Hacker Newsnew | past | comments | ask | show | jobs | submit | krinne's commentslogin

This post is refreshing - smells of the pre AI discussions on the internet. A new language, a new syntax, heavy debate with people who have spent years writing code. I think someone should start a community online where AI isnt allowed.

> I think someone should start a community online where AI isnt allowed.

In case you haven't followed the saga, the latest[1] digg.com relaunch failed because they couldn't deal with the bot onslaught [2]. Whoever finds a reliable way to keep AI out of an online community first is likely to become a very rich person.

[1] Second-to-last, actually, seeing as there seems to be a new homepage right now.

[2] https://www.techspot.com/news/111698-digg-relaunch-fails-two...


I suspect real ways to keep AI out of a community, or really to have an online community at all, are going to be structurally incompatible with making anyone rich. The possibility of getting rich poisons the incentives.

It can't just be money. It also has to remove any notion of score or ranking. There should be NO incentive to artificially increase anything.

Look at Advent of Code. Free site, fun community, but it had a leaderboard. The moment AI was advanced enough, it began dominating the leaderboard. The solution: kill the leaderboard. Sure, you can still solve all the problems with AI and get yourself full points, but you're not competing against anybody, so why bother?

As soon as you can get ahead of others at something, even for something as stupid as a karma score on Reddit or on Hacker News, somebody's going to want to increase it badly enough to start cheating.


Maybe. That's a bit of a different problem: similar mechanic, but much weaker forces with some different pros and cons. Maximizing imaginary internet points is qualitatively different from maximizing actual power in the form of money. I think there's still room to experiment with aligning the incentives on things like karma for community-driven moderation.

If moderation is the problem maybe only allow downvoting? Of course that may bring to the surface a different set of problems (no solution is without drawbacks), but it removes the "numbers go up" incentives.

> In case you haven't followed the saga, the latest[1] digg.com relaunch failed because they couldn't deal with the bot onslaught [2]

Given that they wrote their goodbye post using LLMs and gave up after such a short amount of time, I don't take that at face value the same way I don't believe AI layoffs


We've already got models that can handle it - eg web of trust. We don't want to use them.

Plus "AI" is a spectrum, with "the AI fixed a typo for me" at one end, and "the AI writes my posts for me" at the other.


Web of trust has been a miserable failure because of the GPG's horrible UX and high entry barrier. It's dead, Jim.

I think they were referring to invitation trees like lobste.rs.

That's for children. Make the Pope a certification authority, and he can certify people he trusts, and so on, until the chain reaches you. When you commit a sin, your certificate gets revoked, along with its children.

It would give the web of trust a flair of biblical damnnation, and after your fall you could always seek a new certification authority more aligned with your values, like Milei or Putin.

When a world leader dies, the tree pruning would be almost apocaliptic.


> Whoever finds a reliable way to keep AI out of an online community first is likely to become a very rich person.

I believe it’s the opposite: You have to pay competent human moderators. Like here on HN.


Half the posts here are AI grifting.

That’s a pretty extraordinary claim. Do you have anything quantitative to back it up? HN has bots, I think that is a fair assumption, but half? How do you know that for sure?

[flagged]


TIL that statements of obvious exaggeration are called causal, hyperbolic statements. Who knew?

All kidding aside, I've never heard them called causal hyperbolic before. It's a good term.


Perhaps requiring webauthn credential for any post/comment with a whitelist of permitted webauthn hardware devices which must have touch/interaction enabled.

I'd have to read the FIDO specs, however the only place I've seen webauthn hardware pinning in the wild is with Azure AD/Entra which is ostensibly based on token GUID. If this is the only enforcement mechanism available, it's spoofable.


Then you’ll end up with a forum of only bots because they’ll spoof it and real people won’t put up with the hassle.

FIDO tokens are designed to able (if authorized by the software, your web browser typically offers a pop-up where you can decline this) to prove their membership of a batch of tokens, but not their individual identity.

The Entra feature you thinking of lets somebody say "Only things which can prove they're in this list work". This could make sense if you, as their employer, issue every employee a custom DoodadCorp Doodad FIDO key and so you don't want somebody's Yubikey or off-brand generic device to work. It's stupid and you shouldn't do it in other scenarios, but your "this is how we detect humans" idea is arguably a scenario where that could make sense.

[Edited to add: This feature is called "Attestation"]


Doesn't actually work that well. Browsers hate this, the hardware isn't actually difficult for bots to access, and privacy story is bad. There are solutions being worked on.

