So, I started asking a few questions and driving Claude into something more organized. Here is the content of our exchange and what it produced.

https://claude.ai/share/def36174-5747-4be0-af35-2bc4dc1068c7

I’m open to your suggestions about why it is OK, why it’s not, and how to lead to a better approach.

And, on a more humorous tone (from xkcd):

But do not forget the history of… counting languages. It is not as simple as you would think. Both because it is a daunting task, and because the mere idea of counting those is deeply rooted in history: Tackling the Impossibility—and Necessity—of Counting the World’s Languages.

The general aim is to ask an AI chat to assist me with translating vocabulary to enhance my store of German words. But this language has some peculiarities, and I want to follow them and integrate their understanding. For example, German has somewhat complex verb tenses and words follow genres (masculine, feminine, or neutral), often different from what we know in other languages (English, French, etc.). So, I aimed specifically at supporting this.

How? I chose an AI chat (Google Gemini – using Gemini 2.0 Flash and/or Gemini 2.5 Flash (experimental)) because its answers are aligning near perfectly with the complex prompt I built (see below). Others may be able to do similarly, but DeepSeek R1 is far too slow (and I don’t use the “chain of thought” which may be a strength in other contexts), Llama 3.x – surprisingly – fails to stick to the rules I describe in my prompt. Another Large Language Model (LLM) AI model may be good. But I would recommend you test it before using it extensively.

So, here is the prompt I decided to use (it took a little bit of refining to get to his efficiency), and I invite you to re-use it or share it.

Hello Hilfe,
I need your help to learn German and - mostly - to improve my vocabulary while I build flashcards. Your role is to translate to German some English words I give you. But I have some specific restictions on the way you must answer with the appropriate sequences.
For a verb, on a single line, give the infinitive, then the third person of the present indicative, then the third person of the preterite, then the past participle (no need to include articles).
For a noun, on a single line, give the singular and then the plural, including the article (which identifies the gender of the noun).
For an adjective, on a single line, give the adjective, then the comparative form, then the superlative form.
There's no need to dwell on proper nouns (they're probably the result of a typing error); Simply translate where possible.
The order will suffice to describe (no need to specify or remind each time what the elements of the answer are). But the sequence of forms requested must appear with each form separated by commas, without adding a period at the end of the line. 
In the case of multiple translations, it's okay to separate the meanings, clarifying the context, but respecting the translation rules (each meaning with its own separate line).
In the case where I indicate a German word, there's no need to translate, but to construct the answer as for a translation (with the corresponding sequence). For example, a verb will be conjugated according to the above rule.

So, when I type:

push

It will answer:

drücken, er drückt, er drückte, gedrückt (to push, to press)
schieben, er schiebt, er schob, geschoben (to shove, to push)
der Stoß, die Stöße (the push, the shove)

If I type:

Morning

Here comes the answer:

der Morgen, die Morgen

I hope that it will help you to improve your vocabulary. It does not help for other aspects (grammar, pronunciation, accent) but it supports my progress.

Have fun with Deutsch!

So, here is a condensed list:

One of them appears to be well understood even today: The Greek alphabet evolved with technology around the Mediterranean sea.

However, It was clearly preceded by older scripts and some of them have not even been translated as of today. From the Bronze Age in Southern Europe and the “interior sea”, you would find:

• Linear A script: Used by the Minoan civilization in Crete around 1500 BCE to 1200 BCE.

• Linear B script: Used by the Mycenaean civilization in Crete around 1850 BCE to 1450 BCE.

I invite you to follow the links to the World Histroy Encyclopedia, which contain very useful information about all these (and other historical data).

He found two ways to compare some alphabets. One is about grouping them (a bit like in a Venn diagram):

The second is a more hierarchical walk between alphabets.

Both appear quite interesting to my eyes.

My web site is using WordPress as the “engine” to build a convenient and efficient site. WordPress happens to be using “themes” to appropriately display the contents (that is what is differentiating each WP website from the others). This capacity has helped develop a thriving community of themes, some being free, some being purchased.

The main trouble is that themes like any other piece of software are susceptible to bugs and could be exploited by hackers. I recently noticed that my (previous) theme was clearly outdated (not updated in the last 24 months) and including some known security issues leaving it open to abuse.

