Nathan Bos, Ph.D. – Medium

Nathan Bos, Ph.D.

Nathan Bos, Ph.D.
in
Towards Data Science

Language Models and Spatial Reasoning: What’s Good, What Is Still Terrible, and What Is Improving

A review of capabilities as of July 2024

3d ago

Scorecard. Still Terrible: Box Folding, Abstract Map Navigation, Drawing diagrams. Improving: Stacking objects, Real Map Navigation, Venn Diagrams, Spatial Reasoning in Photos.

3d ago

Nathan Bos, Ph.D.

Do I have to be polite to my LLM?

It won’t change results much, but it might be good for you

Mar 31

Do I have to be polite to my LLM?

Mar 31

Nathan Bos, Ph.D.

What your GPT RAG can’t see

Blind spots give insight into the mind of an LLM

Jan 15

What your GPT RAG can’t see

Jan 15

Nathan Bos, Ph.D.

Will a language model tell me when I’m being lied to?

You’ve recently taken in a teenaged relative. She’s failing in school and says she needs a better study environment. You seek helpful…

Nov 1, 2023

Will a language model tell me when I’m being lied to?

Nov 1, 2023

Nathan Bos, Ph.D.

Language models can explain comedy but can’t write it

Could AI write a good scene for Community or The Office?

Aug 30, 2023

Language models can explain comedy but can’t write it

Aug 30, 2023

Nathan Bos, Ph.D.

Can a language models understand human taste?

Probing GPT and Bard’s judgment about food flavors, textures and preparation. Special bonus recipe for chicken-fried lemon sorbet.

May 29, 2023

Can a language models understand human taste?

May 29, 2023

Nathan Bos, Ph.D.

What does GPT really think about humans?

GPT filled out Big-5 personality tests for some people I made up. The results surprised me. It makes some inferences that I didn’t think it…

May 16, 2023

What does GPT really think about humans?

May 16, 2023

Nathan Bos, Ph.D.

GPT-4 may have general intelligence, but don’t let it near your Easter eggs.

Wherein I test GPT-4’s stacking abilities, find some flaws, and some weird misconceptions.

Apr 9, 2023

GPT-4 may have general intelligence, but don’t let it near your Easter eggs.

Apr 9, 2023

Nathan Bos, Ph.D.

Prompting better ‘Theory of Mind’ performance from Large Language models

And, trying to help GPT-3 from being so gullible

Mar 28, 2023

Variant,Prompt,Result
Transparent Bag,And Why,fail
Transparent Bag,Why First,fail
Transparent Bag,Restate + Why,success
Trusted Testimony,And Why,success
Trusted Testimony,Why First,success
Trusted Testimony, Restate + Why,success
Uninformative,And Why,fail
Uninformative,Why First,fail
Uninformative,Restate + Why,success
Late Label,And Why,fail
Late Label,Restate + Why,fail
Late Label,Get a Clue,success, but at the cost of GPT’s innocence

Mar 28, 2023

Nathan Bos, Ph.D.

Nathan Bos, Ph.D.

Ph.D. Psychology, data scientist, LLM enthusiast. Interests: human-centered artificial intelligence, cognition, language, humor, applied ethics.

Following

Help
Status
About
Careers
Press
Blog
Privacy
Terms
Text to speech
Teams