MicroGPT Lets You Peek With Your Browser

February 21, 2026 by Al Williams 1 Comment

Regardless of what you think of GPT and the associated AI hype, you have to admit that it is probably here to stay, at least in some form. But how, exactly, does it work? Well, MicroGPT will show you a very stripped-down model in your browser. But it isn’t just another chatbot, it exposes all of its internal computations as it works.

The whole thing, of course, is highly simplified since you don’t want billions of parameters in your browser’s user interface. There is a tutorial, and we’d suggest starting with that. The output resembles names by understanding things like common starting letters and consonant-vowel alternation.

At the start of the tutorial, the GPT spits out random characters. Then you click the train button. You’ll see a step counter go towards 500, and the loss drops as the model learns. After 500 or so passes, the results are somewhat less random. You can click on any block in the right pane to see an explanation of how it works and its current state. You can also adjust parameters such as the number of layers and other settings.

Of course, the more training you do, the better the results, but you might also want to adjust the parameters to see how things get better or worse. The main page also proposes questions such as “What does a cell in the weight heatmap mean?” If you open the question, you’ll see the answer.

Overall, this is a great study aid. If you want a deeper dive than the normal hand-waving about how GPTs work, we still like the paper from [Stephen Wolfram], which is detailed enough to be worth reading, but not so detailed that you have to commit a few years to studying it.

We’ve seen a fairly complex GPT in a spreadsheet, if that is better for you.

Microsoft Uses Plagiarized AI Slop Flowchart To Explain How Git Works

February 19, 2026 by Maya Posch 43 Comments

It’s becoming somewhat of a theme that machine-generated content – whether it’s code, text or graphics – keeps pushing people to their limits, mostly by how such ‘AI slop’ is generally of outrageously poor quality, but as in the case of [Vincent Driessen] there’s also a clear copyright infringement angle involved. Recently he found that Microsoft had bastardized a Git explainer graphic which he had in 2010 painstakingly made by hand, with someone at Microsoft slapping it on a Microsoft Learn explainer article pertaining to GitHub.

As noted in a PC Gamer article on this clear faux pas, Microsoft has since quietly removed the graphic and replaced it with something possibly less AI slop, but with zero comment, and so far no response to a request for comment by PC Gamer. Of course, The Internet Archive always remembers.

What’s probably most vexing is that the ripped-off diagram isn’t even particularly good, as it has all the hallmarks of AI slop graphics: from the nonsensical arrows that got added or modified, to heavily mutilated text including changing ‘Time’ to ‘Tim’ and ‘continuously merged’ into ‘continvuocly morged’. This makes it obvious that whoever put the graphic on the Microsoft Learn page either didn’t bother to check, or that no human was involved in generating said page.

Continue reading “Microsoft Uses Plagiarized AI Slop Flowchart To Explain How Git Works” →

The Requirements Of AI

February 19, 2026 by Al Williams 72 Comments

The media is full of breathless reports that AI can now code and human programmers are going to be put out to pasture. We aren’t convinced. In fact, we think the “AI revolution” is just a natural evolution that we’ve seen before. Consider, for example, radios. Early on, if you wanted to have a radio, you had to build it. You may have even had to fabricate some or all of the parts. Even today, winding custom coils for a radio isn’t that unusual.

But radios became more common. You can buy the parts you need. You can even buy entire radios on an IC. You can go to the store and buy a radio that is probably better than anything you’d cobble together yourself. Even with store-bought equipment, tuning a ham radio used to be a technically challenging task. Now, you punch a few numbers in on a keypad.

The Human Element

What this misses, though, is that there’s still a human somewhere in the process. Just not as many. Someone has to design that IC. Someone has to conceive of it to start with. We doubt, say, the ENIAC or EDSAC was hand-wired by its designers. They figured out what they wanted, and an army of technicians probably did the work. Few, if any, of them could have envisoned the machine, but they can build it.

Does that make the designers less? No. If you write your code with a C compiler, should assembly programmers look down on you as inferior? Of course, they probably do, but should they?

If you have ever done any programming for most parts of the government and certain large companies, you probably know that system engineering is extremely important in those environments. An architect or system engineer collects requirements that have very formal meanings. Those requirements are decomposed through several levels. At the end, any competent programmer should be able to write code to meet the requirements. The requirements also provide a good way to test the end product.

Continue reading “The Requirements Of AI” →

Living In The (LLM) Past

February 9, 2026 by Al Williams 23 Comments

In the early days of AI, a common example program was the hexapawn game. This extremely simplified version of a chess program learned to play with your help. When the computer made a bad move, you’d punish it. However, people quickly realized they could punish good moves to ensure they always won against the computer. Large language models (LLMs) seem to know “everything,” but everything is whatever happens to be on the Internet, seahorse emojis and all. That got [Hayk Grigorian] thinking, so he built TimeCapsule LLM to have AI with only historical data.

Sure, you could tell a modern chatbot to pretend it was in, say, 1875 London and answer accordingly. However, you have to remember that chatbots are statistical in nature, so they could easily slip in modern knowledge. Since TimeCapsule only knows data from 1875 and earlier, it will be happy to tell you that travel to the moon is impossible, for example. If you ask a traditional LLM to roleplay, it will often hint at things you know to be true, but would not have been known by anyone of that particular time period.

Chatting with ChatGPT and telling it that it was a person living in Glasgow in 1200 limited its knowledge somewhat. Yet it was also able to hint about North America and the existence of the atom. Granted, the Norse apparently found North America around the year 1000, and Democritus wrote about indivisible matter in the fifth century. But that knowledge would not have been widespread among common people in the year 1200. Training on period texts would surely give a better representation of a historical person.

