LLMs are models, and as we all know, all models are wrong, but some are useful.
2 Likes
To my mind, itâs also important that it also connects âelephantâ to your experience of actual elephants. In other words, at least most thinking seems to be deeply connected to our bodies and experience.
An LLM does not know the meaning of the word âelephantâ in the same way. If it is going to achieve the kind of intelligence humans have, then this lack of real experience must not matter. I donât see how this is true in general. It may be true for certain kinds of thinking; it is true for playing chess and other kinds of games, for example.
Each human brain is constantly recalibrating and learning. Once an LLM is trained it stops learning. Itâs a very different thing from us.
There are many studies that show that emotions are an aid in the strength of these sorts of connections. The stronger the emotion, the more âdetailsâ associated with these connections and the faster (and more autonomous) the recall.
And that will be where LLMâs/AI will never do as well as a human, IMO.
1 Like
Similar to above litmus tests, like Tower of Hanoi, or basic math questions, I think you really need to probe why the bot answers these questions the way it does. It could be:
-It is incapable of understanding
-It is incapable of thinking
-It misunderstands something specific
-It has no common sense
-It is giving a incorrect knee jerk-response, that would be corrected with a second thought.
-It is assuming that the question is intended as a game or joke, and going along with it.
And forgetting. That seem to be a critical piece that hasnât been highlighted. Though we can constantly learn, we only have so much capacity and energy available to maintain that capacity, and so have evolved strategies to cull off âunimportantâ historical data.
1 Like
I like that option. If someone comes up to me at a bar, pub etc. and says something nonsensical I think to myself âThis person is either away with the fairies or making a jokeâ. Either way, I play along with it.
1 Like
Yep, I tend to feel the same way. Our brains evolved for a physical world, they get direct inputs from physical experiences. Thatâs why âcommon senseâ things seem to stump LLMs so far.
Although, for âelephantâ, Iâve never touched one. If I hadnât gone to zoos and circuses, I wouldnât even have first person visual experience with elephants.
I might use stacking blocks, and getting my fingers caught in doors, and dropping a glass as things the I know from first hand experience.
Well, almost all of us. If it was a person it would be said to have anterograde amnesia.
1 Like
UC - Berkeley Law School just banned the use of AI amongst students.
I wonder how many more schools will follow suit.
While I can assume the understandable rationale, the âstudents may not upload course materialsâŚinto generative AI systemsâ constraint seems potentially excessive given some of the more reasonable/valuable uses of AI in the wild, and the need to train future workers in such.
Itâd be better if students constrained to doing such things within the schoolâs own instance of GPT.
I know a philosophy PhD student here whoâs been working at a coffee shopâŚ
Second kid was on the way and he ended up getting some sort of job at the provincial legislature doing something with legislation.
So would it be prohibited to ask AI to provide me a summary of our understanding of topic X and provide me with a bibliography of the key works on this topic?
Eta: if so, thatâs stupid.
Given the AI companies abuse of copyright material, it seems like their rule is to protect the IP of the departmentâs faculty.
ETA: Iâve faced a bit of a similar dilemma where Iâve been asked to prepare a summary of a workshop I organized. I was tempted to upload the presentersâ presentations to copilot to generate a summary and realized I shouldnât as I donât know what the AI is going to do with their content and where itâll reappear.
Here is their argument:
Future lawyers may need to use artificial intelligence (âAIâ) fluently. But the current state of the technology requires that AI use be coupled with the cognitive skills necessary to strategically deploy the technology, to critically assess its work product, and to uphold ethical obligations to clients and to the legal system. In short, thinking remains the sine qua non of good lawyering (and of a quality legal education). This policy seeks to ensure that our courses focus on requisite cognitive skills by default. It provides students with the opportunity to develop the skills they need to conceptualize, outline, draft, revise, and edit their work by forbidding the use of AI for these purposes in connection with work submitted for credit. It also forbids using AI to translate work for credit, thus providing students with the opportunity to develop and exercise their own fluency with legal English. And it prohibits AI use for any purpose in any exam situation.
I think the TLDR of it is that LLMs are completely wrecking education right now and the schools need to declare war.
2 Likes
We can probably draw an analogy with developing techniques to teach math after electronic calculators became widely available. This hard even after knowing how calculators are used in practice. The way frontier, commercial LLMs will be used in practice is still very much unknown.
As for uploading course material: this is a major advantage LLMs have over a simple google search (or similar) in retrieving information. So this rule may be intended in part to keep requiring the same kind of âthinkingâ in research.
1 Like
Papal encyclical on AI (et. al.) is out.
why is it the universityâs role to train future workers? Shouldnât that be the responsibility of the employer?