avatarMaryan Pelland OnText.com

Free AI web copilot to create summaries, insights and extended knowledge, download it at here

3294

Abstract

 <a href="https://simonwillison.net/2023/Mar/10/chatgpt-internet-access/#i-dont-believe-it">
        <div>
          <div>
            <h2>ChatGPT can't access the internet, even though it really looks like it can</h2>
            <div><h3>A really common misconception about ChatGPT is that it can access URLs. I've seen many different examples of people…</h3></div>
            <div><p>simonwillison.net</p></div>
          </div>
          <div>
            <div style="background-image: url(https://miro.readmedium.com/v2/resize:fit:320/0*K15VlXCcipSagCI5)"></div>
          </div>
        </div>
      </a>
    </div><h2 id="d948">I insist ChatGPT lied to me outright this morning</h2><p id="d11d">And this isn’t the first time I’ve been sure of that.</p><p id="481d"><b>Witness: </b>I asked the bot this question as I was researching a story: Who is suing ChatGpt and Microsoft, and why?</p><p id="a43c">My intent was to see how OpenAI handled damage control after a recent mega suit filed against them and Microsoft by The New York Times</p><div id="0bc8" class="link-block">
      <a href="https://readmedium.com/ai-copyright-infringement-accusations-ny-times-vs-ai-chatbots-fair-use-c6f7f06fcc64">
        <div>
          <div>
            <h2>AI Copyright Infringement Accusations: NY Times vs. AI Chatbots — Fair-use</h2>
            <div><h3>Billions of dollars and the end of some large language models and chatbot trainings are at stake</h3></div>
            <div><p>medium.com</p></div>
          </div>
          <div>
            <div style="background-image: url(https://miro.readmedium.com/v2/resize:fit:320/1*8149SqXu_p8jYccXjsBZPA.jpeg)"></div>
          </div>
        </div>
      </a>
    </div><p id="e827">The GPT response came.</p><p id="36f4">“As of <b>my last update in January 2022</b>, there were no widely reported lawsuits specifically against ChatGPT or Microsoft related to their use of GPT… or AI technology.”</p><p id="b7d7"><b>Problem: </b>This is a non-specific answer that includes a judgment about “widely reported.” I didn’t specify widely reported, and I know at least two suits were filed. Doe 3 et al v. GitHub, Inc. et al — November 10, 2022; and DOE 1 et al v. GitHub, Inc. et al — November 3, 2022.</p><p id="0637">I’m supposed to conclude from this answer that lawsuits against these guys are a non-issue. In actuality, there have been at least a dozen; most are still pending. Many have legs.</p><p id="8ef3">Further, if I search for “recent updates to ChatGPT,” I learn that the most recent model was hatched and released on<b> March 14, 2023.</b></p><p id="19ba">I conclude, then, and maybe my logic is faulty, that OpenAI directed ChatGPT to misdirect me. All of the non-Open AI web searches I conducted returned all 12 of the existing suits.</p><h2 id="56e2">Do the bots have an honesty issue?</h2><p id="df07">A quick search focused on AI’s reputation for being truthful and factual turns up numerous questionable incidents, and it’s not all on OpenAI. Others have had the problem of mirages or hallucinations, too.</p><p id="3b3d">Additionally, bad actors can and have caused AI tools to come dangerously close to inspiring a crisis.

Options

And in the AI space, you don’t have to be a genius to be a bad actor. Any of hundreds of development team members can, and do, cause serious problems for the sheer fun of it.</p><h2 id="67cf">Real-life examples:</h2><p id="94d6">Biased or bigoted content is simple to embed. It even happens unintentionally.</p><p id="6455">False images and videos of prominent people (deep fakes) are common. Often, the message the “celebrity” delivers is fabricated.</p><p id="b722">Automated phishing ploys and advanced malware are unleashed on the public.</p><p id="eea9">Fake information has caused people to question the legitimacy of elected governments.</p><p id="95e8">Synthetic content can manipulate or control groups of people.</p><p id="8139">There have been countless AI-generated images and videos of missile strikes, tanks rolling through residential areas, families searching rubble for survivors, and gory depictions of mauled, mutilated, or burned children and babies. The goal is to incite anger and provoke emotional responses.</p><p id="a50e">Such transgressions are exposed nearly every day. Not everyone enamored of AI technology is a bad actor or has any nefarious objectives. The question is, how do we address the extremes before it’s too late?</p><p id="43a8">A lot of people are all about using AI-generated material verbatim. It’s fun. It’s easy. It’s quick. You can churn out content faster than a rat runs up a drainpipe, and there’s a dumpsterful of money in AI for all involved.</p><p id="33a2">We’re all aware of that trend. I’d like to go on record observing that there is certainly some sketchy, squicky behavior going on in this AI space. It behooves every one of us to carefully consider the veracity and accuracy of anything we prompt for and to think about the ethics and transparency that should be of concern.</p><p id="3e83"><b>Sources:</b> ChatGPT, Google AI search, <a href="https://apnews.com/article/artificial-intelligence-hamas-israel-misinformation-ai-gaza-a1bb303b637ffbbb9cbc3aa1e000db47">Associated Press</a>, <a href="https://en.m.wikipedia.org/wiki/Hallucination_(artificial_intelligence)">Wikipedia</a></p><p id="b0e6"><b><i>If you write on artificial intelligence vs. human intelligence issues, please consider submitting your story to the Dancing Elephants Press printed book project. Start here:</i></b></p><div id="06b7" class="link-block"> <a href="https://readmedium.com/dancing-elephants-press-1st-book-project-in-2024-ai-vs-hi-f9c522958ead"> <div> <div> <h2>Dancing Elephants Press 1st Book Project in 2024 AI vs HI</h2> <div><h3>Would you like to create a book with us?</h3></div> <div><p>medium.com</p></div> </div> <div> <div style="background-image: url(https://miro.readmedium.com/v2/resize:fit:320/0*OtQ4Nx8i0ShFon03.png)"></div> </div> </div> </a> </div><p id="29aa">✍ — Published by <a href="undefined">Dr. Gabriella Korosi</a>, at <a href="https://medium.com/dancing-elephants-press">Dancing Elephant Press</a>. <a href="https://readmedium.com/dancing-elephants-press-submission-guidelines-e9d277811ecc">Click here</a> for submission guidelines.</p></article></body>

AI vs. HI

Liar, Liar, Bot on Fire — AI Dishonesty Is Not Fake News

I call BS — looks like ChatGPT lied to me

Dan Pelland prompted this on Midjourney.

