sally_glance

5 ms·

31.
by sally_glance 7mo ago
In software, they can and do perform experiments (make a change then observe the log output). I don't think they possess a "world model" or that it's worth spending too much thought on... My reasoning is more along the l
32.
by sally_glance 7mo ago
Experimentation leads to experience, so I feel like this was included by the parent comment. And in the case of writing software, agents are able to experiment today. They run tests, check log output, search DBs... Sure, they can't hav
33.
by sally_glance 7mo ago
Coming from Ansible with hand-written config templates this was honestly a friction point for me - I felt like NixOS is trying to actively hide what it's actually going to configure. It's gotten better now that I read some nixpkg
34.
by sally_glance 7mo ago
Can we be sure? Maybe it's just very rare for experience, education and memories to line up in exactly the way that allows synthesizing something innovative. So it requires a few billion candidates and maybe a couple of generations too
35.
by sally_glance 7mo ago
You had me at "fuzzy", but lost me at "clean up" - because that's what I usually have to do after it went on another wild refactoring spree. It's a stochastic thing, maybe you're lucky and it fuzzy-matches
36.
by sally_glance 7mo ago
That's a great angle - will handcrafted software of the future become the equivalent of a tailored suit today? One might argue it already is, most companies and individuals do just fine using cloud/SaaS offerings and COTS apps. So
37.
by sally_glance 7mo ago
I don't know, maximum package turnover might be bounded and most likely you were previously not constrained by lack of drivers already... Sure you might try and expand but why would that work better than before? Especially assuming all
38.
by sally_glance 7mo ago
Philosophy territory now... you wrote about technology making labor unnecessary 15 years ago - Aristotele did ~2000 years ago too (same text where he tried to justify slavery but nvm that): "For if every instrument could accomplish its
39.
by sally_glance 7mo ago
Good point, but what if you were previously chaining horse carriage rides and now a car can cover the same distance as 10 of them with a single driver?
40.
by sally_glance 7mo ago
I honestly hope someone will read this comment and vibecode an Atlassian 2.0 platform, preferably open source. But really, I will take closed source and paid as well - just give me something that's on par in terms of features and integ
41.
by sally_glance 7mo ago
The amazing thing is that soon (actually already) we will be seeing people being paid way too much to prompt a LLM to email other people or respond to other peoples emails. And then turn these emails into presentations which will be turned
42.
by sally_glance 7mo ago
Hm I don't think a secondary market would work very well, using fab time productively requires lots of knowledge and collaboration with the provider. Compared to resources like grain or oil where it's basically "just come and
43.
by sally_glance 7mo ago
And they will most likely also be the last to benefit from hypothetical efficiency gains because they haven't been building up expertise (by burning billions) yet.
44.
by sally_glance 8mo ago
Yeah, for companies this works because there are external (government) entities providing and enforcing a framework. For countries, there is nothing like that. The traditional solution is separation of powers inside of the country I guess,
45.
by sally_glance 8mo ago
Great idea, except that I don't think it's easy to make sure we don't grant too much power. Basically this idea is the core of representative democracy. Problem is, the people who have been granted a lot of power are very goo
46.
by sally_glance 8mo ago
I'm pretty sure at least the better woodworking shop managers and QA people all have experience with woodworking and probably would also consider this their craft if asked.
47.
by sally_glance 8mo ago
Cool, I didn't know about this phenomenon. Reading up a little it seems like training multilingual forces the model to optimize it's internal "conceptual layer" weights better instead of relying solely on English linguis
48.
by sally_glance 8mo ago
In the past these trends were cyclical though. We're coming from an expansion phase (mainly driven by the COVID IT and AI craze) and now going through stagnation towards recession (global manufacturing crisis pulling our service sector
49.
by sally_glance 8mo ago
Maybe because you remember that statistic about single person households from a couple of days ago, I think Sweden was 3rd place globally? Though it's also a very happy country statistically, maybe something about the geography and upb
50.
by sally_glance 8mo ago
I also thought this is how skills work, but in practice I experienced similar issues. The agents I'm using (Gemini CLI, Opencode, Claude) all seem to have trouble activating skills on their own unless explicitly prompted. Yeah, probabl
51.
by sally_glance 8mo ago
I'm aware of that particular chapter of history, my master's thesis was on conversational interfaces. I don't think the potential of the algorithms (and hardware) back then was in any way comparable to what's currently g
52.
by sally_glance 9mo ago
Are you not worried that the sibling comment is right and the solution to this will be "more AI" in the future? So instead of hiring a team of human experts to cleanup, management might just dump more money into some specialized A
53.
by sally_glance 9mo ago
I mean, you can't review or cross reference something that isn't there... So interpreting in good faith, technically, maybe they just forgot to also check for completeness? /s
54.
by sally_glance 9mo ago
More like using a calculator but not being able to explain how to do the calculation by hand. A probabilistic calculator which is sometimes wrong at that. The "lots of theory but no practice" has always been true for a majority of
55.
by sally_glance 9mo ago
But you've got to admit that OPs tool does it quicker, except if you like to memorize flags or already have a script specifically for this. And it's much nicer on the eye than most (all?) nmap-guis out of the box.
56.
by sally_glance 9mo ago
Except that the constitution is apparently used during training time, not inference. The system prompts of their own products are probably better suited as a reference for writing system prompts: https://platform.claude.com/
57.
by sally_glance 9mo ago
Only way I can think of is paying for synthesizing training data using SOTA models yourself. But yeah, I'm not aware of anyone publicly sharing that they did so it's also speculation. The economics probably work out though, collec
58.
by sally_glance 9mo ago
Speculation I think, because for one those supposed proxy providers would have to provide some kind of pricing advantage compared to the original provider. Maybe I missed them but where are the X0% cheaper SOTA model proxies? Number two I&#
59.
by sally_glance 9mo ago
Underrated reply - I randomize the default ports everywhere I can, really cuts down on brute force/credential stuffing attempts.
60.
by sally_glance 9mo ago
This matches my experience. You mostly want to not even mention negative things because if you write something like "don't duplicate existing functionality" you now have "duplicate" in the context... What works for

More ›