This would result in hardware farms of such devices being automatically operated, like the existing iPhone farms used for similar purposes.

Pay per interaction model? 1 cent per post.

Probably the only solution. But the penny has been deprecated so it will have to be a nickel.

Only copper was deprecated. Real $0.01 absolutely exists in your bank etc.

Anyway, marketers see a popular site like a physical billboard, where they would pay thousands a month for their message to be seen by thousands of people. If you made it cost pennies to post, and a few more pennies to boost and astroturf, AND that the post would be seen by millions of people, they'd say "By Grabthar's hammer, what a bargain!!" and order a hundred more per day...


I think pay per account (like the Metafilter model) would be better. If you charge $10 for lifetime access and ban slop-posters, it quickly becomes expensive to be a slop merchant

Altman’s orb is as terrifying as it is because businesses might see it as a solution to a real problem—a problem he helped to create.

The reason we didn't see it in the US first is because he was probably gathering data from a pilot system. When he finishes his moonbase and/or volcano lab then we should probably start moving to the Northern Territories or the Yukon.

lobste.rs uses a web-of-trust referral system. I guess it still involves a moderator killing off bad nodes, but it seems to scale well

yeah but I can't post there because I don't know anyone with an account and frankly CBA traipsing around looking for someone who has an account.

does seem like more things will have to go this way though


+1, if anyone wants to help me I'd be honored. mail me at ramon(@)odeva(.)nl

I actually hung out in their IRC for a while hoping to maybe form a social connection and then get an invite, but somehow it just never materialized, and the chat wasn't engaging, so I stopped coming back...

> it seems to scale well

This seems hard to say without knowing how large it is. To me, it seems like a relatively small community, far smaller than HN anyway.


You can see how large it is at https://lobste.rs/stats

Maybe you have to be logged in to see that? Because I don't have a login and all it shows me is a bunch of "no data".

It shows "No data" for each of the 5 metrics though?

Age verification with facial ID? ;)

Isn’t the solution high-quality identity verification? There are piles of digital identity companies out there. They make money selling to banks for KYC compliance. Heck, there are background check as a service companies designed to add trust to gig economy platforms.

I used to think that a small payment could accomplish the same thing, but X selling blue check marks proved that doesn’t help much. Well, at least it’s a much weaker signal than the previous curated version.

The challenge is any barrier to entry high enough to discourage motivated spammers is also high enough to discourage casual users. That disrupts the network effects you’ve traditionally needed to bootstrap a social website.

If I was trying to get a new social site off the ground right now, I would try:

1) secure a good brand from the pre-AI era. Twitter, Digg, Friendster, MySpace. Something that motivates a first look.

2) Require third party identity verification on sign up, configured so the social site is never the custodian of PII, though require enough demographics to support high-value advertising later. Verification is free to the user, ideally provide multiple verification options- one US and one EU at minimum.

3) Target a few core communities and invest. Find the people who moderate historically great subreddits, were active in twitter communities during the good years, etc. get them in your platform. Maybe even pay them.

That should be enough to tell you if it’s going to work or not.


Blizzard's Real ID system would fix all of this. It was ahead of it's time /s

The amazing thing about AI is that you don’t even need AI superfans to shoehorn it into a conversation that doesn’t even touch on AI. Detractors will do it for them.

Yes, it’s similar to Trump. But that makes sense right? AI is changing the world drastically, and so is Trump and his fascist friends.

> a community online where AI isnt allowed.

This is something I think about a lot, especially how one could pull it off without tearing down anonymity online. Having some sort of "proof of humanity" is a hard problem to solve.


not really, you can just ask people to do a shit job identifying traffic lights and motorcycles and they'll do it.

> I think someone should start a community online where AI isnt allowed.

That's lobsters I guess. AI posts got banned there after a 300+ comment discussion, probably the biggest ever on the site.

The exact rule the moderators settled on was "meaningful human authorship" but don't be fooled: a lot of people on lobsters are ideologically opposed to LLMs. Doesn't matter how "meaningfully" the technology was applied. My work was classified as slop simply because AI touched it. People referred to me as an exhibitionist and fetishist when I talked about using AI. Just a heads up for anyone who's thinking of joining.


on my way!

I like the Lobsters stance on AI. Also posts are tagged hence you can decide for yourself if the topic may be interesting despite being touched by AI.

I haven't found a way to join Lobsters yet hence for the meantime I keep posting here only. Not so much of an issue since I am mostly a happy lurker.


Here's the lobsters stance in case anyone is wondering:

https://lobste.rs/s/29pm2f/llm_generated_submissions_should_...

