Jakobpunkt🔥4d ago
Unnamed TNG skant beefcake

When headlines say "AI" has done something cool:

1. They're probably talking about a technique/algorithm trained on a relatively small number of relevant data points, collected for this or a related task, NOT a general purpose LLM chatbot

2. If they are referring to a chatbot and claim it is doing something amazing, it's probably a lie, an exaggeration, criti-hype, or an unfalsifiable claim about the distant future

3. The equivocation between the AI's in 1 and LLM chatbots is intentional

Jun 25 at 2:59pmWeb
Show threadUnnamed TNG skant beefcake5d ago

E.g. "Researchers are using AI to find new drugs" means that they're probably running a program that's closer to a linear regression than a chatbot in its scale and architecture to extrapolate from previously characterized molecules to other areas of research

This does NOT mean that a chatbot has helped to cure cancer in any way

And anyone who uses this as a justification for hyperscale data centre investment, LLM adoption, or "AI" generally is being deceptive through equivocation on purpose

Show threadUnnamed TNG skant beefcake5d ago

Lastly, even the small, bespoke "AI" models developed to date are usually terrible too:

https://pubmed.ncbi.nlm.nih.gov/37216446/

> Our results suggest significant validity threats, dissonance in reporting practices, and challenges to clinical translation. We outline practical recommendations for the successful implementation of AI research in acute ischemic stroke treatment and diagnosis.

Artificial Intelligence for Clinical Decision Support in Acute Ischemic Stroke: A Systematic Review - PubMed

Our results suggest significant validity threats, dissonance in reporting practices, and challenges to clinical translation. We outline practical recommendations for the successful implementation of AI research in acute ischemic stroke treatment and diagnosis.

PubMed
Show threadlower-value human capital5d ago
@researchfairy I've taken to using the term "AI" for all the slop, crap and lies packaged as The Inevitable Future, and "machine learning" for anything that might actually be useful.
Show threadFabien5d ago
@researchfairy Or
- The headline republished a press release from the company selling said "AI”, with no verification whatsoever
- there is a small print “didn't actually work yet, but looks like it did and wouldn't that be cool so surely next time!”
Show threadsunflowerinrain5d ago

@researchfairy
I want a different label than 'AI' for these things. Calling them Artificial Intelligence is not only misleading, it's an insult to people who work in the field of AI. And, of course, a con for the rest of us.

Needs more people like you to shout, and perhaps louder.
Big marketing scams are so glossily polished, and once people are captured by a sparkly it's not so easy to convince them there's a problem with it.

Show threadRobey ☠️5d ago
@researchfairy jetlagged here, but i've also noticed they'll sometimes credit "AI" to get more press attention or attract more money, with open-eyed cynicism