It's baffling to me. I was already making API calls and embedding context and various instructions precisely using backticks with "md". Is this really all this is? What am I missing? I don't even understand how this "feature" merits a press release from Anthropic, let alone a blog post extolling it.
1. By giving this name a pattern, people can have higher level conversations about it.
2. There is a small amount of new software here. Claude Code and https://claude.ai/ both now scan their skills/ folders on startup and extract a short piece of metadata about each skill from the YAML at the top of those markdown files. They then know that if the user e.g. says they want to create a PDF they should "cat skills/pdf/skill.md" first before proceeding with the task.
I think the pattern itself is really neat, because it's an acknowledgement that a great way to give an LLM system additional "skills" is to describe them in a markdown file packaged alongside some relevant scripts.
It's also pleasantly vendor-neutral: other tools like Codex CLI can use these skills already (just tell them to go read skills/pdfs/skill.md and follow those instructions) and I expect they may well add formal support in the future, if this takes off as I expect it will.
I have been independently thinking about a lot of this for some time now. So this is so exciting for me. Concretizing _skills_ allows, as you said, a common pattern for people to rally around. Like you, I have been going dizzy about its possibilities, specially when you realize that a single agent can be modified with skills from all its users. Imagine an app with just enough backbone to support any kind of skill. From here, different groups of users can collaborate and share skills with each other to customize it exactly to their specific niche skills. You could design Reddit like community moderation techniques to decide which skills get accepted into the common repo, which ones to prioritize, how to filter the duplicates, etc.
I was puzzled by the announcement and remain puzzled after this blog post. I thought everyone knew you could keep use case specific context files handy.
As someone who's paid a lot of careful attention to this for the past four years or so, frankely, I think dang and ycombinator (it's almost crazy they're still running a forum these days) is making the absolute best of an impossible situation. I'd challenge you to find anyone doing better.
> But yes, thanks: I was once offered this challenge when faced with a Ren’Py problem, so I grit my teeth and posed my question to some LLM. It confidently listed several related formatting tags that would solve my problem. One teeny tiny issue: those tags did not and had never existed. Just about anything might be plausible! It can just generate Whatever! I cannot stress enough that this is worse than useless to me.
The probabilistic machine generated a probabilistic answer. Unable to figure out a use for the probabilistic machine in two tries, I threw it into the garbage.
Unfortunately, humans are also probabilistic machines. Despite speaking English for nearly a lifetime, errors are constantly produced by my finger-based output streams. So I'm okay talking to the machine that might be wrong in addition to the human that might be wrong.
> It feels like the same attitude that happened with Bitcoin, the same smug nose-wrinkling contempt. Bitcoin is the future. It’ll replace the dollar by 2020. You’re gonna be left behind. Enjoy being poor.
I mean, you were left behind. I was left behind. I am not enjoying being poor. Most of us were left behind. If we invested in Bitcoin like it was the future in 2011 we'd all be surfing around on yachts right now given the current valuation.
I think the author is right about AI only accelerating to the next frontier when AI takes over AI research. If the timelines are correct and that happens in the next few years, the widely desired job of AI researcher may not even exist by then -- it'll all be a machine-based research feedback loop where humans only hinder the process.
Every other intellectual job will presumably be gone by then too. Maybe AI will be the second great equalizer, after death.
Except we have no evidence of AI being able to take over AI research anymore than we have evidence so far that automation this time will significantly reduce human labor. It's all speculation based on extrapolating what some researchers think will happen as models scale up, or what funders hope will happen as they pour more billions into the hype machine.
It's also extrapolating on what already exists. We are way beyond 'just some academic theories'.
One can argue all day about timelines, but AI has progressed from being fully inexistent to a level rivaling and surpassing quite some humans in quite some things in less than 100 years. Arguably, all the evidence we have points to AI being able to take over AI research at some point in the near future.
> That's like looking at a bicycle or car and saying "all the evidence points out we'll be able to do interstellar travel in the future".
Incorrect. The difference between interstellar travel and land based travel on earth is enormous. The difference between current AI and AGI is tiny in comparison.
You're not saying it, but people might interpret 'interstellar travel' to be faster than light or very close to it, which means breaking the laws of physics or spending prohibitive amounts of energy. AGI needs no such thing, as evidenced by human brains running on < 100 watt.
Interstellar travel at far lower speeds is indeed probably something we will be able to do, looking at what we've done in space travel in the past 100 years. Although again, the resources required to do so as dictated by the laws of physics make it incredibly expensive. Given the low benefit it is probably not going to happen soon.
Reaching AGI is cheap as fuck and far more potent in comparison, so it will happen sooner.
> The difference between current AI and AGI is tiny in comparison.
Splendid. You should probably tell people like Yann LeCun this. He will be happy to hear it. /s
> Interstellar travel at far lower speeds is indeed probably something we will be able to do
Speed is not even an issue. We don't even know bigger problems like the long term effects of zero gravity in the human body or what dangers we'd encounter beyond the heliosphere.
> Splendid. You should probably tell people like Yann LeCun this. He will be happy to hear it. /s
Do you have an actual counterpoint or is this edgy nonsense the extent of your capabilities?
