philipswood

7 ms·

61.
by philipswood 1y ago
Those posthumans will be roasting their marshmallows around black holes long after the song in this video ends.
62.
by philipswood 1y ago
Maybe it's more of a 'paraphrase'? >WL#13070: Deprecate && (resp. ||, !) as synonym for AND (OR, NOT) in SQL statements
63.
by philipswood 1y ago
Agreed. I like the paper and I think it raises important considerations, but it does overstate its case a bit. I mean - people do learn new codebases after all. It's just more difficult to do than some think.
64.
by philipswood 1y ago
As Naur describes this, the full code and documentation, and the resulting model you can build up from it is merely "walking the path" (as the blogpost put it), and does not encode "building the path". I.e. the theory of
65.
by philipswood 1y ago
> What LLMs can't do is explaining why we have this and why it needs to change to "a = b - 1" in the new iteration. I did a search on Github for code containing `a=b+1` and found this: https://github.com/ha
66.
by philipswood 1y ago
The paper he quotes is a favorite of mine and I think is has strong implications for the use of LLMs, but I don't think that this implies that LLMs can't form theories or write code effectively. I suspect that the question to his
67.
by philipswood 1y ago
> Theories are developed by doing the work and LLMs do not do the work. They ingest the output of work. It isn't certain that this framing is true. As part of learning to predict the outcome of the work token by token, LLMs very wel
68.
by philipswood 1y ago
Nitpick: "Atoms are called atoms because they have no subdivisible parts". Oops, it turns out we used the name too soon. When people say "universe" these days they mean the "visible universe" (or maybe the visi
69.
by philipswood 1y ago
OK, I hear you. So we're talking about hints - is Anna immortal? or just very long-lived? Assuming the weaker choice still leads to millions of HUMANS dying from the possible cures to the diseases her cohort had. The postdoc was clearl
70.
by philipswood 1y ago
The story has a really tragic implication though... It strongly hints that millions will die in pain from the knowledge that the post-doc "lost" and that the other rats in the cohort died in vain.
71.
by philipswood 2y ago
Do you feel guilty about using a compiler? An assembler? A chip die you didn't design and tape out yourself? What about a library? A framework? An external service or product? I've looked up opcodes on a little insert and hand ent
72.
by philipswood 2y ago
I've had successful chats with ChatGPT that uses Python as pseudocode. I specifically mean pseudocode intended to clarify and communicate - as opposed to be run by a computer. E.g.: https://chatgpt.com/share/67f9fe
73.
by philipswood 2y ago
I recently tried something similar: first I had an in depth chat with Claude in Claude desktop about the system and its requirements and design considerations. Then I asked it to draw up a design doc and component and sequence diagrams and
74.
by philipswood 2y ago
My take is something like the following: With industrialisation blacksmiths were replaced by assembly lines. I'm sure that blacksmiths are more flexible and capable in almost any important dimension, but the economics of factories just
75.
by philipswood 2y ago
> you do actually want that (soft) pushback some of the time. Most modern LLMs will gleefully come up with a complete plan for how we're gonna get the square peg in the round hole. I once accidentally asked a local DeepSeek distille
76.
by philipswood 2y ago
Yes, agreed. It would be good if some spaced repetition was thrown into the mix as well.
77.
by philipswood 2y ago
After watching [Oxford Researchers Discover How to Use AI to Learn Like a Genius]( https://youtu.be/TPLPpz6dD3A?si=FJJ-S6wz0PPrJuSn ) a few days ago, I've been using ChatGPT in "reverse mode" a lot. I give it a
78.
by philipswood 2y ago
He's on to something, but I also find LLM usefulness sharply differs across different languages, tech stacks and problem domains.
79.
by philipswood 2y ago
I think NVIDIAs future is pretty bright. We're getting to the run-your-capable-LLM on-prem or at-home territory. Without DeepSeek (and hopefully its successors) I wouldn't really have a usecase for something like NVIDIAs Project D
80.
by philipswood 2y ago
I think the post arose from: https://news.ycombinator.com/item?id=42886768
81.
by philipswood 2y ago
In Afrikaans we have the (slightly old fashioned): Voor die hand liggend. Something like: It is right in front of your hands.
82.
by philipswood 2y ago
The solar system is mostly a flat spinning disk with everything spinning roughly in the same direction, roughly in a plane. When you find objects in orbits going the opposite direction or not orbiting in the plane or spinning in the wrong d
83.
by philipswood 2y ago
Thanks for sharing. Worth its own submission: https://news.ycombinator.com/newest
84.
Measuring hardware overhang (2020)(lesswrong.com)
10 points by philipswood 2y ago|1 comments
85.
by philipswood 2y ago
AKA: Modern chess software could have beaten Kasparov with contemporarily available hardware years before the Deep Blue match.
86.
by philipswood 2y ago
SaaS is a subscription. Software as an utility is metered, similar to how most cloud costs work. You're billed based on a usage metric. The original cloud vision from Sun and others was called grid or utility computing.
87.
by philipswood 2y ago
So there is: * Software as a possession - pay once to "own" it. * Software as a service - subscribe to it. * Software as an utility - rent it. I'd prefer to see more software as an utility.
88.
by philipswood 2y ago
This vertical is already catered for: Vertu sells expensive phones: E.g. SIGNATURE COBRA LIMITED EDITION KEYPAD PHONE US$504,308.00 They have a concierge service: https://vertu.com/concierge/ Something like ask Siri, b
89.
by philipswood 2y ago
We KNOW that quantum mechanics is incomplete. Everything is definitely not described by quantum mechanics.
90.
by philipswood 2y ago
I look around using a torch and everything I look at, everything I can see, is lit. So the natural and correct conclusion is that all objects are lit, right?

More ›