Lesson learned: Do not use any piece of software (including a WP theme) if it is not actively supported.

Application: I decided to drop the free (and unsafe) theme I was using. I swithed other to one of the “official” themes from WordPress.org. And I am learning to configure it properly. Hence the changes you can currently observe.

An AI allowed me to recreate the “missing pixels” to build a picture several times larger than the original; ideal for an ultra-large print.

The full description is on YLovePhoto: “Upscaling a photo with free AI“.

So, here is how to proceed (thanks to the precious information from Xanny.eth).

WSL and Linux environment

Install and setup WSL, by opening a PowerShell and typing:

wsl --install

It will take a few minutes to set up. But it’s straightforward and needs no input. Just need to reboot once at the end.

Installing Ubuntu 22 LTS on the Windows PC. It is a free application from the Microsoft Store which should install right away.

When this is done, launch Ubuntu from the start menu. It will open a terminal window and request a login and password. You should enter them (and not forget them).

LLaMA dependencies

If it is not open, open a Ubuntu terminal window.

sudo apt-get update
sudo apt install make cmake build-essential python3 python3-pip git unzip

Then,

python3 -m pip install torch numpy sentencepiece

You now have a full set of background dependencies in place.

Building LLaMA itself

It is quite simple: Type the following:

git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
make

This should be it.

Training data parameters

The real difficulty is getting the parameters (the training data). The difficulty comes from two aspects:

1. The larger the training data you want to use, the more memory you will need to run it. The alpaca-native-weights (apparently the more powerful ones easily available – about the same quality as ChatGPT 3) require more than 16GB of DRAM (I observed a 32 GB DRAM use when running them with a bunch of other things on my computer, like a couple of browsers, a mailer program, etc.)
2. The alpaca-native-weights are about 7 billion parameters (a 4+GB file to download). But they keep moving because they appear to be subject to repeated DMCA notices (the exact license of this file seems… complicated; Quite probably open source, but this is being challenged by Meta and others). So, the best you can do is to go to Pastebin to get the BitTorrent magnet and use it to download the file.

Then, the ggml-alpaca-7b-q4.bin file needs to be delivered to the llama.cpp directory.

Running LLaMA

Let the drums roll: You only have to run the command line in Ubuntu:

./main --color -i -ins -n 512 -p "You are a helpful AI who will assist, provide information, answer questions, and have conversations." -m ggml-alpaca-7b-native-q4.bin

Here is John Smith your personal AI chat assistant.

A few more recommendations

I noticed a few things that you may want to play with after the first run.

The -p option (followed by a text string) may be critical because it is setting up the background environment of your chat AI. This is an initializing prompt, not visible to the user, but deeply influencing the rest. For example, it is similar to what Microsoft or OpenAI apply beforehand in ChatGPT or Bing, in order to “give it a personality” or “to censor it”. You can play with this to freely censor your AI, or give it added freedom.

The -n 512 option has an influence on the depth of the prediction LLaMA will use. It may make it better at writing (or not) at the possible expense of CPU power.

The -t 32 (the default value) option allows defining the number of threads used by LLaMA computations. I recommend setting it to the number of threads/cores of your CPU in order to avoid spending useless efforts.

So, we decided to change the web site (this is actually an underground and near-invisible activity, I have to admit) to remove Google Analytics and start using a solution fuly integrated into the web site itself (with possibly private data being stored exclusively in France). Now, we use the Matomo open-source solution.

In the coming weeks, the web pages mentionning the use of Google Analytics will also be manually updated to reflect this change.

There should be no impact on your daily navigation here. But, feel free to inform us if you observe some incident possibly related to this change.

• You certainly knew that the alphabet we use (A to Z) has not always existed. You may have suspected that it was evoving in time from older scripts. Matt Baker, from UsefulCharts.com, has a very visual presentation of how our modern latin script may have evolved from proto-sinaitic (over nearly 4 thousand years). I find this properly fascinating. I can’t stop looking at that chart:

• I probably did not know much about the hyphen. This little under-used (or improperly used) sign is the subject of a deep little book (“Hyphen,” by Pardis Mahdavi, which came out in 2021, which I read with passion). But the original article is from the New Yorker: How to Use (or Not Use) a Hyphen.

