AI & LLMs

Forget iPads. Give every student a Raspberry Pi. They need to become comfortable with command lines. We need more hackers, not consumers.

Just had a thought the other day. Why do our public representatives need to be real people? If their function is to listen to and balance the needs of their constituency, why not make a virtual representative, an algorithmic avatar?

It sounds silly, but consider how the primary issues we run into with representatives are issues of human fallibility, corruption, and trust. A virtual representative would impartially reflect aggregate constituent views.

There's already work being done on building Liquid Democracy processes atop blockchains. This may not be as sci fi crazy as it sounds. https://www.newscientist.com/article/mg23531424-500-bitcoin-tech-to-put-political-power-in-the-hands-of-voters/

Also, this would remove all of the nonsense about electing a public representative just based upon completely irrelevant and superficial traits, such as the way he or she looks or talks.

. . . and to sell you what you never knew you wanted. Of course they will allow bad actors to manipulate truth and dictators to manipulate a populace. A facile vision of social media private enterprise as a utopian fulfillment of democracy/freedom should never have been believed.

And it's ridiculous to expect them not to, while at the same time expecting them to provide you everything for free. Of course they will sell your info to the highest bidder. Of course they will leverage your weaknesses to make you addicted ...

We are facing a reckoning for the Faustian bargain we've made by expecting everything online to be free. You really think a private enterprise like Facebook is actually committed to world peace and connection? They'll sell your soul down the river to make money. . .

We want everything to be free, but won't acknowledge the compromise we make when we join. Maybe it's time to start expecting to pay for the protections we expect.

An underappreciated aspect of effectively using ChatGPT: It takes a well crafted prompt to generate a truly interesting response. Therefore, teach kids the knowledge and skills of crafting great questions for inquiry.

Just remember that when you hear about AI being "trained on data," it's often much of the same data you have been trained upon. If we want our algorithms to be better then us, we may want to restrict their diets.

There's a strong tendency to write off the power of #LLMs because they are 'just' parroting or predicting a sequence of words that most often come together. But that's what language is. Even stripped of cognition, language has a generative power that LLMs have tapped into.

This @emollick post made me think about how using an LLM as a tool for student reflection could be powerful, such as after completing a unit performance task.

More thoughts in 🧵⬇️

https://www.oneusefulthing.org/p/assigning-ai-seven-ways-of-using?publication_id=1180644&isFreemail=true

Wouldn't be surprised if someone out there has already designed some tool to do this, given the surge of innovation in this space right now. Or maybe it's already embedded in an existing LLM and I'm just not aware?

What would make it most useful for such a pedagogical purpose is if the teacher had a dashboard where they could get the LLM started with the same, well honed prompt, then assign the generative discussion activity to each student individually.

I've only been dabbling but I am convinced there are many powerful use cases for AI in the classroom--with the proper constraints in place, once we get past all the anxiety around cheating.

While LLMs may be great for boilerplate forms of writing, their most powerful use is dialogic--and this is where we should look to harness them with students in instructional use.

We think of LLMs inaccurately because we forget that written language is different than spoken language, and they are trained thus far solely on the written form

I have to pinch myself when using @AnthropicAI Claude2-- which is currently free to use. Uploading research papers and asking it questions about the paper -- including across multiple papers -- is an absolute game changer. https://claude.ai/ https://t.co/LvjGLF6POb

Don't think this gets the proper attention. We STILL need a strong base of literacy and knowledge in order to wield these cool new AI tools. Just like we did with Google search and all the other stuff before them. https://x.com/mandercorn/status/1751299301181985093

...if I do my close reading and annotating, and then share my compiled quotes with an LLM, I've found it quite useful for summarizing and synthesis of the key points and patterns I've drawn out. It can also then be more helpful for a basic outline of what I might want to write

I don't use AI tools very frequently in my own life or work yet, but beginning to discover specific uses that can make my blogging more efficient. I've found that I can't rely on LLMs to read and synthesize research if I upload PDFs directly. Too general and vague...But

In other words, I still need to do up front reading and analysis. But it helps me to more efficiently consolidate what I pull out on the back end.

AI, at least in current iterations, best supplements, not supplants, human knowledge and insight.

If the revolutionary advent of the technology of writing advanced our individual and collective capacity for processing and sharing information, in what ways will language models similarly and exponentially scale that capacity?

I am far less interested in the potential of AI for tutoring and far more interested in its potential for supporting planning of instruction or easing administrative tasks for teachers and leaders

The most exciting potential of LLMs may lie not in their ability to respond to and generate human language, but in their ability to help us to process and manage vast quantities of information --and thus further extend our cognitive capabilities. https://languageandliteracy.blog/scaling-our-capacity-for-processing-information

The effectiveness of #AI and #LLMs hinges on the quality of our input.

Just as with previous tools like Google Search, the more precise and informed our prompts, the more powerful and accurate their responses.

Another way of framing this idea:

LLMs can help us further widen or refine our own ideas and language. They are far less useful in just handing them to us.

They mirror and leverage what we provide to them.

I am less interested in the "intelligence" of #AI #LLMs, & more interested in their potential to teach us about our own development of #language and #literacy, and to extend our own cognitive, linguistic, and literary abilities. https://languageandliteracy.blog/the-interplay-of-language-cognition-and-llms-where-fuzziness-meets-precision

The meaning of a concept may not be determined by its individual vector components, but rather by its relationships and interactions with other concept vectors in a vector space. Hence: the advancing power of LLMs. https://x.com/MatthiasMichel_/status/1821575602937139299

There is great potential for #AI to extend our #cognitive capabilities Yet there is a strong case of an equal and commensurate danger that use of #LLMs can reduce our cognitive capabilities. https://write.as/manderson/reviewing-claims-ive-made-on-llms