breuleux

7 ms·

91.
by breuleux 1y ago
Well... what are you trying to do with "1 + 1"? Lisps are "homoiconic" in the sense that everything is a list, but honestly? When I worked with Scheme in the past, I almost never manipulated code as actual lists. It&#x27
92.
by breuleux 1y ago
I think there is potential to do something like that with template strings [1] (upcoming feature in Python 3.14). The choice of {} for interpolation isn't ideal because any code with dict and set literals becomes super awkward, but bes
93.
by breuleux 1y ago
Yeah, I've been using this trick to implement a hot reload library, to only update the specific functions that are changed without re-executing modules: https://github.com/breuleux/jurigged I also use it in a mult
94.
by breuleux 1y ago
Speaking for myself, LLMs are reasonably good at writing tests or adapting existing structures, but they are not very good at doing what I actually want to do (design, novelty, trying to figure out the very best way to do a thing). I gain s
95.
by breuleux 1y ago
> It's not impossible from a scientific perspective ... to create a simulation with intelligent life in it and to save some resources by starting 6000 years ago from a complex seed state rather than simulating 16 billion years of ph
96.
by breuleux 1y ago
As someone who believes knowledge of one's own future is plainly impossible even under determinism (similar reasoning as the halting problem), I actually found myself kind of annoyed by Story of Your Life. It's a good story based
97.
by breuleux 1y ago
The only one I was fairly sure was human was #6, and that was the only one I kinda enjoyed. In any case, as someone who reads a good deal, I agree. I didn't think any of the stories was particularly great (not enough to bother ranking
98.
by breuleux 1y ago
It'll be easy to make new software. I don't know if it's going to be easy to sell it. The more software AI can write, the more of a commodity software will become, and the harder the value of software will tank. It's not
99.
by breuleux 1y ago
> These reduction to absurdity arguments about the temperature of the earth assume we're not going to space. I don't understand why you reject this idea outright. Most people just want to stay home, they don't want to go a
100.
by breuleux 1y ago
> one could be having an agent as a server, where "prompting" is just a command `prompt ...` sent to the agent as a server Yeah, I prefer something like that (which should be strictly easier to make than these elaborate TUIs).
101.
by breuleux 1y ago
Well, that depends. Very inefficient code tends to only be used when absolutely needed. If an LLM becomes ten times faster at answering simple prompts, it may very well be used a hundred times more as a result, in which case electricity use
102.
by breuleux 1y ago
I mean a session that isn't limited to interaction with a single program. For example, if I have an agent in this paradigm, I want to easily interleave prompting with simple commands like `ls`, all in the same history. That's not
103.
by breuleux 1y ago
What bothers me is that what I like about terminals is the scrolling workflow of writing commands and seeing my actions and outputs from various sources and programs sequentially in a log. So what I want is a rich full-HTML multi-program
104.
by breuleux 1y ago
Regardless, it is an extremely uncertain proposition that we can do anything in the present that would have a reliably positive impact on their lives in the far future. It's hard enough to figure something out for the billions of peopl
105.
by breuleux 1y ago
I mean, that's just a popularity contest. People with the greatest media presence will get the most votes, because they are known by the most people. Even if I had a very precise idea of what merit was to me, I have no idea who in the
106.
by breuleux 1y ago
If you have fifty bright and highly competent people, I'm skeptical that adding fifty idiots is going to make much of a difference. Most idiots will accept meritocratic authority if they can be convinced that their needs are taken into
107.
by breuleux 1y ago
If the sample size is low, it could come out at 10% purely at random, but that is still likely to undermine confidence in the system (in the immediate). Pragmatically, I think it makes sense to have quotas for a few protected classes, to ma
108.
by breuleux 1y ago
I feel that adding qualifying criteria is an attempt to solve a problem that hasn't been demonstrated to exist, in a way that hasn't been demonstrated to work. The main threat to a well-functioning society are people acting in bad
109.
by breuleux 1y ago
> A training program that acclimates new members to the system. If terms are say, six years long, then the first year can be entirely devoted to training. A more organic version of this would be to select at random from people who alread
110.
by breuleux 1y ago
I find tab extremely distracting and it was the first thing I turned off. I have no idea how people can tolerate it.
111.
by breuleux 1y ago
I think the issue is going to turn out to be that intelligence doesn't scale very well. The computational power needed to model a system has got to be in some way exponential in how complex or chaotic the system is, meaning that the ef
112.
by breuleux 1y ago
I think the main issue is people using LLMs to do things that they don't know how to do themselves. There's actually a similar problem with calculators, it's just a much smaller one: if you never learn how to add or multiply
113.
by breuleux 1y ago
I don't know about that. Cutting corners will never die. Ultimately I don't think the heuristics would change all that much, though. If every time you review a person's PR, almost everything is great, they are either not usin
114.
by breuleux 1y ago
Sure, the core principles are mostly the same, but the point is that it is getting easier and easier to extract value from these models, which means that the learning curve is getting flatter. The main difficulty for now and for the foresee
115.
by breuleux 1y ago
In theory, but how long is that intuition going to remain valid as new models arrive? What if you develop a solid workflow to work around some limitations you've identified, only to realize months late that these limitations don't
116.
by breuleux 1y ago
They won't. The speed at which these models evolve is a double-edged sword: they give you value quickly... but any experience you gain dealing with them also becomes obsolete quickly. One year of experience using agents won't be m
117.
by breuleux 1y ago
It works, ish . Type checkers have a rule that you must have a non-stub, non-decorated overload, so if using ty for example, you need to --ignore invalid-overload. mypy doesn't seem to recognize the hack, although I could've swor
118.
by breuleux 1y ago
Yeah, I suppose. I wrote "dependent" because I think that's the term of art of it, but you're making me think I should probably change the header to something more intuitive to people unfamiliar with type theory.
119.
by breuleux 1y ago
I use it quite a lot (having made it), most often when I need to recursively process complex heterogeneous data structures. Merging two data structures, treemap, processing a Python AST, etc. Also in any situation where you want to be able
120.
by breuleux 1y ago
Mainly codegen. Most other libraries do something like `return dispatch_dictionary[tuple(map(type, args))](*args)`. Whereas ovld generates a specialized function to the set of actual signatures of the overloads (still with a dictionary, but

More ›