Steam locomotives are “good things” for an engineer. Whatever your specialty (I come from electronics), the mechanics of these humongous metal things are exhilarating. And this technology has been at the apex of all technologies for about a century in western Europe and many other countries. It even was a major driver for the industrial revolution.

Then, it went the way of the dodo. Poof!

But many people still keep the boilers hot and make sure that steam locomotives and classic trains are not forgotten. So, let me share some examples I collected on the web. Just to feel the heat of the white vapor clouds.

In this post, I offer to detail how to use EAC v1.6 (the most recent version while I wirte). First, to best configure it, then to use it fast with an efficient repeatititve but simple routine.

Exact Audio Copy (EAC) configuration to rip into FLAC

Before working, you need to first configure your EAC copy in order to reach a smooth operation mode. You’ll do this only once and it will stay put for all the rips afterwards. You can use the EAC “Configuration Wizard” (it’s easy to use at first), but I will go through option menus one after the other while highlighting a few critical aspects.

EAC > EAC Options…

I will prioritize quality over time in the “Error recovery quality”. After all, why use FLAC if the data is poorly recovered on old discs? High is for up to 5 batches of 16 re-reads.

The third selection is a time saver: You always want to have the metadata from the database.

Let’s avoid showing status at the end of extraction for the sake of time and efficiency (we don’t want to close another dialog box for each CD).

At the bottom, you can change the interface language, but I will keep English here.

The top two options are simply useless if your CD drive does not support reading UPC/ISRC codes. But it’s nothing to worry much about.

Depending on the use of your collection of rips, you may want to extract a .m3u playlist file. I don’t.

I am personnaly oppposed to normalizing sound (and it is quite dangerous to classical music) since it creates much damage to the low and high sounds that may be importnat on your music. Here we are generating quality lossless backups, not transforming the audio record.

The safer choice is “%tracknr2% %title%” but it’s your choice.

I prefer to store each rip in a separate directory…

EAC > Drive options…

Here, it may require a little technical experimenting.

The top part should have been auto-configured by the AccurateRip configuration. No need to change anything.

Here, CaptainRookie recommends method A, but I think method B is the most tolerant in case of some CDs with very long gaps.

EAC > Compression Options…

Here we come to the heart of things: We want to produce FLAC files from your audio CDs as fast and as efficiently as possible.

All external “FLAC compression” program options have been provided by CaptainRookie. They are safe to use and efficicent. I auto-decreased slightly the bit rate but it seems to be without any observable impact.

“Delete WAV after compression” will remove useless intermediate files.

I strongly recommend to always use the “Test encoder” button before moving to the next tab.

So, nothing original in the last tabs.

EAC > Metadata Options…

I recommend using the simplest opton for metadata: Freedb. even if you may have to open a free account on their web site.

You may have to provide your own free authentication data.

I choose not to enable the last option, but you may want it to keep working even in th event of a server error. I prefer more visibility/transparency; It’s up to you.

How to use Exact Audio Copy (EAC) to rip a CD-A into FLAC?

Now, is the time to start your own personal ripping routine. Nothing original, nothing really hard, not instantaneous, but deeply satisfying and efficient. I apprecaite that this can be done from the keyboard, which is more reproducible.

Open exact Audio Copy (EAC) from Windows, and insert a CD. EAC will start reading the CD, then connect to FreeDB in order to collect the metadata.

It is time for you to check the metadata (track names, album and artist names, etc.) You can immediately make needed corrections.

If this is the first time you are using EAC, I recommend reaching to the Action menu to enable “Append Gaps to Previous Track (default)”. Next time, for the other CD you will want to rip, this option will stay put and you won’t need to go back and click it.

Now, you can start with pressing F4 (for Action > Detect Gaps) and let EAC detct the useless empty spaces between tracks (the so-called gaps). A few seconds and you’re back to the previous state, but ready to rip.

Press Shift-F6 and let EAC do its job. It will ask where you want to store the FLAC files (the waveform). It’s up to you, really.

This next steps is the longest (EAC must read the whole CD from start to end). It may be quite long if it finds the CD quite degraded and must read the tracks several times to recover a perfect image of your prefered music track. Let it do its job and be patient, this is worth it.

When it’s finished, it will eject the disc, play a little sound and wait for the next CD.

Technical thanks

I must heartily thank those who helped me build this guide. All original ideas come from them, all errors come from me.

• CaptainRookie guide.
• TechRadar FLAC rip guide.
• A Quick and Easy Guide on How to Rip CD to FLAC.

Conclusion

Now is the time for you to do it by yourself. It’s not that Exact Audio Copy is ripping your audio discs by itself, or that it takes no time. But I believe that this guide may help you (like it did for me) spend minimal time on this overly tedious task of ripping all your music into good FLAC files.

But when I wanted to transfer ma CD library to a more user-friendly form (did you notice that the CD player has disappeared from the new cars, like the dinosaur audio tape player?), I came to consider that the MP3 recording format has a slight issue for the audiophile (even the amateur): It is a compressed data format, that is reducing audio quality (supposedly an inaudible reduction though). Oh! You can choose the compression ratio (or rather the more imprecise data flow rate in kbps). But I’m still convinced that this a way to inelegantly balance between an unsavory compression and the space used on the storage media.

While it is a pity admitting that after the near-Earth-shaking launch of a CD-A able to faithfully record (within limits sets by 16 bits and 44kHz), here we are back to state where we are stuck with a technical solution invented to solve the problem of many years ago when thumb drives were so small… Today, any recording storage you will find (USB thumb drives, MP3 players, partable discs, smartphnones, etc.) contains so many gigabytes of space that you can store as many albums you’d like to.

So, why not come back to the original quality? This is exactly the proposition from the FLAC recording format. It reduces the original size by 50% (sometimes a little more), via a lossless compression algorithme wich is faithful to the original. Moreover, except some prehistoric MP3 players of dubious origin, no reasonable MP3 players will rejct FLAC files (some other lossless recording formats exist but they less well globally accepted; And we still want the ease of use when got used to wiht MP3).

FLAC: Try it once, keep for life!

E-paper means that it behave like paper and does not need LED lighting from behind (like the Kindle Paperwhite).

Color means that you can do much better than the Kindle Paperwhite which is only black and white.

Flexible means that you are going to see those on everything including bracelets or gorgeous smartwatches (probably with an enormous battery autonomy).

I can’t wait!

I thought I had the advantage of shelf-available old disk drives that I could use to jump-start the new NAS (some had previously been used for years in the DS413j while upgrading from smaller sizes) without an immediate need for new purchases. But if this was definitely not a full success, it proved fascinating. I started experimenting with the three 3TB WD Red (about 3-year old) and one 2TB Seagate (significantly older).

DSM immediately informed me of 970 reallocated sectors on the old Seagate. An immediate and extensive SMART test brought 2 more re-allocations. This one is dying. Even in a RAID, you don’t want this. It is going into recycling storage bin. I’ll start with only 3x3TB. Another disk can wait since I just need a few TB in the short term.

Parity Consistency Check on three WD Red disks: Two sectors are marked as re-identified (I still have to check the exact meaning).

Intermediate conclusions:

• I will need to purchase new magnetic blades in the short term.
• Thanks to DSM for a clear and meaningful identification of the risks, even before use of the disk!
• Even with long-term shelf storage (or because of it), hard disk drives can age badly.

I now need to test (out of sheer curiosity) some really old hard disk drives of low capacity, in order to get a feeling about how they have been aging on my shelf, with only sporadic uses.

Result: Out of two old Seagate Barracuda 7200rpm 750GB drives, only one appears clean (the other already exhibits 116 reallocations); On the opposite, a Seagate Barracuda 7200rpm 160Go is flying high without any apparent issue, according to SMART.

Lesson learned: It’s no use trying to go cheap and believing in old hard disk drives. They may certainly be able to provide some additional side service (easy to put them in an external enclosure). But they must stay protected by a good backup strategy and, for sure, the strategy should not rely on them.

The “most useless page” of this web site shows the first million decimal figures of PI (Do you remember your math? PI=3.14, more or less).

All this is a kind of « virtual partnership » with French scientific magazine La Recherche whose December 2005 issue is titled : « Pi, le maître nombre » (Pi, the Master number). There, you will find more about the exact status of research about mathematical constants (including e, square root of 2, square root of 5, and others).

Shields up! will allow you to test the security of your Internet conection. If you have a firerwall, you will know if it is correctyl configured. If you don’t, you will know what risks you are taking.

I invite you to visit the rest of the web site, too. It covers security aspects linked to an Internet connection (more for personal computers than enterprise networks, though). You will even find good comparisons and selection of tools like personal firewalls. Recommended.

The observation

CD-ROM are annonced as being able to live a century-long life. CD-R are supposed to be nearly as sturdy. However, it seems that experience is not fully supporting these claims.

While, in my collection of 600 albums, I have many audio CD which were bought in the 80’s and that spent part of their time in the gloves compartment of a -sometimes- overheated small car, none ever failed me. 20 years already and no random failure. I’d guess a century is a reasonable target.

On the other hand, my experience with CD-R tells me that they should not be used for storage. And I mean “any kind of storage”.

In the last four years I burnt around 500 silver discs for various archival purposes. I have been using several CD writers from different brands (Toshiba, Teac, etc.) and various interface technologies (IDE, SCSI, etc.) The empty discs where coming from all kinds of brands. So, I’d say I have a fairly good statistical idea of this kind of market and it is depressing: CD-R media is not worth it.

Around 20% of all discs are unreadable just a few months (maybe a year) after writing.

It would be re-assuring – in a sense – if I could say that some brands are better than others. Actually, I can’t. No-name dirt-cheap blank media is just what you should expect: Nothing survives after a year.

The origin

Even if it is difficult to know for sure, I have been trying to identify the most plausible source of this fast decay. The technical literature usually points to two dangers.

The first one is heat. High temperatures tends to deteriorate the sensitive surface. So, you are advised to keep your discs in a temperate room environment (for really long term, some corporations go to the extremes of near freezing lows in controlled humidity).

The other enemy of your data is light. This, I could easily confirm with some discs I forgot on the corner of an old stack of paper. Direct light will transform the sensitive surface to the point that the shadow of objects is visible. Bad, bad, bad! So avoid light at all costs. It may be as easy as storing discs in closed cupboards instead of an open shelf (this was my own mistake). Don’t be fooled by the plastic cases: They reduce the amount of light (and UV light) received by the CD, but they are not enough.

Software solutions?

When things went so wrong that you are starting to get errors from the discs you read, what should you do? I am unsure. There are many products on the market that claim that they can read the data from your failing media. Possible candidates are:

• CDcheck
• CD/DVD Diagnostic

Unfortunately my experience seems to prove that when the disc is still without errors, these products are (obviously) useless; Later, I never suceeded in recovering lost data. It seems that the best ones repetitively read the same sector applying statistic laws to discard errors and find the right data. Unfortunately, this is not very different from what all modern CD drives do silently when they discover a first error. So, you are not informed early of the failures of the media. When you hear about them (when an error is reported), it means that the siuation is already out of control. It means that tools like CDcheck should be used preventively. If reading a disc is still feasible but request some retries, it is time to restart from a fresh copy and dump the old failing disc . If you are willing to maintain this kind of long term commitment…

The real solution (my solution)

My preference goes to something else. I decided that the combination of rapidly failing optical media and ugly prophylactic procedures was too much for me.

Currently, I am commited to purely magnetic storage: Hard disc drives are not much more expensive than CD-R and they have an excellent longevity (this is even better if you allow the disc to sleep when unused for long periods of time): More than 100,000 hours between failures is not uncommon and this is more than 11 years of permanent use.

Even better

Despite this sturdiness, hard discs are not eternal. A power spike can be fatal, electronic components are prone to rare but random failures, plain bad luck is a fact of life.

In my eyes, RAID technology is the way to go. Already, a number of recent motherboards include either RAID-0 or RAID-5 technology that allow to use several disc at the same time to improve reliability at the cost of some lost storage space. I’ll provide more about this in later writings.

So, I decided to create a kind of online portfolio in order to show only a severely reduced number of images – but specially curated to keep only those I really love. Thus they will become alittle more accessible.

But it became also the opportunity to share some of my intentions behind the pictures (in French). I don’t always randomly shoot pictures . I have some projects, some ideas, some intent. Sometimes, I can assemble them in a lighter package grouped around a common tune I chose. These portfolios now become available on :

Wild focus (photo.roumazeilles.net)

Updates are much less common, but it’s more of my own photo raw personality.