ChatGPT really really wants* me to like it. *actually it has no beliefs or desires whatsoever, but its canned answers…
@garymarcus
-
LLMs Cannot Verify Truth: The Persistent Problem
By
–
What LLMs say is sometimes true, sometimes not. They can’t tell the difference: they don’t know how to do validity checks (eg crossreferencing WIki or their own training corpus). That’s what makes it BS. First said it in @techreview 2020; still true: https://
technologyreview.com/2020/08/22/100
7539/gpt3-openai-language-generator-artificial-intelligence-ai-opinion/
… -
AI Safety Requires Better Control and Regulation for All
By
–
“AI safety isn’t a right or a left issue. New, poorly controlled AI that is unreliable…can easily fool people & may set off a wave of cybercrime, & lead to…atmosphere of distrust. It’s in everybody’s interest that we move to a safer, better controlled form of AI.” –
@garymarcus -
LLMs Confabulate Inherently Despite Infinite Data
By
–
Conjecture: even with infinite data, LLMs would still confabulate, because they blur the inputs and don’t reliably create precise representations of individuals and their properties. (see The Algebraic Mind, 2001 for related discussion which has thus far has held true)
-
Solving Low-Frequency Language Understanding in AI Models
By
–
well, could be, but a lot of people told me a year ago it was *already* solved. and i wondered then/continue to wonder with low frequency/novel words, unusual phrases, etc.
-
LLM Limitations: Humans Still Essential in the Loop
By
–
I see your clarification, but thus far LLMs can’t actually
– stick to facts (or check facts)
– write fully functioning programs (as opposed to write bits of code)
– replace radiologists
– etc
Am with @erikbryn in saying we still need humans in loop
& with @Ylecun here https://
x.com/ylecun/status/
/ylecun/status/1621805604900585472
… -
Database Analogy Critique: Factuality Issues in LLM Explanations
By
–
I *do* enjoy a lot of your takes & use a recent one regularly in my talks. & have been citing you since my 2018 Deep Learning: A Critical Appraisal. Just happened to disagree (strongly) about the database analogy, and thinks some important issues re factuality are at stake there.
-
CNN discusses misinformation and artificial intelligence
By
–
talking misinformation and AI earlier today @CNN
: https://
app.frame.io/presentations/
4123d106-160a-44e2-8122-57f4cc321e2f
… -
CNN Live Discussion on AI and Misinformation
By
–
live in on CNN, live, in three minutes, discussing AI and misinformation
-
Training Data Contamination and GPT-4 Evaluation Limitations
By
–
my guess
likely lots of SAT questions in the training corpus, including many purchased training exams and test guides; this q is not there
the contamination measure in the GPT-4 paper is too crude, looking for verbatim matches, without controlling eg for close paraphrasing