#genml — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #genml, aggregated by home.social.
-
I have a couple of contexts in which I need to provide formal input on the use of #AI for #science. I've written the following and would welcome suggestions if I have misstated or misrepresented any aspects of the #genML craze.
1/5
Machine learning (ML) and artificial intelligence (AI) are variably defined overlapping terms that are generally applied to software applications that deliver information outputs that go beyond data captured in prior databases or knowledge systems. The term AI is currently normally applied to applications that simulate the behaviour and responses of human agents.
ML approaches are a significant tool of great importance both for science and applied use. Successful applications have exploited diverse technologies including semantic datasets, process models and, above all, neural networks to estimate unknown values for variables of interest. Such applications include, among others, weather forecasting, future climate prediction, text recognition, audio processing, and image classification (including processing of biomedical imagery or species identification tools). Common features of such applications include well-curated source datasets that include known values for the variable of interest associated with relevant co-variables; prediction or estimation of output values (temperature, species name, etc.) that include significantly less information than the inputs; and well-defined truth criteria that allow the success rate of the application to be quantified. The information produced by such models, when combined with success measurements, is in effect a probability for a given value in a given context based on co-variables and on the training data.
Other successful AI/ML applications such as protein-folding models and computer chess may have very different internal models that have been developed to target a specific domain but still have well-defined and measurable success criteria (proximity of modeled protein structures to actual molecules, victory in chess matches).
-
I understand it's probably AI bots randomly stringing together words that connote cuteness and getting AI tools to generate AI video clips that they can use to generate AI clickbait ads (and I have no idea where this link on the Independent actually went), but wouldn't you want to cut the video before your "robot rabbit" starts morphing into a cat?
-
Oh, FFS - I hadn't realised they've done that. I lost respect for #TheGuardian generally because of their pro-establishment campaigning against Corbyn. I have been making a monthly contribution (not subscription) directed to #TheGuardianAU, but partnering with #OpenAI and buying into #genML hype is a red line for me.
On #TheGuardian and #OpenAI, here's a little from #Crikey:
https://www.crikey.com.au/2025/02/20/guardian-artificial-intelligence-openai-news-corp/
Sorry, @VoordeMus - are the lay-offs separate or explicitly linked?
-
EPFL's Center for Digital Trust conference on disinformation, elections and AI opens with a keynote by @sisiwei of the @themarkup.
(https://c4dt.epfl.ch/c4dt-conference-on-disinformation-elections-and-ai/)
1/🧵