TL;DR Plenty of ideological opposition which will loudly call you out for any usage of AI, and also quite a few nuanced takes which will no doubt be overridden by the vocal opposition. It got to the point the site itself told me to leave.

I don't feel comfortable posting my projects there anymore, even though they do have meaningful human authorship. Still have my account, but I'm essentially a lurker now. I'll participate if someone else independently posts my stuff there.

Feel free to reach out and introduce yourself if you want an invite.


Ironically, the top comment (this) is now about AI.

> a new syntax

How is the syntax new?

It looks like lispy - see the outer parens in the examples given.


Heh, every time you show a average developer lisp for the first time the reaction is the same. Little do they know conditionals, GC, REPLs, macros and more comes from the syntax and language dreamed up in the 50s/60s.

I don't see why Lisp's history would necessarily imply the family is worth learning in 2026. What (other than macros) do lisps offer that other modern languages don't?

You don’t program in Lisp, do you? I used to be confused by the smug Lisp weenies. Now I am one. And the difficult thing I’ve found over the years is that Lisp is sort of unexplainable. You either “get it” or you don’t. Yes, it has macros, but macros are a bit overrated. I’ve been programming in Lisp for decades and I rarely write macros. I think the thing that is difficult to convey is how powerful Lisp’s core execution environment is while at the same time being just a page of code that a CS undergraduate can understand. Literally everything else is a library. And those libraries can create syntax, generate code on the fly, and do many other powerful things. But most people won’t “get it” until they take the plunge. I didn’t. Until I did. And now, I don’t feel a need to defend Lisp at all. It won’t go away. You can’t kill it. The folks that “get it” will always have it, and those that don’t “get it” will reach for their Blub language again and again. Such is the way of the world.

Yeah, I didn't get it either until I implemented my own lisp.

> the difficult thing I’ve found over the years is that Lisp is sort of unexplainable

I've found that getting rid of the parentheses helps.

  f(x)
  (f x)
  ["f", "x"]

  (print (< 10 20))
  ["print", ["<", 10, 20]]
Lisp code is just normal Python lists which get evaluated by an interpreter function. Like this:

  code = ["print", ["<", 10, 20]]

  def eval(code):
      # magic

  eval(code)
  True
Filling out that eval function is a great way to learn lisp.

These articles are very good and accessible:

https://www.norvig.com/lispy.html

https://norvig.com/lispy2.html


Getting lisp is analogous to spiritual enlightenment. If someone doesn't have the eyes to see and ears to hear, there's little you can do for them, except pray.

Is the magic a property of the broader language-family (and could be experienced with Janet, Racket, whatever), or Common Lisp specifically? When people praise the core execution environment they're typically praising Common Lisp specifically.

What's the quintessential "now I get it" experience, in your mind?


Any Lisp will do. I had mine with CL and now use Clojure. Racket is great. I’ve never used Janet, but it looks great, too. Pick one. But make sure you engage with the community and ask about editors and tooling. Don’t just fire up VSCode and start typing parentheses. If you do that, you’ll just be frustrated. And nobody who programs in Lisp does that. We use lots of plug-ins and have a live REPL right in our editors at all times. In other words, do what Lisp programmers do. If you try to “do Lisp” the same way you “do JavaScript” or whatever, you’re just going to get frustrated.

If I could explain the moment, I would. But I really can’t. That said, one aha moment for me was reading McCarthy’s original Lisp paper and realizing the whole core of the language was a single page (17).

https://www.informatik.uni-bremen.de/agbkb/lehre/pi3/folien/...


> What's the quintessential "now I get it" experience, in your mind?

Learning that https://calva.io/paredit/ exists and moving your cursor along the AST or moving expressions around with nice hotkeys. Then making simple macros for infix notation and SQL and so on, which operate on the AST too. Realizing that there is no "architecture" because any repeated code or pattern can be easily abstracted away with a macro. Realizing that you can just describe your problem on paper, making up the perfect notation, then implement that notation in a few hours.


There are reasons why not that many programmers “get it”, and it’s not because the others are uninformed. It’s a matter of valuing different things.

Hmm, that'd be weird, how do you know you "value something different" if you haven't "got it"? You'd need to "get it" first, then you can understand if you value something different or not, otherwise how would you know?

The usual insinuation is that if you don’t like it, you “just don’t get it”. I meant not “getting it” in that sense.

Absolutely! But it’s also because they don’t really understand one of the things. It’s the Blub Paradox.

> You don’t program in Lisp, do you?

