agnosticmantis

24 ms·

31.
by agnosticmantis 4mo ago
I believe sycophancy is a side effect of RLHF and whatever reward function it explicitly and implicitly optimizes.
32.
by agnosticmantis 4mo ago
I haven't done it yet due to privacy concerns, but I would totally do it with a private local model that's as intelligent as current frontier models and is not sycophantic and perhaps is finetuned on the psychology literature. My
33.
by agnosticmantis 4mo ago
How would these compare to a MacBook Pro M5 in terms of performance and price?
34.
by agnosticmantis 5mo ago
> If AI is good enough that it can boost productivity by 20% then it is good for society in general because the gains will be redistributed (as it has always been) This redistribution has never been as automatic and inevitable as you see
35.
by agnosticmantis 5mo ago
Both can be true: insiders bet on a sure thing, outsiders naively gamble, the house gets a cut no matter what.
36.
by agnosticmantis 5mo ago
> Ideas like a linear timeline progressing from A to B (rather than a cyclical one) Seriously? So we’re either linear Christians or circular Hindus, and nothing else ever existed? And nobody else believed in linear time?
37.
by agnosticmantis 5mo ago
Likely because Grothendieck was a vegetarian, at least later in his life.
38.
by agnosticmantis 5mo ago
It may not be mainly or solely due to LLM pollution, but rather the fact that every publisher, (social) media company, newspaper, etc. clammed up and started charging (licensing) fees sometime in the last couple of years. So maybe there&#x2
39.
by agnosticmantis 5mo ago
> I lived through the great MTBF vs MTTR (mean-time-between-failure vs. mean-time-to-recovery) reckoning of infrastructure during the transition to cloud and cloud automation. What's the historical context for this MTBF vs. MTTR rec
40.
by agnosticmantis 5mo ago
Willing to bet they'll lobby for regulatory capture and raise the drawbridge for the little guys.
41.
by agnosticmantis 5mo ago
Now discuss whether a bonobo, a dog, a cat, a mouse, an ant, a bacterium is conscious. And you’ll find it’s not as clear cut.
42.
by agnosticmantis 6mo ago
My $0.02: Competing against exp(t/2) + exp(t/2) is much much easier than exp(t/2+t/2)=exp(t). (If anthropic didn't exist, ØpenAI would suck up all the capital and talent in the room. Anthropic's existence has
43.
by agnosticmantis 6mo ago
"You could see a single robotaxi being worth, or providing, about $30,000 of gross profit per year. ... A Tesla is an appreciating asset..." - Elon Musk during Tesla's Autonomy Day in April 2019.
44.
by agnosticmantis 8mo ago
> Checkpoints are a new primitive that automatically captures agent context as first-class, versioned data in Git. When you commit code generated by an agent, Checkpoints capture the full session alongside the commit: the transcript, pro
45.
Reza Pahlavi: Iran is ready for a democratic transition(washingtonpost.com)
3 points by agnosticmantis 9mo ago|2 comments
46.
by agnosticmantis 9mo ago
Archived (non-paywall) version: https://archive.is/BObYG
47.
by agnosticmantis 9mo ago
One would have to control for all the layoffs, tarrifs, interest rates etc too. On the other hand, people are working much harder today than 3 years ago (remember people not showing up to work and posting on TikTok about how little work did
48.
by agnosticmantis 9mo ago
I’ve started experimenting with Quarto[0] for scientific/technical publishing on a personal website, and it’s been quite easy to use so far. I especially like that it has builtin support for LaTeX, markdown, code blocks and Jupyter not
49.
by agnosticmantis 9mo ago
He shamelessly advertised companies in his portfolio, whether it was always having a can of coke or candies on the table or stopping at a McDonald's in a documentary. I refuse to believe that his lifestyle was what was on display. He l
50.
by agnosticmantis 10mo ago
I find it hard to see an upper bound to demand for compute/power. If models/hardware become 1000X more efficient/cheaper/smaller, people will be running 1,000,000X more jobs with even bigger models on even more intensive
51.
Comparison of Waymo Rider-Only crash rates by crash type to human benchmarks(tandfonline.com)
2 points by agnosticmantis 10mo ago|0 comments
52.
by agnosticmantis 11mo ago
This says Gemini 2.5 though.
53.
by agnosticmantis 11mo ago
The Pipe Query Syntax in GoogleSQL implements this elegantly as well: https://docs.cloud.google.com/bigquery/docs/reference/standa...
54.
by agnosticmantis 1y ago
> Once AGI is declared by OpenAI, that declaration will now be verified by an independent expert panel. So OpenAI will declare AGI as soon as ChatGPT is a better AI lawyer than any Microsoft could hire.
55.
RFC: Evolving PyTorch/XLA for a more native experience on TPU(github.com)
3 points by agnosticmantis 1y ago|0 comments
56.
by agnosticmantis 1y ago
This looks very cool, I’m gonna try it later today. Out of curiosity, why the name Erdos? AFAIK Erdos was neither a statistician, data scientist nor AI researcher. He sure solved many probability/combinatorics problems and famously had
57.
by agnosticmantis 1y ago
"His decision seems to have been based less on his famous scientific thinking and more on a very personal feeling. He wanted to retain his autonomy until the end and to shape his own end." So you could say it was more system 1 thi
58.
by agnosticmantis 1y ago
Only difference is soon you'll have Mickey Mouse tell you (and your children) to buy same old crap that the influencers were shilling.
59.
by agnosticmantis 1y ago
Business idea: disrupt all these apps by creating an agent that watches all the brainrot for us so we all get our stolen attentions back.
60.
by agnosticmantis 1y ago
How about the donor piggy though? If this becomes commonplace and animals are bred/raised just for their organs, we get into murky ethical territory. (Yes, people already eat the organs for food, that's murky too. But industrial s

More ›