ConferenceCall 2024 02 28: Difference between revisions
Ontolog Forum
No edit summary |
|||
| Line 21: | Line 21: | ||
== Agenda == | == Agenda == | ||
* 12:00 - 12:05 Introduction '''[[RaviSharma|Ravi Sharma]]''' | * 12:00 - 12:05 Introduction '''[[RaviSharma|Ravi Sharma]]''' [https://bit.ly/49BPE11 Video] | ||
* 12:05 - 12:35 '''Gary Marcus''' | * 12:05 - 12:35 '''Gary Marcus''' [https://bit.ly/42Y5T5X Video] [https://bit.ly/3SXW3MR Discussion] | ||
* 12:35 - 13:05 '''[[JohnSowa|John Sowa]]''' ''Without Ontology, LLMs are clueless'' | * 12:35 - 13:05 '''[[JohnSowa|John Sowa]]''' ''Without Ontology, LLMs are clueless'' [https://bit.ly/434EMGy Video] | ||
** Abstract: Large Language Models (LLMs) are a powerful technology for processing natural languages. But the results are sometimes good and sometimes disastrous. The methods are excellent for translation, useful for search, but unreliable in generating new combinations. Any results found or generated by LLMs are abductions (hypotheses) that must be tested by deduction. An ontology of the subject matter is necessary for the test. With a good ontology, errors, hallucinations, and deliberate lies can be detected and avoided. | ** Abstract: Large Language Models (LLMs) are a powerful technology for processing natural languages. But the results are sometimes good and sometimes disastrous. The methods are excellent for translation, useful for search, but unreliable in generating new combinations. Any results found or generated by LLMs are abductions (hypotheses) that must be tested by deduction. An ontology of the subject matter is necessary for the test. With a good ontology, errors, hallucinations, and deliberate lies can be detected and avoided. | ||
** [https://bit.ly/3uHQv10 Slides] | ** [https://bit.ly/3uHQv10 Slides] | ||
Revision as of 04:32, 3 March 2024
| Session | Foundations and Architectures |
|---|---|
| Duration | 1.5 hour |
| Date/Time | 28 Feb 2024 17:00 GMT |
| 9:00am PST/12:00pm EST | |
| 5:00pm GMT/6:00pm CET | |
| Convener | Ravi Sharma |
Ontology Summit 2024 Foundations and Architectures
Agenda
- 12:00 - 12:05 Introduction Ravi Sharma Video
- 12:05 - 12:35 Gary Marcus Video Discussion
- 12:35 - 13:05 John Sowa Without Ontology, LLMs are clueless Video
- Abstract: Large Language Models (LLMs) are a powerful technology for processing natural languages. But the results are sometimes good and sometimes disastrous. The methods are excellent for translation, useful for search, but unreliable in generating new combinations. Any results found or generated by LLMs are abductions (hypotheses) that must be tested by deduction. An ontology of the subject matter is necessary for the test. With a good ontology, errors, hallucinations, and deliberate lies can be detected and avoided.
- Slides
- 13:05 - 13:30 Discussion
Conference Call Information
- Date: Wednesday, 28 February 2024
- Start Time: 9:00am PST / 12:00pm EST / 6:00pm CET / 5:00pm GMT / 1700 UTC
- ref: World Clock
- Expected Call Duration: 1.5 hour
- Video Conference URL: https://bit.ly/48lM0Ik
- Conference ID: 876 3045 3240
- Passcode: 464312
The unabbreviated URL is: https://us02web.zoom.us/j/87630453240?pwd=YVYvZHRpelVqSkM5QlJ4aGJrbmZzQT09
Participants
Discussion
Questions for Gary Marcus
12:46:10 From bhaugh : Can you recommend any specific approaches to combining symbolic and neural AI?
12:49:48 From ToddSchneider : Where does context fit into the ‘landscape’?
12:50:48 From ToddSchneider : Where does 2nd (or higher) order cognition or reasoning come into play?
12:52:47 From Gary Berg-Cross : Are you familiar with the Sci American article: "How AI Knows Things No One Told It" (MAY 11, 2023 BY GEORGE MUSSER) and if so what did you think of its approach to its claims back then? https://www.scientificamerican.com/article/how-ai-knows-things-no-one-told-it/
Questions and Comments for John Sowa
13:14:40 From Janet Singer : John addresses the language pragmatics as well as semantics. How much of a mistake was it to assume a ‘semantic web’ could be successfully developed without the context of the pragmatics?
13:23:13 From ToddSchneider : Should images be considered ‘symbols’?
13:32:16 From Janet Singer : @ToddSchneider Don’t want to answer for John but diagrams could be icons as opposed to symbols which need rules of a language for interpretation/use (after Peirce)
13:26:29 From Phil Jackson : what about the mental models of people who are blind from birth? are these models based on 2-d images or instead based on 3-d models of spatial volumes, or 4-d models of spatial volumes moving in time? Or some other kind of model, that also includes sounds?
13:33:23 From Gary Marcus : Sorry I have to go! Thanks for having me! and I am very sympathetic to most of what John said!
13:34:42 From Gary Berg-Cross : A friendly "central executive" needed to align with some of "our" values gets us into the vague area of social concepts.
13:39:27 From Marco Neumann : Verisimilitude seems to be the predominant goal for the current generative AI effort. And in our media mediated world this may be a sufficient threshold after all. But if AGI is a future goal, how would you test it? Isn't the Turing test just a test of believability as well?
Resources
Previous Meetings
| Session | |
|---|---|
| ConferenceCall 2024 02 21 | Overview |
| ConferenceCall 2023 11 15 | Synthesis |
| ConferenceCall 2023 11 08 | Broader thoughts |
| ... further results | |
Next Meetings
| Session | |
|---|---|
| ConferenceCall 2024 03 06 | LLMs, Ontologies and KGs |
| ConferenceCall 2024 03 13 | LLMs, Ontologies and KGs |
| ConferenceCall 2024 03 20 | Foundations and Architectures |
| ... further results | |