Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
NiloCK
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
6 ms
·
61.
▲
by
NiloCK
4mo ago
What do you mean with this? I doubt there is any large demographic of users paying subscription fees for the joy of abusive role play.
62.
▲
by
NiloCK
4mo ago
A great many today find themselves surrounded by people staring at phones, who express irritation any time they are asked to look up. I saw a video a while back on one social media site or another where someone sitting in a car recorded thr
63.
▲
by
NiloCK
4mo ago
Like it or not, the "merge request" (eg, open a PR) is the Schelling point of relevant information. I expect that At scale here refers to size of software projects, and not only code velocity. Software projects of large enough s
64.
▲
by
NiloCK
4mo ago
No no no no no. Big misunderstanding here. We just meant in the city of Cannes.
65.
▲
by
NiloCK
4mo ago
I appreciate the generosity, but you're gonna want to meet me first.
66.
▲
by
NiloCK
4mo ago
I think it's telling how split the opinions are around all of this. A lot of people distinctly disliked 4.7. Are the dividing lines around personality? Working domains? Opinionated software stuff? Who knows?
67.
▲
by
NiloCK
4mo ago
A rambling comment: I think this is the first time we've had a third minor version bump on a frontier Anthropic model. (I count the 0.5s as major here, because they've been issued non-sequentially and also corresponded to massiv
68.
▲
by
NiloCK
4mo ago
I'm mostly unfamiliar with the current offering of last.fm, but the name is familiar from way back. Glad to see something well-liked reclaim some independence. At a glance, they're providing an interface to YT sourced content with
69.
▲
by
NiloCK
4mo ago
Ahh - the dependency graph recipe card. These are excellent. I've imagined something like this forever. Always annoyed that recipes put ingredients in a giant undifferentiated list and then give an instruction like "mix the dry in
70.
▲
by
NiloCK
5mo ago
The thing that drove me away from manual edits was that I found myself confusing the LLM all the time. It would read or write, some code, I'd twiddle with things, and then the LLM's future references to the same code would be a me
71.
▲
by
NiloCK
5mo ago
80-90% reduction, over the course of 170 million miles driven on the famously very controlled city streets of LA, SF, Austin and Phoenix. On average, I wouldn't expect the regulatory agencies to be very friendly toward outright fraud
72.
▲
by
NiloCK
5mo ago
It's great that you believe this, but are you hiring? I don't intend this to read as pure snark, but someone's abstract value isn't much good to them if the job market itself can't / won't recognize it.
73.
▲
by
NiloCK
5mo ago
Over a given driving distance, compared to humans, Waymos produce a 90% reduction in serious injury, 90% reduction in pedestrian strikes, 83% reduction in airbag deployments, 85% reduction in cyclist strikes [1]. We currently sit in the bal
74.
▲
by
NiloCK
5mo ago
It's about many things, including reaction speed, visual awareness, specific expertise and informed decision making wrt braking or acceleration power. All of these are better in a modern self-driving car (I do not know whether Tesla fa
75.
▲
by
NiloCK
5mo ago
Doesn't it benefit me if the models I use improve?
76.
▲
by
NiloCK
5mo ago
Maybe you're sitting pretty right now, but try posting this from your deathbed, or that of your kid. The lack of compassion that people display here is shocking to me. "Don't automate science, because there are junior scienti
77.
▲
by
NiloCK
5mo ago
This is wonderful. Having models attempt an SVG letter S remains one of my personal/informal LLM benchmarks. They are still pretty bad at it.
78.
▲
by
NiloCK
5mo ago
What do you think of things like the changes to infant mortality or life expectancy between the industrial revolution and present day? EG, my own oldest child needed a surgery at birth that would have been logistically impossible even 50 ye
79.
▲
by
NiloCK
5mo ago
I find it forgivable if it's within minor version bump. (NB that x.5 is now a defacto major-version bump for LLMs for whatever reason). Even with LLMs, posts like this don't just fall out of a coconut tree. If you have a set of ta
80.
▲
by
NiloCK
5mo ago
A stan is a supporter/booster of whatever . I do not remember the origin. Pol here is abbreviated politician .
81.
▲
by
NiloCK
5mo ago
I'd say that there is no such statistically significant data. Practically nobody teaching K-12 has subject-matter masters degrees. It's just not part of the career trajectory. As unusual as a nurse having an M.A. in history or som
82.
▲
by
NiloCK
5mo ago
> No, it is emphatically not. D Fraud requires intent to deceive. I'm about as pro AI-as-a-research--and-writing-assistant and anti AI-witchhunt as they come, but I simply cannot parse what I've quoted here. Posting slop to arx
83.
▲
by
NiloCK
5mo ago
I don't think the product people and the RSI -adjacent people are the same people.
84.
▲
by
NiloCK
5mo ago
I grant that there's a definition of abstraction that LLMs don't fall into. But people describing LLMs as another abstraction layer aren't all misunderstanding this. Instead, they are using the term ... more abstractly. EG:
85.
▲
by
NiloCK
5mo ago
I've already posted a couple of times here but I'm pretty jazzed with this publication. Some thoughts: 1. It's amazing how strong the obvious in hindsight is for this research. LLMs have been (rightly) characterized as insc
86.
▲
by
NiloCK
5mo ago
Notable here that the training run didn't have access to the 'plaintext' context that the LLM was working in. It'd be quite a coincidence if the training runs discovered an invertible weights>text>weights function
87.
▲
by
NiloCK
5mo ago
You've never learned about a service or product from a TV ad?
88.
▲
by
NiloCK
5mo ago
Future model training runs will have a copy of this research, and know "to defend against it". EG, could a misaligned model-in-training optimize toward a residual stream that naively reads as these ones do, but in fact further enc
89.
▲
by
NiloCK
5mo ago
Are the training arenas for the Activation Verbalizer and Activation Reconstructor models well described here? If they are co-trained only on activationWeights->readibleText->activationWeights without visibility into the actual
90.
▲
by
NiloCK
5mo ago
Humanity!
More ›