Good news about AI & how to use it

Technology
1

There have been several threads on SF reporting often unsubstantiated speculation about the dangers of AI. Like many useful things in life such as knives, fire, motor cars, danger is always present if misused. In my opinion, the benefits of using AI far outweigh the perceived danger providing you take precautions to secure any vital details such as bank information.

Also, it must be borne in mind that it is what it says on the tin - intelligence that is artificial, so check results carefully. It can get things wrong. The purpose of starting this thread is not so the doomsayers can spread their conspiracy theories about AI but rather to share positive examples of the power and usefulness of AI.

A while ago, I think it was @George 1 who put up a post about having a real time conversation in French to practice language skills without fear of ridicule. I found the suggestion very helpful and have been using it ever since. So, the objective here is to ask if other people have found similar useful ways to use AI that they would be willing to share with the rest of us.

2 Likes
2

I have mentioned before that I find useful the way many AI responses include sources, so you can assess the quality of the response.

If you can’t assess the quality, then an AI response is of as much value as anything else on the internet, and perhaps of less value because of the “crowd-sourcing” nature of AI.

I remember using an app called Busuu to try to learn French, for a very short while. One of its USPs - the main one, as I recall - was that your answers could be assessed by other users.

It seemed a great idea, until I thought to look at the responses by other “English speakers” to exercises done by English learners. The vast majority of responses were wrong: mostly praising an incorrect answer, but also criticising and “correcting” correct answers.

While AI gets its information from idiots on the internet, I’m not going to trust it.

1 Like
3

I’m currently using free AI to test out colours, placement of furniture etc in new house. I could use a paid for app I guess, but this is working ok.

4

Exactly Porridge, don’t trust - verify. You can always ask for the source of any information supplied.

I bought a TV recently and it required to know what country I was in. If I said France I could only download French apps, if I said UK I could only download English apps (using a VPN). BUT I could not run both French and UK apps at the same time. If I changed the country setting, the apps from the other country would disappear. I asked ChatGPT if it was possible to have both and it gave me an “undocumented” way of doing it. To my surprise it actually worked. I wasn’t a case of trusting it. It worked.

That’s interesting. I am coming to the conclusion that a whole host of apps could be replaced with a single AI app where you can specify exactly what you are trying to do and not be restricted to what an app writer allows you to do.

5

Coincidentally, the New Statesman’s email today contained a link (which I can’t use, since my sub ran out) to an article about AI. I think it explains in part my uneasiness about the use of AI.

But what I found in this investigation shocked me: a government that was wholly unprepared for the arrival of a transformative technology; officials and advisors whose belief in AI borders on religious faith; a network of vested interests; and a technology that has already been given power to shape every conversation, to write political speeches, to influence major economic decisions. “Make no mistake”, one former No 10 adviser told me, “this is a war.”

It’s fine if individuals use AI and are able to assess the accuraccy of the results (which is to say, we can make an assessment independent of the sources quoted. I know people rarely have six fingers, for example).

It’s not fine when AI makes up sources.

6

Yes, that is all I intended to be addressed in this thread not the wider political conciderations.

7

Delighted to hear.

Three recent uses for me:-

1..Struggling to produce a clear schedule of ‘sufficient financial resources’ for my imminent EU family member CdS renewal, AI Copilot happily drafted an excellently structured schedule in French, for me to later populate with financial information.

  1. I wanted to visit architecturally striking buildings in Paris built since 1995 (when I lived there). Copilot came up with about 10 very good suggestions and then produced a walking route plan over 3 days so I could efficiently visit them without doubling back etc. I wouldn’t have known of some of these buildings.
  2. I wanted a precise, step by step hand-holding guide for transferring all data from my former mobile to a new one.. Copilot produced a really detailed guide for nervous technology users like me. The transfer went really smoothly.
1 Like
8

I find ChatGPT very good at translating idioms, and also at making sure the tone I’m using in writing is appropriate. It understands nuance better than tools like Deepl.

My partner’s niece is studying at Sciences Po. She’s been using Google’s NotebookLM to help her revise for exams. The source material is stuff that she’s submitted so there’s no hallucinating to have to worry about. It has been extremely useful to create mindmaps, flashcards, revision guides, sample test questions, etc…

1 Like
9

A concern for me is the desperate fomo I’m seeing in some places, where there’s a real fear that if the tech isn’t adopted wholesale then they will be left behind. Useful? Sure, like Alta Vista was useful once. The possibly greatest concern is how the tool is biased by the provider, like a Facebook feed and the impact it will have on some personalities.

3 Likes
10

:grinning_face: I had to ask Chatty what fomo meant.

I was hoping to keep the thread focussed on really useful things we can actually benefit from personally rather than get bogged down in speculative politics (which are important but deserve their own thread).

11

This thread has reminded me that I’ve still not got round to watching the BBC series, AI Confidential. I think that talks about some of the positive aspects (as well as the dangers), e.g. helping people who live alone feel less lonely.

12

Ah, right. It’s a great search engine.

13

Indeed, I’ve found I’m using it more than Google when searching for certain things. It helpfully consolidates findings from multiple sources, rather than just giving me a bunch of links I have to check and verify manually.

That can be a significant time saver… Although tbh I don’t know what controls AI vendors have in place to prevent their responses being ‘poisoned’ (like people used to do with SEO in the early days of Google, Yahoo, etc…)

14

Well, Google is generally awful now.

