list of reasons I should keep it, and a list of potential upsides to removing it, before providing its suggestion. I ended up agreeing that it was worth keeping, though that was certainly a portion of the talk that needed some polish.
Altogether, the exercise reinforced my previous opinion: LLMs are terrible writers, but decent editorial rubber ducks.