You shouldn’t play 20 Questions with an LLM. Why? Because LLM behavior less resembles a simulation of a person than a superposition of many such simulacra. In @Nature
, Shanahan, McDonell, and Reynolds eloquently illustrate what this means:
RESEARCH
-
Don’t play 20 Questions with LLMs due to superposition of simulacra
By
–
-
Mathematics of Machine Learning Summer School Video
By
–
(15) 2019 Mathematics of Machine Learning Summer School – YouTube https://
bit.ly/3Q94q6S #AI #MachineLearning #DeepLearning #LLMs #DataScience -
GPT-3 retroactively labeled GPT-3.5 after text-davinci-003
By
–
It was, at the time, called GPT-3. The label “GPT-3.5” was applied retroactively after text-davinci-003 was released. text-davinci-002 gets the question wrong, text-davinci-003 and subsequent 3.5-series models get it right.
-
Bostrom’s Vulnerable World Hypothesis: Technology Risk Framework
By
–
The Vulnerable World Hypothesis, Nick Bostrom : https://
nickbostrom.com/papers/vulnera
ble.pdf
… -
AI Risk Analysis: Comparative Doom Probability Assessment
By
–
You merely need to figure out that
P( doom | AI ) < P( doom | no AI ) -
French AI Research Ecosystem Growth Through Education
By
–
I also taught a course at Collège de France in 2016.
It helped a lot of French students to get started in deep learning. FAIR-Paris has been hosting 20 to 30 resident PhD students, graduating 5 to 10 per year since 2018.
They have seeded the French ecosystem of AI research, -
AI Tools With Safety Flaws Should Not Drive Society
By
–
If I built a car & said, “this vehicle was built using information that may be inaccurate regarding physics, safety, or environmental protections” I wouldn’t let it drive you or your family. Yet lots of people creating these tools want them to drive society knowing these errors.
-
AI Tool Disclosure Standards: Beyond Mere Warnings
By
–
If I published a paper that says, “this paper may have inaccurate information about people, places or facts” the response from reviewers would be, “keep doing your homework & provide reliable resources.” Disclosure can’t be “this tool is crap but I told you.”
-
ChatGPT LLMs reliability concerns for factual accuracy
By
–
Recognizing ChatGPT & LLMs are helping people in their work, & I appreciate end users feeling helped, this first point of disclosure below, screen capped 15 minutes ago, makes no response trustworthy. Responses can be interesting/useful like gaming scenarios. But not facts.
-
NLP Podcast: CEO Discusses LLMs Strategy and Straive Partnership
By
–
My first #NLP podcast with Anand S , CEO & Co-Founder of Gramener. He shared his views & experience with LLMs and also explained the strategic reasons behind Gramener's decision to say "Yes" to Straive… #datascience #ai #chatgpt #llms https://
youtube.com/watch?v=X0exQy
ppcb0
…