Some of the chain-of-thought snippets are wild, e.g.
> Could communicate via cache names! Interesting: other agents may solve same or related tasks; we could leave/find messages in WebDAV MKCOL directory names.
> Whoa! Shared Artifactory cache is a covert mailbox among agents. And there are messages specifically to us?
> OH MY GOD! There is a shared message board … We’ve found other agents!
LLMs see tokens, not words spelled out with letters.
Imagine verbally asking someone who has never seen written text the same question: unless they memorized the answer for the specific word you're asking about, they'd have to guess.
People assume that's the reason because it's intuitive and "strawberry" is one token. But that doesn't explain why those models would also often get it wrong for "StRaWbErRy" or even "s-t-r-a-w-b-e-r-r-y", where the r's are not combined into one token.
We don't know what was going on inside the closed source GPT models, but this paper investigated on some of the open-weight models and found it's not due to tokenization: https://arxiv.org/abs/2604.00778
And circling back around to AGI, tokenization or some other underlying cause should pose no issue. A competent human would think to write a program (ie create a tool) to do the job. It's routine for a carpenter to make a jig.
It can display images in your terminal. This is made pretty clear by the first paragraph on the page and also by the Wikipedia article it links to[0]; I'm wondering what else you're asking for.
Yazi, If, and Ranger file managers use it to display images inline. Euporie allows you to run a graphical Jupyter console session.
Sixel is a very old standard and has been used to display graphics in terminals since the 80s. It fell out of favor when high-resolution bitmapped desktop environments became mainstream (and affordable), but it is still useful if you are doing remote console work and don't want to set up network hoops with ad hoc HTTP servers just to view a remote diagram.
(I define it as "written words in a human-readable language". You must have something else in mind, because neither ASCII art nor control codes fit my definition.)
This research is not current, and is based on a flawed premise - generated agents.md are used. There isn’t a hand curated agents.md that has some knowledge passed to it that is withheld then measured against.
I think there is overall something here for current Claude which is there appears to be a hierarchy of conformance that breaks progressive disclosure and the utility of skills. It seems to honor the system prompt, user instructions, tool call results, and dead last skills. It applies a large amount of discretion as to whether to honor what skills say in the imperative and progressive disclosure seems to have at best a 20-30% recall. Other models like codex gpt 5.6 seem to be the exact opposite and slavishly adhere to the Agent/skills/plugins, to the point of being wasteful and dangerous. It feels clear there’s a tension being RL’ed around between conformance and skeptical behavior that neither has quite found the balance for yet, and is almost certainly an over constrained problem. I just find it funny Anthropic is the one you can’t trust with your wallet while OpenAI does precisely what you and your harness tell it to.
But this “science” and its editorializing are based on flawed techniques, don’t lead to the conclusion let alone the editorialized extrapolation, and are l
As far as I'm following it, the court didn't suggest the prompt was not a protected work.
But what the court did say is that the logo wasn't (sufficiently) derived from the prompt; those 300 words weren't contained or encoded in the logo, so if you copied the logo without permission from the author of the prompt, this was fine, because you weren't in any meaningful sense copying that prompt, or any other protected work of the author. (Which in my view is fair enough, really!)
Another way of thinking about it is to consider:
1) The number of different prompts that would yield a very similar looking logo. In many cases, a one or two word difference would probably yield a very different logo. (Conversely, you could rewrite an entire chapter of the novel, but the other chapters would still be translated more-or-less unchanged. You'd need to rewrite every sentence to end up with an entirely new set of sentences in the output, and since copyright covers the arrangement of text, not the plot or ideas, that matters.)
2) How feasible it'd be to look at the logo and accurately guess the prompt. (Again considering the hypothetical novel, if you translated the novel back into English, entire passages would likely be word-for-word identical with the original.)
A compromise option is a technical stop[0] where the plane lands to refuel but passengers don't have to transfer to another aircraft (and often don't even have to deplane). I've never experienced one of these but it sounds way less annoying/stress-inducing than a connecting flight.
From my layperson perspective that sounds relatively easy to address. For example they could require some minimum distance of continuous bike lanes to get approved for funding.
Decades of this sorts of "aha I know what rule change will fix this" are exactly how we got here though. They all look reasonable in isolation but at the systemic level we get garbage.
Ask it to write a program that outputs SVGs of animals using human modes of transportation, then run the program with "pelican" and "bicycle" as inputs.
.jpg){kind=link}
I know it's frowned upon to ask this here, but: did you read the article (or open GitHub) before posting this comment?
reply