Not anymore. I started with Racket and went through the Little Schemer. I did Clojure for a while. I even used Babashka to write all my scripts, then later rewrote them in other languages.

I gave it a good try. Maybe it wasn't enough to properly "get it"?


Aw man I love babashka. I will say the lack of static types in clojure is pretty brutal for me. Especially when combined with the obtuse error messages. But I still love babashka and the whole REPL driven world.

What did you end up rewriting your bb scripts in?


Python, Go, or Rust, depending on the complexity. Python (with strict typing) for the simplest ones, or those where startup time wasn't a concern, Go for the medium-complexity ones that I could do with only the standard library, and Rust for everything else. Besides lisps in general still feeling alien to me, I also really like static types.

If there's one thing that I sometimes wish Lisp had, it's types. Most of the time, I don't need or even want them. But when you're doing a big refactor or changing the shape of your primary data structure, it would be nice to have the compiler be able to assist you in detecting locations where you've cross-wired something. But other than that, I don't care. And yes, Clojure's error messages could be better, but they have been getting better over time.

> If there's one thing that I sometimes wish Lisp had, it's types.

Let's write some very silly code to turn an integer into a list of digits in Common Lisp:

  (deftype Digit ()
    "A non-negative integer smaller than 10."
    '(Mod 10))

  (defun integer->digits (integer)
    "Turns a given INTEGER into a list of digits."
    (declare (type Integer integer))
    (labels ((digit-loop (integer digits)
                (declare (type Integer integer)
                         (type List digits))
                (if (< integer 10)
                    (list* integer digits)
                    (multiple-value-bind (quotient remainder)
                        (truncate integer 10)
                      (declare (type Integer quotient)
                               (type Digit remainder))
                      (digit-loop quotient
                                  (list* remainder digits))))))
      (declare (ftype (Function (Integer List) List) digit-loop))
      (digit-loop (abs integer)
                  nil)))

  (digit-loop 2026) ; => (2 0 2 6)

  (digit-loop "2026")
  ; The value
  ;     "2026"
  ; is not of type
  ;     INTEGER
  ; when binding INTEGER
  ;
  ; Type HELP for debugger help, or (SB-EXIT:EXIT) to exit from SBCL.
  ;
  ; Restarts:
  ;   0: [ABORT] Exit debugger, returning to top level.

Yep, I know, but most Lisps do not have those declarations available and even in CL they aren’t used very often. But yes, they would help in CL.

That's a very reasonable try. Your statements are not unfounded. If I may ask, what's your daily driver now and why do you favor it over Lisps?

Python (with types), Go or Rust, depending on the complexity. I really missed static types in Clojure. I think no Lisp does static typing well. I guess there's Coalton, but it's too niche.

this is incredibly smug, but fun to read :) I briefly "got" Clojure but forgot again. Maybe I'll give this Janet thing a try.

LOL, indeed. Clojure is fun. I haven't used Janet, but I appreciate seeing some of the good ideas that it stole from Clojure (stealing being the sincerest form of flattery, and all that). IMO, one of Clojure's greatest gifts, above and beyond other traditional Lisps like CL and Scheme, is its focus on immutable data structures. When I started playing with Clojure, I was skeptical. I figured performance was going to be horrible. Now, I can't live without them. It's one of those subtle features that just changes how you program. It's one reason I choose Clojure over CL and Scheme today. Janet seems to have both mutable and immutable data structures, which is nice. Clojure has transients, but that's sort of partially mutable. That said, with Clojure, one of the nice things is that you can always drop back to Java's full mutability if you want, but that's obviously relying on the platform and not Clojure the language.

Honestly... it's entirely possible to "get" Lisp and at the same time not really see anything that compelling about it for doing your own work.

I "get" Lisp just fine, have made my own hobby Lisp interpreters, have written programs in Lisp, am an emacs user, etc. etc.

And yet if you handed me a terminal and an editor and asked me to write a program, I would never reach for Lisp to do it. My eyes don't like it. (Also I like static types).


S-expressions. Defining data in JSON or XML is way worse than S-expressions.

I agree that S-expressions for defining data are pretty nice, but you don't need to actually use a Lisp to use them. You can just use them like you would XML.

Just saying "macros" is a bit reductive: in CL, you have access to the full language at (read) parse time, compile time and runtime. Said macros also mean that logical OR/AND short-circuiting isn't a compiler black box, you can implement such behaviour easily yourself.

------

The syntax is actually a big pro for a lot of people. I love its streamlined look that basically reads like Python once you let your IDE indent properly and learn to see "through" the parentheses (CL, Scheme).