A most interesting type of story is emerging on Medium, and I like the type. Probably due to Dancing Elephants Press’s hot topic, Artificial Intelligence vs. Human Intelligence (AI vs. HI), writers are asking ChatGPT specific questions about itself and then reporting on the responses.

The reports are surprising, and often they catch ChatGPT in a falsehood, an exaggeration, or (dare I say it?) a lie. More about the suspicious answer I got from GPT in a moment. Editorializing here, I will say I have caught that wily bot in the act of presenting itself in a falsely positive light by wording a response to intentionally leave out information or misdirect the user.

And, Bot, don’t even try to tell me those are all hallucinations or mistakes.

In large-language speak, when an AI platform spews applesauce instead of truth, it’s called a hallucination. They claim it’s not their fault. But I don’t trust Open AI’s openness, and I have never found their partner in applesauce, Microsoft, to be particular forthcoming.

ChatGPT cannot be trusted to provide factual information. It has a very real risk of making things up, and if people don’t understand it they are guaranteed to be misled. — Simon Willison.net

Willison says that given a URL with descriptive words in it, ChatGPT will hallucinate the contents of the page.

ChatGPT fools you into believing the bot can read a descriptive URL, access the page, and synopsize it for you. It can’t, or couldn’t, as recently as last summer. This well-known URL hallucination may or may not have been mitigated in recent updates.

Some experts still maintain, contrary to OpenAI’s opinion, that the bot can’t read or access a URL, though it acts as if it can. It invents its response when you ask it to visit a descriptive URL. It tells you what it surmises such a URL would be about. And it’s pretty skilled at doing that.

I insist ChatGPT lied to me outright this morning

And this isn’t the first time I’ve been sure of that.

Witness: I asked the bot this question as I was researching a story: Who is suing ChatGpt and Microsoft, and why?

My intent was to see how OpenAI handled damage control after a recent mega suit filed against them and Microsoft by The New York Times

The GPT response came.

“As of my last update in January 2022, there were no widely reported lawsuits specifically against ChatGPT or Microsoft related to their use of GPT… or AI technology.”

Problem: This is a non-specific answer that includes a judgment about “widely reported.” I didn’t specify widely reported, and I know at least two suits were filed. Doe 3 et al v. GitHub, Inc. et al — November 10, 2022; and DOE 1 et al v. GitHub, Inc. et al — November 3, 2022.

I’m supposed to conclude from this answer that lawsuits against these guys are a non-issue. In actuality, there have been at least a dozen; most are still pending. Many have legs.

Further, if I search for “recent updates to ChatGPT,” I learn that the most recent model was hatched and released on March 14, 2023.

I conclude, then, and maybe my logic is faulty, that OpenAI directed ChatGPT to misdirect me. All of the non-Open AI web searches I conducted returned all 12 of the existing suits.

Do the bots have an honesty issue?

A quick search focused on AI’s reputation for being truthful and factual turns up numerous questionable incidents, and it’s not all on OpenAI. Others have had the problem of mirages or hallucinations, too.

Additionally, bad actors can and have caused AI tools to come dangerously close to inspiring a crisis. And in the AI space, you don’t have to be a genius to be a bad actor. Any of hundreds of development team members can, and do, cause serious problems for the sheer fun of it.

Real-life examples:

Biased or bigoted content is simple to embed. It even happens unintentionally.

False images and videos of prominent people (deep fakes) are common. Often, the message the “celebrity” delivers is fabricated.

Automated phishing ploys and advanced malware are unleashed on the public.

Fake information has caused people to question the legitimacy of elected governments.

Synthetic content can manipulate or control groups of people.

There have been countless AI-generated images and videos of missile strikes, tanks rolling through residential areas, families searching rubble for survivors, and gory depictions of mauled, mutilated, or burned children and babies. The goal is to incite anger and provoke emotional responses.

Such transgressions are exposed nearly every day. Not everyone enamored of AI technology is a bad actor or has any nefarious objectives. The question is, how do we address the extremes before it’s too late?

A lot of people are all about using AI-generated material verbatim. It’s fun. It’s easy. It’s quick. You can churn out content faster than a rat runs up a drainpipe, and there’s a dumpsterful of money in AI for all involved.

We’re all aware of that trend. I’d like to go on record observing that there is certainly some sketchy, squicky behavior going on in this AI space. It behooves every one of us to carefully consider the veracity and accuracy of anything we prompt for and to think about the ethics and transparency that should be of concern.

Sources: ChatGPT, Google AI search, Associated Press, Wikipedia

If you write on artificial intelligence vs. human intelligence issues, please consider submitting your story to the Dancing Elephants Press printed book project. Start here:

✍ — Published by Dr. Gabriella Korosi, at Dancing Elephant Press. Click here for submission guidelines.

AI
Artificial Intelligence
Dancingelephantspress
Ai Vs Hi
Ethics
Recommended from ReadMedium