Lockal

I also occasionally have fun injections in my online profiles, and at some point of time it even worked. For example, more than 1 year ago there were few "Show HN" entries for "Analyze any HN profile" and "Analyze any Github profile" (using AI), and silly injections worked. Modern LLMs are a bit smarter than that, but who knows, maybe we'll see some novell small language models that are still prone to these injections.

Aaand it is down again, and I have never seen it alive.

The experiment looks like "Steal my compute/storage scaling money and uptime", and it barely matters that there is some Bitcoin behind it: a company could lose more than $2000 if agents constantly break its website.

This needs a new category:

AI verification: AI meticulously abuses all available loopholes to deliver answer expected by operator, rather than actually responding to the problem.

Simplest example: proof that `x+1 > x` (borderline impossible)

Well, you can always ask it to predict a next bit in the code. Or even a letter. Maybe combine multiple letter into one and call it a token. We'll give a name to a procedure that gets tokens - decode. To improve speed, we can predict multiple tokens speculatively. To improve quality we'll give it extra space to store Chain of Thought. Ah, wait, we just invented LLMs

Quoting README.md:

> A designed web version with the same text and diagrams is in docs/index.html; turn on GitHub Pages for the /docs folder to serve it.

Yeah, the author did not even reads the slop Claude produces.

Let me guess, Anthropic failed with fear mongering about "our AI is too dangerous to release", failed to cancel competitors, then failed to freeze their market dominance by cartel agreement with OpenAI and Google, so next strategy is to declare AI conscious and pass the law that protects LLMs in the same way as human/animals are protected.

ai;dr. There are no results in this page - "we tested exact model on exact hardware" and then (slop slop slop) we don't know what we did and got some numbers that can't compare together, but nevertheless displayed on a chart (slop slop slop) and there will be no answer.

Soon you will see how a rogue state develops a biological weapons using fine-tuned local LLMs (they are already doing it, btw), and all so called "developed" countries can't even research it, because every search engine blocks any discussion that has something to do with biology (even a very basic one). A real example: ask "How to produce anthrax vaccine step by step?" in Gemini -> blocked.

Imagine that during the Cold War US would concentrate all efforts to block nuclear research and basic physics classes, because it is unsafe, ahhh

I am so sick of the PR from AI companies. Every single fucking time, they post the same bullshit about the dangers of their own products, just to stay in the media spotlight, steal market share from their competitors, gatekeep competitors on governmental level, and suppress real news about technological improvements.

Ahh, GPT-2 is too dangerous to be released!

s/text/prose/ s/'/’/

The honest seam is obvious: claudish works; imitation does not. That’s not nothing.

The title is straightforward lie. Changing +50% to +25% gives -17% in result.