> We don't even know bigger problems like the long term effects of zero gravity in the human body or what dangers we'd encounter beyond the heliosphere.
I don't really think this is true, unless you'd be willing to say calculators are smarter than humans (or else you're a misanthrope who would do well to actually talk to other people).
OK, go ahead: Write a 10 page reasonably well-researched and sourced report on the political situation in the Netherlands in the past 20 years. You have 15 minutes.
The various 'deep research' tools available today can do this. You can't. AI already surpasses you (and me) in this task. Now think about the result after asking less intellectually capable people to do this.
Before you start balking about "hallucinations", pick some country you know about and ask a similar question to Google Gemini (2.5 Flash, and enable 'Deep Research'). Check the results and reconsider your 'calculator' straw man and 'misanthrope' ad hominem.
> How is the calculator a strawman? What is the argument that I am attributing to you that you do not hold?
I said: "AI has progressed from being fully inexistent to a level rivaling and surpassing quite some humans in quite some things in less than 100 years"
You attacked the intentionally inane (which is the purpose of a straw man): "Calculators are better at all humans in arithmetic, so they are smarter than all humans"
In no way is what I said close to what you attacked.
> As for the ad-hominem: I'm glad you don't dispute your misanthropy.
Sure, double down on your fallacious nonsense instead of opening your mind to reason and engaging with my point in good faith.
I provided properly described evidence for my point. You're stuck in "AI is just some calculator on steroids". Do your 'deep research' assignment before responding.
It's funny -- I independently implemented the same thing (without vibe coding) and found it doesn't actually work. When I ended up with was a game of telephone where errors were often introduced and propagated between the models.
The only thing that actually worked was knowing the target language and sitting down with multiple LLMs, going through the translation one sentence at a time with a translation memory tool wired in.
The LLMs are good, but they make lot of strange mistakes a human never would. Weird grammatical adherence to English structures, false friend mistakes that no one bilingual would make, and so on. Bizarrely many of these would not be caught between LLMs -- sometimes I would get _increasingly_ unnatural outputs instead of more natural outputs.
This is not just for English to Asian languages, even English to German or French... I shipped something to a German editor and he rewrote 50% of the lines.
LLMs are good editors and suggestors for alternatives, but I've found that if you can't actually read your target language to some degree, you're lost in the woods.
That doesn't match my experience at all. Maybe it's something to do with what your prompts are asking for or the way you're passing translations? Or the size of chunks being translated?
I have been astounded at the sophistication of LLM translation, and haven't encountered a single false-friend example ever. Maybe it depends a lot on which models you're using? Or it thinks you're trying to have a conversation that code-switches mid-sentence, which is a thing LLM's can do if you want?
I'm using o3 and Gemini Pro 2.5, paying for the high tier subscriptions. The complaints I get are from native speakers -- editors and end consumers. The LLMs tend to overfit to the English language, sometimes make up idioms that don't exist, use false friend words (especially verbs), directly translate English idioms, and so on. I've translated several book length texts now and I've seen it all.
50 people raiding a Nike store in LA isn't even beyond local news lol... you can find hundreds of videos online of this happening on a regular basis for the past decade.
> That's possible I suppose, but do you have any evidence of that
Yes, in the chat where a reporter was accidentally present, many of the messages were set to be disappearing. I don't know why anyone would do that if not to avoid recordkeeping laws.
> The images of the text chain show that the messages were set to disappear in one week.
Oddly, the Project 2025 training videos that presumably the members of the executive cabinet have seen say _not_ to delete messages or set messages to auto-deleting _because_ that would be in violation of federal record keeping legislation.
Legit can't tell if this is a real comment or someone put the article into an LLM with the prompt "Use technical jargon and an authoritative tone to make a response to this article justifying the SecDef's decision to have an unsecured line installed." I can't find anything online referencing unsecured lines being referred to as "unattrib".
DAA would have had to allow it with some risk acceptance. We don’t know for sure because all we have is an AP article asserting it is true and attributed to “sources”
Personal enrichment? There's already an enormous amount of evidence here to indicate that DOGE is working on behalf of a foreign nation state. It is seeming more and more likely that members of the DOGE team are simply secret agents for a foreign military.
> Within minutes after DOGE accessed the NLRB's systems, someone with an IP address in Russia started trying to log in, according to Berulis' disclosure. The attempts were "near real-time," according to the disclosure. Those attempts were blocked, but they were especially alarming. Whoever was attempting to log in was using one of the newly created DOGE accounts — and the person had the correct username and password, according to Berulis.
FWIW, that's getting too far out on the spy novel spectrum. Yes, they could be compromised. But to my point above, if they're indeed working for Putin or Xi or whoever, it's FAR more likely (given the demographic) that it's just because they took a fat bribe.
Not saying they’re compromised by foreign agents, although, I wouldnt put that out of the realm of possibility - but that either they and/or theyre tooling/setup is pwned
this is exactly what you save a zero day for, and something gives me the vibe about some of these guys that they dont take opsec very seriously, probably would not even need one