The model uses texts from 1800 to 1875 published in London. In total, there is about 90 GB of text files in the training corpus. Is this practical? There is academic interest in recreating period-accurate models to study history. Some also see it as a way to track both biases of the period and contrast them with biases found in data today. Of course, unlike the Internet, surviving documents from the 1800s are less likely to have trivialities in them, so it isn’t clear just how accurate a model like this would be for that sort of purpose.

Instead of reading the news, LLMs can write it. Just remember that the statistical nature of LLMs makes them easy to manipulate during training, too.

Featured Art: Royal Courts of Justice in London about 1870, Public Domain

[Yang-Hui He] Presents To The Royal Institution About AI And Mathematics

February 2, 2026 by John Elliot V 2 Comments

Over on YouTube you can see [Yang-Hui He] present to The Royal Institution about Mathematics: The rise of the machines.

In this one hour presentation [Yang-Hui He] explains how AI is driving progress in pure mathematics. He says that right now AI is poised to change the very nature of how mathematics is done. He is part of a community of hundreds of mathematicians pursuing the use of AI for research purposes.

[Yang-Hui He] traces the genesis of the term “artificial intelligence” to a research proposal from J. McCarthy, M.L. Minsky, N. Rochester, and C.E. Shannon dated August 31, 1955. He says that his mantra has become: connectivism leads to emergence, and goes on to explain what he means by that, then follows with universal approximation theorems.

He goes on to enumerate some of the key moments in AI: Descartes’s bête-machine, 1617; Lovelace’s speculation, 1842; Turing test, 1949; Dartmouth conference, 1956; Rosenblatt’s Perceptron, 1957; Hopfield’s network, 1982; Hinton’s Boltzmann machine, 1984; IBM’s Deep Blue, 1997; and DeepMind’s AlphaGo, 2012.

He continues with some navel-gazing about what is mathematics, and what is artificial intelligence. He considers how we do mathematics as bottom-up, top-down, or meta-mathematics. He mentions about one of his earliest papers on the subject Machine-learning the string landscape (PDF) and his books The Calabi–Yau Landscape: From Geometry, to Physics, to Machine Learning and Machine Learning in Pure Mathematics and Theoretical Physics.

He goes on to explain about Mathlib and the Xena Project. He discusses Machine-Assisted Proof by Terence Tao (PDF) and goes on to talk more about the history of mathematics and particularly experimental mathematics. All in all a very interesting talk, if you can find a spare hour!

In conclusion: Has AI solved any major open conjecture? No. Is AI beginning to help to advance mathematical discovery? Yes. Has AI changed the speaker’s day-to-day research routine? Yes and no.

If you’re interested in more fun math articles be sure to check out Digital Paint Mixing Has Been Greatly Improved With 1930s Math and Painted Over But Not Forgotten: Restoring Lost Paintings With Radiation And Mathematics.

Continue reading “[Yang-Hui He] Presents To The Royal Institution About AI And Mathematics” →

The CURL Project Drops Bug Bounties Due To AI Slop

January 26, 2026 by Maya Posch 44 Comments

Over the past years, the author of the cURL project, [Daniel Stenberg], has repeatedly complained about the increasingly poor quality of bug reports filed due to LLM chatbot-induced confabulations, also known as ‘AI slop’. This has now led the project to suspend its bug bounty program starting February 1, 2026.

Examples of such slop are provided by [Daniel] in a GitHub gist, which covers a wide range of very intimidating-looking vulnerabilities and seemingly clear exploits. Except that none of them are vulnerabilities when actually examined by a knowledgeable developer. Each is a lengthy word salad that an LLM churned out in seconds, yet which takes a human significantly longer to parse before dealing with the typical diatribe from the submitter.

Although there are undoubtedly still valid reports coming in, the truth of the matter is that the ease with which bogus reports can be generated by anyone who has access to an LLM chatbot and some spare time has completely flooded the bug bounty system and is overwhelming the very human developers who have to dig through the proverbial midden to find that one diamond ring.

We have mentioned before how troubled bounty programs are for open source, and how projects like Mesa have already had to fight off AI slop incidents from people with zero understanding of software development.

LLM-Generated Newspaper Provides Ultimate In Niche Publications

January 26, 2026 by Tyler August 18 Comments

If you’re reading this, you probably have some fondness for human-crafted language. After all, you’ve taken the time to navigate to Hackaday and read this, rather than ask your favoured LLM to trawl the web and summarize what it finds for you. Perhaps you have no such pro-biological bias, and you just don’t know how to set up the stochastic parrot feed. If that’s the case, buckle up, because [Rafael Ben-Ari] has an article on how you can replace us with a suite of LLM agents.

The AI-focused paper has a more serious aesthetic, but it’s still seriously retro.

He actually has two: a tech news feed, focused on the AI industry, and a retrocomputing paper based on SimCity 2000’s internal newspaper. Everything in both those papers is AI-generated; specifically, he’s using opencode to manage a whole dogpen of AI agents that serve as both reporters and editors, each in their own little sandbox.

Using opencode like this lets him vary the model by agent, potentially handing some tasks to small, locally-run models to save tokens for the more computationally-intensive tasks. It also allows each task to be assigned to a different model if so desired. With the right prompting, you could produce a niche publication with exactly the topics that interest you, and none of the ones that don’t. In theory, you could take this toolkit — the implementation of which [Rafael] has shared on GitHub — to replace your daily dose of Hackaday, but we really hope you don’t. We’d miss you.

That’s news covered, and we’ve already seen the weather reported by “AI”— now we just need an automatically-written sports section and some AI-generated funny papers. That’d be the whole newspaper. If only you could trust it.

Story via reddit.