Cerstin Mahlow @CerstinMahlow@qoto.org

@moultano Humans have an underlying knowledge model. They have beliefs about the world, and choose whether to represent those beliefs accurately or inaccurately using language.

LLMs do not have an underlying knowledge model, they don't have a concept of what is true or false in the world. They just string together words they don't "understand" in ways that are likely to seem credible.

It's not a matter of making better LLMs; it'll take a fundamentally different type of model.

Another post introduces the pilot project @peterkaminski & @band & I are working on: using #MassiveWiki to map the #ToolsForThought landscape. Apart from exploring decentralised collective intelligence, we hope to help newcomers adopt #TfTs, which create the "raw material" (networkable knowledge) required for collective intelligence: https://mathewlowry.medium.com/mapping-tools4thought-using-collective-intelligence-tools-9b5cd00a6309

If you want to convince yourself that getting large langauge models to stick to the "truth" is a viciously difficult problem, ask one of them to write a fictional story for you (they're really good at this)

Then ask yourself how something capable of doing that could be engineered to be able to distinguish fact from fiction

⁣

Sehr interessanter Vortrag, große Hörempfehlung: "Würden Sie Ihre Stasi-Akte lesen, wenn Sie eine hätten?"

Mittagssalon mit Ralph Hertwig: Warum sich viele entschieden, ihre Stasi-Akte nicht einzusehen und wie sich dieses „Gewollte Nichtwissen" erklären lässt.

https://www.bbaw.de/mediathek/archiv-2023/wuerden-sie-ihre-stasi-akte-lesen-wenn-sie-eine-haetten

If anyone is wondering if Bing+GPT is somehow going to be better than Google Bard, remember that both LLM systems will be dependent on the underlying search algorithms.

Here is Bing (prior to GPT integration) doing the wrong thing. Bing search is retrieving articles about how Google summary boxes are wrong and then giving the wrong answer.

Both will have the same failure modes. The only difference is that Google stepped in it first.

Source: https://twitter.com/stilgherrian/status/1623576572015050753

In eigener Sache: Dienstagabend, 22 Uhr auf http://radiostonefm.de präsentiere ich eine Stunde Musik von und mit dem letzten Sommer viel zu früh verstorbenen Hammond-Organisten Joey DeFrancesco (1971–2022):

@jazz

https://radiostonefm.de/naechste-sendungen/7822-230214-gypsy-goes-jazz-138-joey-defrancesco

#MusicAgainstMadness

Hey friends listen I was going to cross post my whole #odysseyreactions but...it's a lot and I have work to do so if you WANT to see it (and you should because it's funny), please follow the whole thread here, thanks. https://twitter.com/BeeBrookshire/status/1616447081027338240

Sorry about that.

People want #statistics to tell them the answer to questions like "what percentage of children are homeless?" But #Bayesian stats answers questions like "given a model of geographic distribution of homelessness and how a survey was conducted and the data from the survey **how much do I know** about how many children are homeless?"

Sometimes the answer is "not much" and this is a very precise and correct answer.

#ChatGPT

Interesse und Freude an #Tontechnik und #Literatur?
Das @DLAMarbach sucht für seine #Bibliothek #Mediengestalter Bild und Ton (m/w/d -Teilzeit 50 %) für die vielfältigen historischen und aktuellen Audiobestände.
https://www.dla-marbach.de/ueber-uns/stellenangebote/aktuelle-stellenangebote/bibliothekar-oder-archivar-m/w/d-kennziffer-21-55/

lmao, "we found out your epic posts were overloading the mainframe sir"

Some engineer did a fantastic job of saving probably their entire team's jobs by pulling that whopper of a lie out of their ass

In this weekend's WSJ Review section: The onomatopoetic "tick-tock" for the sound of a clock goes back for centuries, but in the early '70s it became journalistic jargon for a blow-by-blow account. Now of course it's easily mistaken for "TikTok." https://on.wsj.com/3YHJc2q

Dans le cadre du « Cercle de Serendip » https://wp.unil.ch/serendip/ j’interviendrai le 28 février 2023 (17 h 15, UNIL, salle POL-303 « Le Forum ») :

L’incertitude historiographique, mal nécessaire, ou bien inévitable ?

Nos connaissances du passé sont toujours incomplètes. En général, ce fait est considéré comme un défi pour l’historiographie. Mais si c’était une chance ?

#UNIL #incertitude #historiographie #humanitésNumériques

The call for papers for @ACM #DocEng23 (August 22–25 in Limerick, Ireland) is out:

http://doceng.org/doceng2023/cfp

❝The 23rd ACM Symposium on Document Engineering (DocEng’23) seeks original research papers that focus on the design, implementation, development, management, use and evaluation of advanced systems where document and document collections play a key role. DocEng emphasizes innovative approaches to document engineering technology, use of documents and document collections in real world applications, novel principles, tools and processes that improve our ability to create, manage, maintain, share, and productively use these. In particular, DocEng 2023 seeks contributions in the area of collaborative work with documents. Attendees at this international forum have interests that span all aspects of document engineering and applications.❞

#DocEng #DocumentEngineering