10,967 karma · joined March 11, 2010
Looking to get in touch? edanm@btlms.com.
I'm the Chief of Staff to the CTO at Irregular, a frontier AI Security company. We are working to prevent harm from AI models.
I was also previously a 2x entrepreneur of bootstrapped companies (professional services). I've been involved in many different fields in software.
Feel free to reach out about anything!
Can you provide a link showing that the UN is saying this? I kind of doubt the UN ever called it a "Nazi-style" genocide, and I don't think anyone serious is alleging anything of the kind (including serious people/groups who do call it a genocide - I don't think they'd characterize what's happening as a "Nazi-style" genocide, and nothing they say implies they think it.)
That there's a learning curve, especially with a new technology, and that only the people at the forefront of using that technology are getting results with it - that's just a very common pattern. As the technology improves and material about it improves - it becomes more useful to everyone.
An author once wrote an intro to some short story. The story is part of a much larger futuristic scifi universe in which people have developed telepathy and other things through genetic means. And the specific short story was the first one he wanted to publish, and it was about a specific planet in that world, in which the whole story is basically a telepath coming into town and interacting with the population.
And the publisher returned the note that this wasn't scifi, it was fantasy. Because of course he did - stripped of the broader futuristic setting that the story takes place in, it's just a story of a wizard coming into town. Never mind that there are solid science fiction explanations for the "magic" - you don't get that in the short story.
AFAIK you are using the term "for cause" incorrectly. Firing "for cause" is usually more serious than simply firing for poor performance or similar. It's usually for something like breach of contract, for getting in legal trouble, for theft, wilful misconduct, etc. It also usually results in losing severance, options, etc.
Simple firing for poor performance is not firing "for cause".
A common sf theme is "here is this change to the laws of physics, what would our universe then look like". Eg Arrival (and the story it's based on), tons of books by Egan, any book with FTL.
I agree, and sometimes the line is drawn between SF being "things that are theoretically possible" vs. Fantasy where things are impossible. But then you have things like Egan's Clockwork Trilogy, which is "what if the laws of physics actually worked a bit differently in this specific way" but which I assume anyone would consider SF. As opposed to Brandon Sanderson's books, which could be described in a similar way, but are usually categorized Fantasy.
At the end, it's mostly a marketing and feeling thing. As one of my favorite authors put it, the different between SF and Fantasy sometimes comes down to - are you putting a tree or a spaceship on the cover of your book?
1. Many people who like one genre also like the other.
2. Many authors write in both genres.
3. There's a lot of similarity in the genres, and lots of things that are hard to categorize. More true lately btw.
Just as an aside though, I personally was an avid almost-only-SF reader for the first 30-ish years of my life, but lately have been reading a lot of fantasy as well. I highly recommend trying, especially more modern fantasy - I feel like the lines are even blurrier between them today, and a lot of the best work today has shifted from SF to fantasy. (I still love SF and there's a lot of great SF as well, to be clear.)
(The lines get blurrier when talking about imagined historical fiction, or even things like alternative fiction.)
If you do it, I'd love to hear your results.
In general, I wonder if we're at the point where an LLM watching you interact with your computer for twenty minutes can improve your workflow, suggest tools, etc. I imagine so, because when I think to ask how to do something, I often get an answer that is very useful, so I've automated/fixed far more things than in the past.
RAG is confusing, because if you look at the words making up the acronym RAG, it seems like it could be either of the things you mentioned. But it originally referred to a specific technique of embeddings + vector search - this was the way it was used in the ML article that defined the term, and this is the way most people in the industry actually use the term.\
It annoys me, because I think it should refer to all techniques of augmenting, but in practice it's often not used that way.
There are reasons that specifically make the "embeddings" idea special - namely, it's a relatively new technique that actually fits LLM very well, because it's a semantic search - meaning, it works on "the same input" as LLMs do, which is a free-text query. (As opposed to a traditional lookups that work on keyword search or similar.)
As for whether RAG is dead - if you mean specifically vector-embeddings and semantic search, it's possible - because you could theoretically use other techniques for augmentation, e.g. an agent that understands a user question about a codebase and uses grep/find/etc to look for the information, or composes a search to search the internet for something. But it's definitely not going to die in that second sense of "we need some way to augment LLMs knowledge before text generation", that will probably always be relevant, as you say.
