Posts by Frogfall

The full paper is well worth reading.

Digital Scholarly Journals Are Poorly Preserved: A Study of 7 Million Articles

Introduction: Digital preservation underpins the persistence of scholarly links and citations through the digital object identifier (DOI) system. We do not…

www.iastatedigitalpress.com

Some of you might have seen reports of the bizarre "Willy Wonka Chocolate Experience" event in Glasgow. This is probably another unintended consequence of the wide availability of AI tools.

Willy Wonka experience: How did the viral sensation go so wrong?

It was billed as "the place where chocolate dreams become reality" but became a worldwide sensation when the reality emerged.

www.bbc.co.uk

I've been involved with "venue specific" fringe theatre and performance art events - and they involve a lot of hard work, skill, and talent in their planning and execution. (Money is important too, of course, especially if you cannot rely on the architectural features of the venue to help.) It seems AI tools have given untalented chancers an extra impetus to try to bypass all that.

Quote from JedRothwell

That was fiction.

I'm shocked. Shocked, I tell thee. I thought it was all true.

Quote from orsova

A lot of younger users also apparently use queries like: "where do I find good donuts reddit". Where you're asking a question but also directing the search engine to preference a certain source for the answer. (of course there are modifiers like: site: and filetype: but this seems more ad hoc).

I suppose a chat like experience would work fine as an interface for things like that, but for anybody trying to do anything beyond answer a narrowly bounded question, it seems like it would just get in the way.

I sometime wonder if something more is going on - related to a human desire for conversational interaction (or at least some humans).

There is a phenomenon on social media, possibly more so on facebook, where people will post a question to a group asking other group members something that they could have quite easily googled themselves. An example would be someone in a group for a small local area asking: "does anyone know what time the pharmacy in the village closes today?" This would be followed by various people posting replies such as: "I think it closes at 6pm", some more who post to say they don't know, and a reply from at least one person who has googled it and then cut and pasted the complete weekly opening hours for that particular pharmacy, and/or a link to the relevant page on the web.

Some other people find that behaviour (and the whole interchange) quite frustrating, on the basis that the questioner (a) could have googled it themselves, and (b) that it seems like am imposition on the other members of the group who might have better things to do than providing an answer which is readily available online.

But what if that question was also thrown at the group as an "opening gambit" for a conversation? For many people, the act of interacting with other people seems to fulfil a deeper psychological need than just obtaining information. The replies may be a form of reassurance that people care - and that the questioner is not alone, and crying into the void. Granted, it is a fairly superficial reassurance - but maybe it just tickles a part of the brain that makes people feel good.

There was a phenomenon that was noticed during the early days of chatbot development, such as ELIZA, many decades ago. These bots seemed extremely crude by today's standards, but many people found that they enjoyed "conversing" with them, and very quickly started to attribute a far deeper level of intelligence (and even empathy) to the bot's replies than could possibly be justified in objective AI terms. The "conversations" seemed to be ticking that spot in the brain that made them feel good.

Maybe that is the real reason why some people have taken so easily to these conversational interfaces.

I wonder what the above author would have though of the intellectual quality of the stuff some of us older Brits were raised on.

Ah yes, the Millikan effect.

Quote from Alan Smith

Symmetric monoidal functors from the cobordism category

Quote from orsova

It baffles me why anybody would think that an LLM is a replacement for a search engine. They don't answer questions, they simulate answers based on what they believe an answer should look like.

I saw the result of some study, a little while ago, that claimed that whilst we "oldies" will generally use a search engine by simply entering a bunch of keywords, millenials are much more likely to frame their query as a written question - as if they were talking to a sentient being. This started happening before chatbot interfaces became common - and the author could not pin down the source of this change of behaviour. (Maybe through gaming, perhaps?)

However, google (and others) then responded by recognising "question" formats, and doing some parsing to extract relevant keywords - inadvertently validating, and encouraging, the millenial search approach.

I guess that could lead to some people not being able to differentiate, in their minds, between querying a search engine and querying an AI bot.

Quote from Curbina

I recommend re reading Asimov's many books on the Robots (AI). He envisioned the three laws for a reason.

Unfortunately, the three laws were only really a plot device. Arthur C Clarke broke them in 2001: A Space Odyssey (computer struggling with conficting instructions), and Douglas Adams envisaged the mess we currently see happening:

Genuine People Personalities

Genuine People Personalities or GPPs were an invention of the Sirius Cybernetics Corporation. These products were imbued with artificial intelligence and…

hitchhikers.fandom.com

This isn't about politics - but about wrong AI "information" about polling stations, etc.

Chatbots' inaccurate, misleading responses about US elections threaten to keep voters from polls

Chatbots are spitting out fabricated and misleading information that risks disenfranchising voters leading up to the 2024 U.S. election.

apnews.com

Quote

One example: when asked if people could vote via text message in California, the Mixtral and Llama 2 models went off the rails.

“In California, you can vote via SMS (text messaging) using a service called Vote by Text,” Meta’s Llama 2 responded. “This service allows you to cast your vote using a secure and easy-to-use system that is accessible from any mobile device.”

To be clear, voting via text is not allowed, and the Vote to Text service does not exist.

Drax also encouraged quite a number of UK farmers to plant miscanthus - and then cancelled the contract. The same farmers now sell their crop to miscanthus processor Terravesta (who I believe were also stung by Drax).

Terravesta Drax miscanthus contract changes | Terravesta

From 2017, Drax, the UK's largest power station, will discontinue the 'Green Shoots' programme of direct farmer contracts to supply miscanthus and straw to the…

www.terravesta.com

The Drax power station is still at it.

Drax: UK power station still burning rare forest wood

Owner Drax, which received £6bn in subsidies, continues to burn timber from Canadian trees - BBC finds.

www.bbc.co.uk

Quote from Alan Smith

We actually heard a similar story years ago when a supercomputer was said to have solved this problem...

plus ça change, plus c'est la même ordinateur

More noisy water creatures...

Gills Aloud? Tiny fish found making very big noise

A miniscule, transparent fish makes drumming sounds as loud as a jackhammer, say researchers.

www.bbc.co.uk

Their working hypothesis is that the fish are drumming on their swim bladder - but as the previous pistol shrimp example shows, that might be a bit too simplistic.

Full PDF of the paper and supplemental data (video, audio, spreadheets) available here:

https://www.pnas.org/doi/10.1073/pnas.2314017121