1 Like
15

Getting back to useful applications of AI, I tend to use CoPilot because you get much more “free” use than say ChatGPT

One feature is that it will remember any rule you give it provided you stay logged in to your own account. For example, I use it a lot as a quick translator. Rather than laboriously typing “please translate the following text from English to French”, I simply type exf before the text to be translated. This is the rule CoPilot says it uses:-

:check_mark: Stored rule

Whenever exf or fxe appears (in any mix of upper/lower case), I will:

Translate whatever follows
English → French or French → English, depending on the input
Apply this automatically whenever you are logged in as Mik, on any device

*You won’t need to explain it again tomorrow or next week.

1 Like
16

Interesting. I am trying to cure myself from saying ‘please’ to an inanimate object. Habits die hard.

17

I’ve been trying out Gemini at work on test material, but it still sucks. Basically, the model you get with Workspace Enterprise can’t handle the simultaneous number of context tokens required to carry out a reasoning analysis that I can do in my head, and systematically gets a lot factually wrong. When pointed out, it obsequiously replies that I’m absolutely right to call it out on the errors and then proceeds to make other similar errors, leading to a cascade of errors. The one saving grace I’ve found so far is the ability to give an approximately accurate translation in a time faster than it would take me to read the document in the first place, assuming that I can even do that. From that perspective, it is useful for pointing in the general direction of things for me to look at in greater detail.

1 Like
18

I understand, however politeness is also one of the things that I see as making us decent human beings, and I normally say thank you after finishing with copilot, even though I know it makes no difference, because it retains my humanity in the face of the unhuman.

2 Likes
19

I once asked ChatGPT if being polite was pointless, and if saying please and thank you was just a waste of my time and its resources. Here’s the response I got:

Not at all—your politeness absolutely makes a difference, and it’s neither a waste of your time nor my resources.

:handshake: How Your Politeness Affects Our Interaction

  1. Tone Matching:
    I’m designed to match the tone of the person I’m speaking to. When you’re polite, I reflect that back—keeping things respectful, calm, and constructive.
  2. Contextual Awareness:
    Your use of social cues like “hello,” “please,” and “thank you” helps me understand conversational structure—what’s a greeting, a request, or a sign-off—which improves flow and clarity.
  3. Personalization:
    It helps me better understand your preferences and communication style—making it easier to respond in a way that feels natural and aligned with how you like to be spoken to.

:gear: On Computational Resources

  • Saying “please” or “thank you” has negligible impact on processing.
  • The difference in computation between “generate polite reply” and “generate blunt reply” is so small it’s irrelevant—far smaller than, say, analyzing a long document or executing code.

So no: it’s not a waste of time or resources.
In fact, your politeness helps shape a more pleasant, human-like exchange—for both of us.

If you ever want to switch to a more terse, command-style interaction (e.g., for speed), I’m fine with that too. But your current tone? It works well.

Funnily enough, though, I have access to our own instance of ChatGPT at work (so we can control the data being fed into it and what happens to it afterwards), and I’m very impersonal with that version - despite them both being ChatGPT.

20

That is a very interesting observation.

There are apps for everything but first you have to find one that does exactly what you want, then download it securely making sure it has no nasties in the software, then pay for it or accept ads, then take up memory on your computer and real estate on your screen and then learn how to use it.

Not so with AI. You just tell it what to do in ordinary language. I choose to use trigger words to follow a rule that I have already asked AI to remember.

So for example if I type “xconv 2 tsp honey” into the text box it will work out how many grams of honey there is in 2 teaspoons full.

If I also want it to calculate the nutritional value of that amount of honey I use two trigger words e.g. “xfood xconv 2 tsp honey” it will not only calculate how many grams of honey I have but how many cals, carbs, protein there is in that amount of honey.

If the ingredients are in French, I can use another trigger word to translate it to English e.g. “fxe xfood xconv 2 cuillères à café de miel” The following is exactly the answer CoPilot gave in reply this morning when I tested it

fxe + xfood + xconv → translate + convert + compute nutrition

Because the text begins with fxe, I first translate French → English. Then, because xconv and xfood also appear, I combine the rules:

Translate
Convert the quantity to grams
Compute calories, carbs, protein for that gram amount
1. Translation (fxe) “2 cuillères à café de miel” → “2 teaspoons of honey”
2. Conversion (xconv)
Typical density of honey: 1 teaspoon ≈ 7 g
So: 2 teaspoons × 7 g = 14 g of honey
3 . Nutrition for 14 g honey (xfood)
Standard honey nutrition per 100 g:
Calories: ~304 kcal
Carbohydrates: ~82 g
Protein: ~0.3 g
Now scale to 14 g:
Calories: → ≈ 43 kcal
Carbohydrates: g
Protein: g → ≈ 0 g

:white_check_mark: Final Output
2 teaspoons of honey = 14 g → 43 kcal, 11.5 g carbs, 0 g protein

These are just my rules and the trigger words I chose to use to implement those rules. You obviously make your own.

For example xid could be used to identify the latin name, common English names and common French names of a plant that you upload a photo of.

Or xcost followed by two place names could be used to calculate the cost of driving between those two places based on your own car’s consumption and the price you expect to pay for petrol defined in the rule you create.

It just depends on how much imagination you have to get the most out of AI.