If you follow my streams, you'll know that I abuse DeepL (like for this page) for translating French texts into English and other languages, even when it wouldn't hurt to use my brain a little.
If you follow Embarcadero's online conferences or attended the TMS Training Days and Delphi Summit 2024, you'll know my level of spoken English.
I'm pretty good at video, even if it takes up a lot of my energy and time to cut up after a recording. I usually do two of them and mix the bits that seem to come across best. Orally, I still have trouble finding the right words and my pronunciation remains uncertain.
I had planned to really get back into it this year. My attempts with Duolingo and other mobile applications are clearly ineffective. I'll need to spend some time immersed in England before my next interventions so that I'm more or less satisfied with myself.
This is to say that I'm not giving up on the idea of creating content in English (or other languages) but I'm staying focused on French despite a smaller potential audience.
Why am I going to use AI?
I've just finished a training course on AI at Simplon as part of the European Time 2 Skill project, to get a more grounded view of what we can do with it apart from over-consuming precious resources (water, energy). I don't have any use cases for my day-to-day work, but I do have some for the sites, videos and software I edit.
Future changes for my blogs
I was already planning to publish some articles in other languages, but I'm taking far too long to think about how best to adapt my backoffice for this. I didn't want to overwhelm the feed aggregators (notably Begin End and Delphi Feeds) but why not?
Make a blog for each language or put everything in one place and let the browsers translate what I don't translate myself? It's an open question, so it's up to you to tell me what you prefer.
Webinars, courses and tutorials in several languages?
For videos, I've finally tried out HeyGen(*). These tests were conclusive. I took out the Team subscription. It allows me to translate videos of less than 30 minutes. That's enough for courses, tutorials and maybe “Le Temps d'une Tomate”, but not for reruns of presentations that systematically exceed 30 to 60 minutes.
So the question is: am I better off investing in subtitling existing videos and translating them, or in fully automating the translation of relevant videos (subtitles + voice)?
(Don't expect to find my live coding integrals on Twitch in other languages!)
I'm going to do a test with the Delphi Getting Started course available on Udemy(*): an automatic translation of the soundtrack while retaining my voice after correcting the transcription of the videos into French. This series will be available on Udemy in English, Italian, Spanish, Portuguese and German.
I'll also be uploading several videos automatically translated (without proofreading the transcription or translation) from French to English on YouTube. Here again, the question is whether to stay with my current channels or create a new one for each language. For visibility, the first option seems better. For user experience, I'd opt for the latter. It remains to be seen whether fighting with the recommendation algorithms is worth the cost.
What about streams and conferences?
I'm still undecided about the next time I speak in English.
I'll probably record the presentations in French, as I used to do from time to time, and test their automatic translation before redoing an English version if I'm not satisfied. The boss is not opposed. His always kind encouragement encourages me to make a human version whenever possible. I'll decide on a case-by-case basis, according to timing and subject matter.
For Twitch I stay in French.
HeyGen(*) offers virtual avatars for streaming. They've announced progress on this subject for next year. So it's not impossible that I'll be streaming in both French and English in a few months' time, at least for test purposes. In the meantime, I'll be looking for an automatic subtitling solution for the streams.
A chatbot or a spell?
In the series of short-term experiments, I've trained a chatbot on my Delphi-related sites (this blog, Serial Streameur and Delphi Books). It's on trial until December 24 (the length of the trial period for the tool used).
When I do searches on a classic search engine (Google via Ecosia, for example) I tend to align keywords, but over time I've started to formulate coherent sets that can resemble sentences. The results are less and less relevant, but I do find answers.
The advantage of an LLM in the form of a chatbot would be to have an analysis of requests to provide a more textual response than a set of links to rummage through. The disadvantage is that AIs can make things up as they don't understand the meaning of what is being asked. The other disadvantage is that answers are written instead of providing links to the pages used to answer.
These days, there's a tendency to be (too) content with what ChatGPT provides. This not only makes the work of the authors of the sites from which it draws its replies invisible, but also contributes to the consumption of precious resources.
The choice is between user comfort and respect for other people's work (or our own).
The results of the chatbot installed as a test on this page are pretty good on the questions I tested, and I'm wondering whether I'm not going to propose an equivalent solution in addition to or as a replacement for the search that currently goes through Qwant (or wait for Octave Klaba and his teams to finish redesigning this search engine and propose something around it).
Here again, I need your feedback on the subject.
Do some tests and tell me if you think it's worth spending time on this or if it's better for me to install a classic keyword search engine for all my sites (and maybe others linked to Delphi).
Time and money
As you know, all this takes time and costs money. If you like what I do online, the best way to help me keep going is to subscribe to Zone Abo.
If you're in a company or educational establishment, you can take advantage of yearly or lifetime offers for all users of the same domain name.
(*) Affiliate link.