The original language where everything is an expression and it shows. Where Python still needs an ugly ternary and made match a statement, Lisp has had the perfect IF and COND since the dawn of time.

Symbols are still a cool and useful concept that almost no other language I know of got.

The numerical tower - despite some holes - is amazing. Built-in rationals and "correct math" as sane default (i.e. 1/2 not returning 0) never get old.

------

And if you let me rave about CL specifically (e.g. DECLARATIONs as "#pragma done well", restarts, CLOS/MOP, runtime READ/COMPILE, etc...), there are a lot of cool features barely copied anywhere that'd improve other languages, but these aren't part of "what make Lisp Lisp".


It’s been a few months, but I built a tool by Janet lang to communicate with an LLM via HTTP. Of course, I probably had Claude Code write it for me. It turned out better than I expected.

I was really impressed by how small the executable file was. I’d only ever done web development with Node.js up until then.


Amazon has this thing of stack ranking and firing the lowest performing employees - its just a pattern that works for them as it creates urgency with all the employees who remain. This is probably all there is to it.

Is this available on OpenRouter ? Cant really trust thehindu for this kind of journalism - they arent really known for original tech content


I think OpenAI is just trying to grow recklessly as much as possible. Anthropic seems to be more thoughtful about how they are moving.


Very interesting. This should be useful in India also, if you ever get around to it.


But doesnt existing AI systems already learn in some way ? Like the training steps are actually the AI learning already. If you have your training material being setup by something like claude code, then it kind of is already autonomous learning.


Most, if not all, commercially available AI models are doing offline learning. The cognition is a skill that is only possible on online learning which is the autonomous part the authors refer to, that is, learning by observing, interacting.

In that sense the "autonomous" part you said simply meant that the data source is coming from a different place, but the model itself is not free to explore with a knowledge base to deduce from, but rather infer on what is provided to it.


> The cognition is a skill that is only possible on online learning which is the autonomous part the authors refer to, that is, learning by observing, interacting.

This is the "Claude Code" part, or even the ChatGPT (web interface/app) part. Large context window full of relevant context. Auto-summarization of memories and inclusion in context. Tool calling. Web searching.

If not LLMs, I think we can say that those systems that use them in an "agentic" way perhaps have cognition?


> This is the "Claude Code" part, or even the ChatGPT (web interface/app) part. Large context window full of relevant context. Auto-summarization of memories and inclusion in context. Tool calling. Web searching.

From what I've been learning in my uni, this is said pre-programmed. Cognition is really the ability from, out of no context, no knowledge of what you are capable of, to learn something. These tool calling and web searching are, in the end, MCP functions provided by the LLM provider themselves.

It's an entire academic discussion, about how things start. For example babies: they someone have a knowledge base on how to breath, how to cry, but they have absolute no knowledge on how to speak and it learns by the interactions with the parents.

LLMs try as much as they can create this by inference and pre-programmed functions, but they don't have a graph of memories with utility to weight their relevance in the context. As others said, the context window dies as soon as you close the session.

They also don't have the epistemic approach that is to know that another agent knows about something just by observing the environment they were all put in.


No, no they don't. Actual learning survives beyond "sufficient context window".

Start a new chat, and the "agentic" system will be as clueless as before


They can write to the filesystem, and future instances can read it (and write more). The agentic system does not remain as clueless as its LLM's first instantiation.


Do you need a notebook to remember who you are? The point is to update the model weights so it learns.


No, but I don't need floating point weights either. My evolved biological systems work very differently from artificial digital systems.

My point is that it's not the "model" that is the thing that is demonstrating cognition here, it's the "system" that uses the model and stores information and can retrieve it later. To that system, these notes are more the equivalent of my memories than my notebooks.


These are not "memories", as with every new session these are entirely new, never-before seen files that the "system" may or may not use.

And it is not "learning" (which was the initial claim) as the system never learns beyond what was already there in the training data, and any new information you supply are new data, from scratch, that is immediately forgotten once the session is over.

It's easy to prove: start a new session in your project and ask "what is this project about". Two days from now, in a new session, ask the same thing. Observe how in both cases it re-reads the files, greps source code etc. Meanwhile a system with actual memories and learning wouldn't have to do that.


Memento notebooks are not learning


If you let the AI train on your prompts it will actually learn indirectly. It is still offline learning though.


I wasnt able to find a way to access this - is this something accessible only to enterprises ?

Would love to take it for a spin, if that is even possible.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: