0:00
Welcome back to the Titch's AI Cafe. I'm Kane, a teacher and a parent who wants AI to give us
0:04
back our time. I share practical classroom-tested AI strategies to cut down your workload and prove
0:10
your feedback and reduce stress so you can get your evenings back and be more presently a friend of
0:15
family. No jargon, just actual steps, grab your mug and let's get going. This week delivered one
0:22
of the biggest AI platform updates for the years so far, though it is only March being March the
0:27
nice today. The release of Chatchepity 5.4 comes out as a significant upgrade to what came out
0:33
previously. So the most interesting feature educators in this new is the new thinking plan. So when
0:40
the model tackles complex requests, it can outline its approach before producing the final answer.
0:45
I think this is really good. The old show you're thinking type arrangement. In particular,
0:50
terms this allows teachers to intervene early, for example, if you ask Chatchepity to design a
0:55
lesson sequence or summarize an historical document you set, you can adjust the structure before
1:00
it actually completes the response and fine tune it, saving you a bit more time. It doesn't mean
1:06
you've got to pay attention while it's doing. You can't just set and forget and then come back
1:09
and modify it, but like I said, the opportunity is there, particularly if you're working on
1:12
actively as it goes through. Another big improvement of the model is the ability to handle now
1:19
extremely long documents. With the context window approaching a million tokens, basically,
1:25
it's going to allow you to analyze very large research collections. Things write extended
1:30
curriculum documents in total units of work in a single prompt, whereas before I couldn't handle
1:35
that amount of characters. And for teachers work regularly with, I guess, lengthy syllabus materials,
1:42
you go a lot of content and you want to put things like academic articles, historical sources.
1:47
This could hopefully streamline your prep time because you're doing it all in a single prompt
1:51
or discussion at one time, which is an interesting thing. I was looking at people, a lot of people
1:55
jumped over the cloud for ethical reasons, but they're apparently using up their limits very
2:02
quickly in cloud compared to why Chatchepity works and how they calculate things. Interesting point.
2:08
The other thing that came out for Chatchepity is there's an L version for Excel, so instead of obviously
2:14
moving dropping spreadsheet data into a Chatbot, now you can put the inside Excel itself with
2:19
basically a plug-in to interpret the datasets formulas and explain patterns within the spreadsheet.
2:25
And I think that's actually really handy. I use Excel a lot to do analysis and often I do put it
2:30
in there so I can now do it in Excel, so I'm looking forward to that particular feature.
2:35
And this will just speed up my timeline, my process doing that. So once again,
2:40
competition between the different platforms is a tense file. And like I said, the ability to
2:44
now export or your chats and move to other platforms is really opening up those opportunities.
2:51
Anthropic, which is clause-making expanded memory capabilities in cloud,
2:56
even for its free use to act so that basically you can have a long-term conversation memory.
3:02
And cloud does a bit different because it'll often scan the entire chat each time it's
3:06
white people hitting their limits, if you got a very big chat. But this allows the system to remember
3:12
preferences and recurring information over time better than what other platforms actually do.
3:18
So in this case, good-make your AI assistant, gradually learn the context of your classroom
3:23
where you correct them and focus a lot faster and a lot easier than different platforms.
3:28
What's another development? GitHub Copilot, which is Copilot, and now includes Grox. If you
3:33
didn't know, GitHub is actually owned by Microsoft. But now you can put the Grox coding model
3:39
as part of the automatic model selection if you want to use that there. So this development,
3:43
I guess, increasingly students in computing classes using these tools now can interact with more
3:49
than one AI model within the same tool, giving them choice, therefore, depending on the projects,
3:54
which one gives you the best results. And creates opportunities in the classroom,
3:58
around discussing a different system, solve the same problem. And that's very relevant in all
4:03
systems, even AI. And before finishing up today's update, one unusual story from the research
4:10
community provides a reminder that AI adoption always carries risks. Some security researchers recently
4:16
demonstrated large-scale prompt injection attacks hidden in web content. Remember I spoke about
4:22
before we're using AI tools to actually go out onto the web. The idea of injection attacks is
4:29
they can actually inject malicious code actually into your program there through different websites.
4:35
So these attacks can manipulate AI tools. So wouldn't you click, say, summarize button or similar
4:40
features? The research has found that this data can actually be loaded into. So as always,
4:48
treat AI outputs critically. Always your expert opinion. Verivice your sources, just as you
4:55
would with any other research. But when you look at this, these incremental changes, like I said,
5:02
talking about Claude, people are moving for ethical reasons over to Claude. And that's
5:06
absolutely fine. There's lots of different AI platforms out there. And I'll just reiterate,
5:11
there's no one best model or best platform simply because there are ways developing new
5:18
updates, competition features, compared to the competitors to try and get that natural advantage.
5:24
Therefore, why one might be as good now next week and you release my command and it'll leap frog
5:30
the one you're thinking of shifting to. Well, like I said, find one you actually prefer,
5:34
what one one that actually does the job for you, stick to it, learn it the best that you can.
5:38
All the advice I give really is clickable to every single platform when I do my activities
5:43
because they, at the core of them, they do essentially do the same thing just in slightly
5:47
different ways. So until then, keep up the good work. If you want to come on and talk about your
5:52
journey, drop me an email. Till then, look after yourself.