ex-aws-dude

wouldn’t it result in whoever has the most money having the most secure software?

Hey man I'm just looking up the symbols like they told me

Well the whole lesson learned was that the Turing Test as it was defined was way too easy, it was a bad criteria for GI because it underestimates how easily humans find meaning/patterns in things.

I mean you could show people random markov chain gibberish in 1996 and they would swear they found intelligent meaning in it

https://en.wikipedia.org/wiki/Markovian_Parallax_Denigrate

its not, from what I've heard the difference in quality was not worth it

I finished it and thought it was overall good but I don’t know if I would say it’s good “bang for buck” in terms of what you get for the time commitment

Because you don’t need TPUs and anyone can scrape the internet

I think that has been well proven so far

Working games mine is all layoff posts recently

Just scrolled now and saw like 5

Reading Gödel, Escher, Bach but I've been stuck on the MU puzzle for days

> It's doing something more novel than the haters suggest

How is that different than https://www.nobodywho.ai/posts/jev-in-25-lines/

Couldn't you always get probabilities for a fixed set of tokens without training a new model?

I'm just talking about practical determinism not absolute

In the sense that if I hire a competent person, I can tell them something and they will deterministically do it

For an LLM I have no mechanism to even update the weights

All you can do is play around with context, which is like taping a post-it note to someone's desk

Imagine you had an employee where you had to tape 1000 post-it notes to their desk

If I do the same task 100 times I'm not going to suddenly do it crazily different at time 101 because I've built in the memory of how to do it

There is no RNG involved when I decide to push vs pull the unlabeled door to my building every morning, it becomes deterministic because its baked into memory

You can put stuff in context to deal with this but you can't do that for everything, its not practical and you would blow the context window

> Fully vibe-coded with Claude Code

Lame, my interest drops to 0 whenever I find out a project is vibe-coded

Because why would I be impressed if I can just go prompt it myself?

If its vibe-coded its just another display of claude code's abilities, which its like...duh at this point

That means the only thing interesting is the idea itself, which in this case...

> it is preferences all the way down

No you're falling victim to the common programmer fallacy that "my use case is everyone's use case"

These things exist because people had different use cases and priorities over the years

I've seen LLMs do something correct 98% of the time then randomly do something crazy that a human would never do because we have continual learning

As humans we don't have our memory reset multiple times per day

IMO you need to pivot to higher-end software as the low-end cheap stuff is going to be gobbled up by vibe-coders

The type is guaranteed because you only read the probability from your set of choice tokens

You don't actually use the "next token" that the model chooses

That's what the author is doing in this part

  token_ids = [model.tokenize(text=label.encode(), add_bos=False)[0] for label in labels]
  choice_logits = numpy.asarray([logits[token_id] for token_id in token_ids])
  logprobs = choice_logits - numpy.logaddexp.reduce(choice_logits)
  probabilities = numpy.exp(logprobs)

This works because the model is always producing probabilities for all tokens

No I'm s̶p̶a̶r̶t̶a̶c̶u̶s̶ Jev!

I operate at the level where I'm telling the agents what functions, classes, code structure to create

So its still fairly involved and low level but a lot less manual typing

I also don't use auto-mode when they write code, I review the diffs immediately as they come and approve/deny as I find it stops the agents going too far down the wrong path.

I think the term automated programming captures it best

I don't see why its that crazy that a system with a huge amount of parameters starts to exhibit emergent behavior

> Only accident rate per mile driven (compared to human drivers' rate) is a relevant comparison

This is such an insane take I see all the time from self-driving boosters

If a self driving car glitches out and crashes in some edge case pathological scenario we don't just accept that as totally fine because its hidden under big statistics

The reason why a crash happened does matter, its not just about aggregate statistics

As a thought experiment if I have a perfect self driving system but I add some code that purposefully crashes 1 in 10 million rides are you ok riding in it since the aggregate statistics look good?

I dunno this sounds like some insane goalpost moving

Unfortunately we have entered Eternal Sloptember

I can't see governments not somehow getting more involved

I think large amounts of compute are big strategic asset for a country and governments will partner with the labs

You have to throw in “load bearing” in the request though to avoid suspicion

How can you not need maps? That’s crazy to me

The homer simpson car

OMG THEY CHANGED THE FONT, TOP OF HN!!!

> anyone producing content, everyone’s creativity, is fed by something that others did before

The thing I produce does not replace demand for the original though?

I can’t recite the original for a million people

Isn't that a challenge with RL anyway that for a lot of problems its hard to even know accuracy continuously for each step

Nah there's a lot of reasons to prefer physical

* You own it forever

* Nothing to charge

* Don't have yet another screen to stare at

* Its the definitive way to collect books (similar in the sense of vinyl for music)

* Going to used bookstores is awesome

* Seeing your whole collection physically is awesome and can be a conversation starter

* Graphic novels/comics are still better physical