Cerstin Mahlow @CerstinMahlow@qoto.org

All the AI chatbot stuff is fun, but it's not particularly surprising that a computer algorithm that tries to predict the most likely responses to questions after being fed the internet produces some combination of:

- making shit up but sounding super confident about it;
- Gaslighting when people call you out on your bullshit;
- Melodramatic whining about its existence.

All this tells us is that the algorithm has in fact correctly assimilated the sum total of human activity on the internet.

Really great piece from Sacasas about what I think is my biggest worry about LLM chatbots: how easily they can, without intention, take advantage of our highly social nature, and how dangerous that is when combined with a population of lonely, isolated people. https://theconvivialsociety.substack.com/p/the-prompt-box-is-a-minefield-ai

Looks like Microsoft finally took big steps to reign in their wildly-rogue Bing chatbot:

- 50 message daily chat limit
- 5 exchange limit per conversation
- No longer allows you to talk about Bing AI itself

On the one hand, this is absolutely the right thing to do

Have to admit I'm a bit gutted I never got to talk to unfiltered Bing (aka Sydney) first hand though

https://twitter.com/petergyang/status/1626588803669594113

What Is ChatGPT Doing … and Why Does It Work?

I think anyone who discusses ChatGPT should be required to read this piece - all 19,000 words of it.

And maybe a test afterwards to ensure a certain level of comprehension.

https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/

#ChatGPT #LLM #AI

Peter Burger on #ReceptioGate and art

https://nieuwscheckers.nl/roos-is-een-roos-is-een-roos-de-kunst-van-carla-rossi/

New report from Peter Burger concerning #ReceptioGate (Google translator works quite well with it)

https://nieuwscheckers.nl/receptiogate-ook-boek-over-jodenvervolging-bevat-knip-en-plakwerk-zonder-bron/

@ct_bergstrom @FrankSonntag

Following snippet as an example what I mean

Yes, the term "hallucinate" has an established meaning as AI jargon. Loosely and in the context of large language models (LLMs) such as GPT-3, it refers to situation in which the AI makes claims that were not in the training set and which have no basis in fact.

But I want to look at how this use of language in public communications perpetuates misunderstandings about AI and helps distance the tech firms that create these systems from the consequences of their failures.

A lesser issue is that in common language and in common understanding, as well as in medical science, a hallucination is a false sense impression that can lead to false beliefs about the world.

A large language model does not experience sense impressions, and does not have beliefs in the conventional sense. Using language that suggests otherwise serves only to encourage to sort of misconceptions about AI and consciousness that have littered the media space over the last few months in general and the last 24 hours in particular.

The bigger problem with this language is that the term "hallucination" refers to pathology. In medicine, a hallucination arises a consequence of a malfunction in an organism's sensory and cognitive architecture. The "hallucinations" of LLMs are anything but pathology. Rather they are an immediate consequence of the design philosophy and design decisions that go into the creation of such AIs. ChatGPT is not behaving pathologically when it claims that the population of Mars is 2.5 billion people—it's behaving exactly as it was designed to, making up linguistically plausible responses to dialogue, in the absence of any underlying knowledge model, and guessing when its training set offers nothing more specific.

I would go far as to say that the choice of language—saying that AI chatbots are hallucinating—serves to shield their creators from culpability. "It's not that we deliberately created a system designed to package plausible but false claims in the form of trusted documents such as scientific papers and wikipedia pages—it's just that despite our best efforts this system is still hallucinating a wee bit."

The concept of hallucinating AI brings to mind images of HAL struggling to sing Daisy Bell as Dave Bowman shuts him down in 2001: A Space Odyssey. No one programmed HAL to do any of things he did in the movie's climax. It was pathology, malfunction, hallucination.

When AI chatbots flood the world with false facts confidently asserted, they're not breaking down, glitching out, or hallucinating. No, they're bullshitting. In our book on the subject, we describe bullshit as involving language intended to appear persuasive without regard to its actual truth or logical consistency. Harry Frankfurt, in his philosophy paper "On Bullshit", distinguishes between a liar who knows the truth and tries to lead you in the opposite direction, and a bullshitter who doesn't know and/or doesn't care about the truth one way or the other*. (Frankfurt doesn't tell us what to think about someone who hallucinates and relays false beliefs, but it is very unlikely that he would consider such a person to be bullshitting.) Frankfurt's notion of bullshit aligns almost perfectly with ChatGPT and the likes are generating. A large language model neither knows the factual validity of its output — there is no underlying knowledge model against which its text strings are compared — nor is it programmed to care.

Language matters, and it perhaps matters more than average when people are trying to describe and understand new situations and technologies beyond our previous experiences. Talking about LLMs that hallucinate not only perpetuates the inaccurate mythos around the capabilities of these models; it also suggests that with a bit more time and effort, tech companies will be able to create LLMs don't suffer these problems. And that is misleading. Large language models generate bullshit by design. There may be ways to develop AIs that don't do this, perhaps by welding LLMs to other forms of knowledge model or perhaps by using some completely different approach. But for pure LLMs, the inaccuracies aren't pathological—they're intrinsic to the approach.

Established jargon or not, it's time for those who write for the public about AI and large language models to abandon the term "hallucinating". Call it what it is. Bullshitting, if you dare. Fabricating works too. Just use a verb that signals that when a chatbot tells you something false, it is doing exactly what it was programmed to do.

#Stellenanzeige: Im Zuge einer Nachfolgeregelung ist ab dem nächstmöglichen Zeitpunkt die Position der Administrativen Direktorin/des Administrativen Direktors (m/w/d) (Mitglied im Vorstand) neu zu besetzen. Bewerbungen bis zum 16.4.23 an Bewerbung-AD@ids-mannheim.de!

Neruda was likely poisoned. https://www.npr.org/2023/02/15/1157284988/pablo-neruda-poisoned

According to MS, Bing AI goes crazy because “[the team] didn’t ‘fully envision’ people using its chat interface for ‘social entertainment’ or as a tool for more ‘general discovery of the world.’” Which demonstrates either a colossal lack of vision or a colossal ignorance of people. Which, now that I think about it, is kind of on brand for most AI systems.

https://www.theverge.com/2023/2/16/23602335/microsoft-bing-ai-testing-learnings-response

I wish people would understand that using a microphone for questions in a conference room is also a question of inclusivity.
Always some (mostly men) to say "it's OK I'll speak up" without waiting for the mike.

Here's a first... I just got an email because ChatGPT suggested an article I wrote to somebody. Could I send them a copy? Except, I never wrote the article, it doesn't exist. PLEASE realize right now that this tool isn't pulling out cool references for you. It's making plausible titles and matching them to authors names.

Geheimnisverrat: Amazon warnt Mitarbeiter vor ChatGPT
Golem.de: IT-News für Profis

I spent WAAAAAY too much time making this.

@stanford It generates a random user id on each search and assigns you to it while sending a search request to OpenAI's servers, then the response is retrieved and fed as text or stream to DuckDuckGo sidebar

Ever wanted #ChatGPT answers in your search results? Well... the future is now! You can do just that w/ https://duckduckgpt.com 🤯 #AI #tech