When Claude Explains, then it doesn’t anymore!

I’m not sure when was it that Anthropic launched its Claude Explains sort of blog, but between June 3 and 5 every single technical website wrote something about it.

When I visited it in the evening of June 5, I noticed that all posts seemed to have been classified under “Implement code functionality.” Also, that it was structured in two major sections:

  • A top one, with posts displayed in sets of two, after which one had to repeatedly hit “See More” to add two more posts. The titles were written in that wooden language that didn’t make me click on any of them.
  • A bottom one, with posts displayed in sets of five, after which one had to repeatedly hit “See More” to add five more posts. They all were Python tutorials aimed at beginners and covering punctual topics. I thought them to be an interesting choice.

The problem I had with this approach is that these Python lessons (or whatever you call them) were utterly impractical: no paginated navigation, no full list, nothing! The best proof that a company able to develop a very decent AI agent can still lack individuals with common sense able to design a usable website.

I got so pissed off by this retarded design that I wasted 1304 credits on Manus AI to make it extract and index the links to the 130 Python tutorials as they were available on June 5! It even made a nice website with the links grouped in 12 categories.

Let me stress it again: only the links were extracted—the tutorials are still hosted by Anthropic.

It was only the next day that I noticed how unreliable Manus AI can be:

  • The 130 pages included a duplicate, so they were actually 129. Or maybe Manus “forgot” a distinct page and replaced it with an already existing one. Who could tell in what form an AI hallucination can manifest itself?
  • 32 of the 129 links (24.8%!) were broken! In most cases, the AI added articles (“a-”/“an-” or “the-”) or it pluralized nouns in the URL to match the respective titles! A few times, it even inserted several words. When the URL and the title weren’t consistent enough, Manus decided to “correct” the URL!

I then decided that, what the heck, why not try telling Manus to correct its shit? I asked it to check and correct the links (and the website), and the cost increased from 1304 to 2073 credits: 769 credits to fix what it broke itself!

Except that it only fixed 2 of the 32 broken links! Yeah, this is how much one can trust the AI these days…

While I was correcting the URLs myself, by hand (first by heuristically trying to figure out the real URL, then by using Google search into the site), and as I was 4 links to the end… 💥

Claude Explains disappeared before my eyes!

I quickly determined what happened. Anthropic has abruptly removed the access to the Claude Explains section of the website by modifying each hosted file at this time stamp:

<!-- Last Published: Fri Jun 06 2025 15:31:39 GMT+0000 (Coordinated Universal Time) -->

What they did was to insert this:

// Redirect all traffic to Anthropic
window.location.replace('https://www.anthropic.com');
</script>

All the files are, however, still on the server, and they can be retrieved, e.g., with wget. (At least, they were there when I last tried it.)

By the way, all the “Claude Explains” pages are dated “May 30,” but one shouldn’t trust them much: this page is labeled “7 Jun” but it was visible since June 6! Unless the displayed date is China’s or Japan’s!

When a zombie can live, it should live!

So what did I decide to do? I resolved to retrieve 129 files about Python for which I had the URLs; there might be more of them. Should Anthropic restore access to Claude Explains, everyone could access whatever else is there.

Then, I had to clean them up. Each page was about 9 times larger than it needed to be, and then it loaded JS and CSS and whatnot. The only good design decision was to generate a TOC for each page (at its left), but the relevant content was too narrow to my taste, so it needed scrolling, and scrolling, and scrolling…

The page template was atrocious, like it’s the case with most corporate websites, no matter how “clean” they pretend to be. As long as I decided to keep offline versions of these pages and to make something usable out of them, I needed to clean all the Anthropic-related garbage that polluted those pages.

That was extremely feasible without the help of any AI, with the following two exceptions:

  1. Given that I had to perform a “replace in all files” based on regex, I had to be reminded of patterns such as [\s\S]*? and (?=end_marker).
  2. To save some time, I asked a chatbot to create the CSS for me.