Even the watered-down version of wave was something I used at my host startup, it was effectively our project management tool. And it was amazing at that.
I don't know how it would fare compared to the options available today, but back then, it shutting down was a tremendous loss.
(Will now be checking out your blog.)
(I'm not saying you're wrong, btw. People aren't always rational.)
Other than a vague sense of "fairness", can you articulate why insider trading should be illegal?
I'm talking about "shrinkwrap" software like Word or something. There's nothing even close to testing for that this is not just "system testing" it.
Why this is inherently different?
It bugs me, because the acronym should encompass any form of retrieval - but in practice, people use RAG to specifically refer to embedding-vector-lookups, hence it making sense to say that it's "dying" now that other forms of retrieval are better.
If I'm following you, then most "mathematical" CS is based on constructivist foundations? E.g. while a halting problem Oracle might "exist" in the mathematical sense, it's not considered to "exist" for most purposes of deciding complexity classes, etc.
> The real question isn't whether this oracle exists. It is what you want the word "exists" to mean.
I was going to say the same thing. I'm not sure what "exists" means in some of these discussions.
To be honest, I don't feel like I know enough about the constructivist philosophy. What would be a good place to start if I want to learn more about it?
I haven't yet read your PA proving Goodstein sequences article, though I have skimmed it and it is, indeed, super interesting.
And for the record, Godel, Escher, Bach was probably the single most important influence on me even starting to get interested in computation, etc.
I might be confused here, but isn't an Oracle to decide the halting problem something that everyone agrees doesn't exist?
The whole idea is for this to be a thought experiment. "If we magically had a way to decide the halting problem, how would that affect things" seems like a normal hypothetical question.
On the other hand, I think it's really cool to teach laypeople about things like "sizes of infinities", etc. They are deep math concepts that can be taught with relatively simple analogies that most people understand, and they're interesting things to know. I know that I personally loved learning about them as a kid, before I had almost any knowledge of math - it's one of the reasons that while I initially didn't connect with other areas of math, I found set theory delightful as a kid.
I just feel like if you need to first walk people through a bunch of philosophical back and forth on constructionism, you'll never get to the fun stuff.
Errr, I'm just assuming the axioms of ZFC. That's literally all I'm doing.
> In what sense do [numbers that can't be finitely specified] exist?
In the sense that we can describe rules that lead to them, and describe how to work with them.
I understand that you're trying to tie the notion of "existence" to constructability, and that's fine. That's one way to play the game. Another is to use ZFC and be fine with "weird, unintuitive to laypeople" outcomes. Both are interesting and valid things to do IMO. I'm just not sure why one is obviously "better" or "more real" or something. At the end, it's all just coming up with rules and figuring out what comes out of them.
For sure there are valid arguments on whether or not to use certain axioms which allow or disallow some set theoretical constructions, but given ZFC, is there anything that follows that is unprovable?
I've been a happy user of autojimp for many years, and just made the switch to Zoxide last week because it's maintained. I feel vindicated now :)
So a specific BB(n) is just a number and is computable.
I think this is true of the "children" question, but I actually disagree that this is what makes the Monty Hall question so confusing.
For one thing, I vaguely recall this being asked directly, and even after people agree on all the definitions explicitly, they still consider the answer wrong. (See e.g. some mathematicians like Erdos refusing to believe the correct answer without actually running simulations on computers... by that point you clearly have a real definition.)
For another, when I personally talk to people about Monty Hall, even after I explain the correct answer, and explain all the nuances, people tend to still have a hard time accepting the correct solution and claim to find it counterintuitive (as did I!).
But money is an abstraction of wealth, and wealth absolutely can be destroyed, in multiple ways:
1. It can be physically destroyed - if I break a window, that's wealth that is destroyed. That window now needs to be replaced, which costs materials and labor, which could've gone to building something new instead.
2. It can be spent on things that end up not used. If five years from now, those millions of GPUs are no longer in use, we created them for nothing instead of creating more of something people would use.
3. Wealth can be spent on the less important things, rather than the more important things. This is not exactly wealth being destroyed, just built more slowly, because instead of building lots of new wealth (via innovation, say) we're creating less valuable things.
I don't think any of the above are relevant to AI, btw.