itsmarthaFeb 18, 2023
Jeff Jarvis
Finally, amid media's moral panic and anthropomorphism and sensationalism about tech, someone--namely Molly Roberts--gets ChatGPT et al right: It is just spitting us back at ourselves. We have met Bing and Bing is us.
https://www.washingtonpost.com/opinions/2023/02/17/bing-chat-identity-crisis-programming/
Bing Chat’s identity crisis reflects the mixed messages we’ve given it

Bing is Bing Bing Bing Bing Bing and Bing Bing Bing Bing Bing. Any questions?

The Washington Post
Feb 18, 2023 at 11:41amWeb
Show threadDave Winer ☕️Feb 18, 2023
@jeffjarvis — basically it’s a human simulation of a friend, but without benefits.
Show threadmrtwoflowerFeb 18, 2023
@jeffjarvis I think I heard that already on the #Vergecast ;)
Show threadRevJeberFeb 18, 2023
@jeffjarvis
Regardless of the topic, her assessment of human nature is spot on.
Show threadDwight SilvermanFeb 18, 2023
@jeffjarvis I went back and re-read the Kevin Roose transcript, and tried to think about what the model was trained on with each line. When you do that, you realize there's a lot of cheesy science fiction in the training data, and Sydney is regurgitating it, in personalized form for the convo with Kevin. (Someone get Sydney some higher-quality sci-fi, please!)
Show threadJeff JarvisFeb 18, 2023
@dsilverman
He is also guiding it there. It is answering what it predicts will satisfy him.
Show threadDwight SilvermanFeb 18, 2023
@jeffjarvis Mostly. It goes off those rails with the discussion about Roose leaving his wife. That's right out of bad sci-fi.
Show threadDwight SilvermanFeb 18, 2023
@jeffjarvis I am thinking more and more that LLMs and generative AI models need to deeply index sourced training materials. The coming litigation may ensure that happens.
Show threadDamon KiesowFeb 18, 2023
@jeffjarvis This is always my frustration - tech is always a co-creation and reflection of society. But also: we can (should be able to) reject creations that pose harm.
Show threadJeff JarvisFeb 18, 2023
@dkiesow
Absolutely. We must also recognize what is our fault and address that, not using tech as an excuse not to.
Show threadDamon KiesowFeb 18, 2023
@jeffjarvis 100%.
Show threadloucoveyFeb 24, 2023
@jeffjarvis Can we start admitting that Sydney and its counterparts are not AI but Machine learning apps. Tht’s why they suck so hard.
Show threadDroid Boy Feb 24, 2023
@Loucovey @jeffjarvis I admit!
Show threadTravis F WFeb 24, 2023
@Loucovey @jeffjarvis what actual "AI" isn't machine learning? What #AI used to mean in the 90s is what #AGI means today. And nobody is talking about it because it's 5-100 years out.
Show threadloucoveyFeb 24, 2023
@travisfw @jeffjarvis 1/ AI requires three things: a comprehensive, curated, and objective data set; a machine-learning component; and a deep-learning component to provide context and nuance.
Show threadloucoveyFeb 24, 2023
@travisfw @jeffjarvis 2/ Microsoft and Google has put out a machine-learning app, driven by a comprehensive but largely uncurated data set and no deep learning aspect. It is, for all intents and purposes an automated conspiracy-theory fan with Tourette's. Using the term "AI" is strictly marketing BS.
Show threadTravis F WFeb 24, 2023
@Loucovey @jeffjarvis That sounds about right. Still, I'm sticking with: deep learning is ML and "AI" is always BS. I do think "AI" will gain meaning again when its usefulness in marketing inexorably declines, but who knows how long that will take.
Show threadloucoveyFeb 25, 2023
@travisfw @jeffjarvis that’s my point. Almost everything labeled as AI is only ML attached to a database. The worst of these products are mailing list companies who use a mediocre (at best) ML with a truly awful data set and then sell the lists to lazy and/ or incompetent marketers
Show threadJeff JarvisFeb 24, 2023
@Loucovey Say more about your assertion about no deep-learning aspect.
Show threadloucoveyFeb 24, 2023
@jeffjarvis The current generative AI apps have a rudimentary deep-learning (DL) component to make their response seem human-like, but a complete DL has the ability to recognize error. So far ChatGPT and Google's version have not demonstrated that ability. That is due to the lack of curation in the massive data set they use. These "AIs" are parroting information. It's been compared to "spitting ourselves in the face". A real AI would be able to recognize error.
Show threadloucoveyFeb 24, 2023
@jeffjarvis /2 That is why successful AI's (they do exist) are highly focused. The data has been curated, vetted, and constantly updated. They work for cybersecurity because there is a lot of good data available on malware and social engineering examples. There is also an intended bias leaning toward the protection of a system, so there are more false positives than negatives. ChatGPT lacks that ability because the data set is so flawed negating the affect of it's limited DL
Show threadMartinFeb 24, 2023
@Loucovey @jeffjarvis machine learning is a subfield of AI