I’m rather satisfied with the result, which can be consulted online, and whose files can be downloaded from GitHub. A ZIP archive with everything takes about 855 KB!

🌐 Online: derludditus.github.io/ClaudePython

🌐 Repository: github.com/DerLudditus/ClaudePython

I cannot vouch for the absence of errors generated during the processing, in addition to whatever Claude and its human supervisors might have created.

What other dumb things has Anthropic done lately?

This is not the first shitty decision taken by Anthropic in the last few days. The other major recent news about Claude is this one: Anthropic has made the Claude 3.x models (Claude 3.5 Sonnet, 3.7 Sonnet, and 3.7 Sonnet Thinking) unavailable to Windsurf (Codeium)! And Claude 4 wasn’t available to Windsurf in the first place.

For the time being, GPT-4.1 is heavily discounted, and Gemini 2.5 Pro is mildly discounted.

Here’s Windsurf’s statement on the matter. Sure thing, should you pay upstream to the makers of your preferred AI, the BYOK (bring-your-own-key) mechanism allows you to continue using any and all Claude 3.x and 4 models.

TechCrunch:

Anthropic co-founder and Chief Science Officer Jared Kaplan said his company cut Windsurf’s direct access to Anthropic’s Claude AI models largely because of rumors and reports that OpenAI, its largest competitor, is acquiring the AI coding assistant.

“We really are just trying to enable our customers who are going to sustainably be working with us in the future,” said Kaplan during an onstage interview Thursday with TechCrunch at TC Sessions: AI 2025.

“I think it would be odd for us to be selling Claude to OpenAI,” Kaplan said.

The comment comes just a few weeks after Bloomberg reported that OpenAI was acquiring Windsurf for $3 billion.

Yeah, sure. It sounds to me like this:

PepsiCo, Inc. can no longer sell cans of Pepsi Max to the makers of a brand of mini-fridge because that brand will be acquired by Coca-Cola. Therefore, when customers buy such a fridge, they will have to purchase Pepsi cans by themselves—the subscription through the fridge maker is no longer possible.

That’s honest American business to you.

In other news, Reddit sued Anthropic for allegedly “scraping user comments to train chatbot Claude” (AP, Reuters, NYT, CNBC). I couldn’t care less, because I consider Stack Overflow more pertinent, but the comments on Reddit can indeed be valuable, too.

My short fuse hasn’t grown any longer

I cannot use AI products that misalign my chakras. I hate the tards and the scumbags. So far, I already ditched Perplexity on May 4. Starting right now, on May 7, I give up using Claude, too!

What’s left in the box? Let’s just talk browser and Android apps, but the same engines could be used via API calls too:

  • 🟢 ChatGPT is usable until it says something like this:
  • 🟢 Grok is usable unless it produces garbage, and until it tells me I have to wait, too (but it’s usually more generous than ChatGPT).
  • 🟢 I should use Copilot more frequently for short questions, as a replacement for Claude. I’m already becoming better friends with Copilot 😉
  • 🟡 I should try to value Mistral more, although it’s lagging a lot behind others. It even lacks the imagination to give me synonyms or rephrasings in English!
  • 🔵 I definitely should spend more time with Qwen3 and DeepSeek to discover the fields in which they might excel. Bar their cretinous censorships, they’re often pretty OK. For instance, Copilot simply refused to answer a question about the general elections in the United Kingdom (the right to vote for non-citizens who are however citizens of a Commonwealth country and permanent residents), whereas Qwen3 and DeepSeek gave correct answers. Copilot sometimes censors too much!
  • 🟠 Finally, I should probably resolve to use Gemini now and then. The design of its web interface is so atrocious that my dick gets smaller at the sight of those stupid fat letters: M (Menu), S (Search), A (Choose your model), E, G, H, S (Settings), A (Add files), N (Canvas: Create docs and apps), M (Microphone), not to mention how it looks like once it answered, with mingled letters and an extra one: H (Files in this chat). How retarded must one be to come with such a design?

Oh, and Manus, as promising as it looked like, is a pile of crap.