peterdsharpe
peterdsharpe on Hacker News
This would have been true until ~50-200 years ago, and the advent of modern capital markets. Now, credit is center stage and cash is the sideshow.
Say I'm a lab.
I train a new model. I publish my training data and my training code/workflow.
I then secretly do additional post-training to poison it. I do not release the data or workflow for this.
Now, I publish the poisoned weights and an md5 hash of the poisoned weights.
To you, the user - you have no way of knowing this happened.
You might ask "why not publish intermediate training checkpoints" but those are nondeterministic (in the sense of being not bitwise-reproducible). Anything with shuffling, random initialization, distributed races, driver versions changing roundoff, compiled vs. eager, and a million other factors introduce slight differences that yield totally different training trajectories. So publishing checkpoints does not yield verifiability.
Nope. keepandroidfree.org
This is not a constructive way to engage with people. Please reply in good faith on HN - there are other platforms if you prefer otherwise.
$5 does not make sense, and there must be some qualifer on that we're missing.
Napkin math:
(27 minute flight) * (25000 lbm TOGW) * (Earth gravity) / (~10 average L/D) * (160 KTAS average) / (70% propulsive efficiency) / (97% coulombic efficiency) * ($50/MWh) = $30
(Add residual kinetic energy upon landing too)
Side note: Heart also has a weight issue here they'll need to overcome if they want to certify under FAR Part 23 (as their 19-pax spec implies); TOGW limit for that is 19,000 lbm.
There is a word for that. It is "heaviest", not "largest".
`uv sync --upgrade`? It's in the docs
Why use this over Equinox's Module, plus jaxtyping? That seems like the cleaner (and more established) solution, but maybe there are pros/cons I'm not seeing.
Death of the AI race? This is the starting gun.
"Financially, what will take me to $1B?" -Greg Brockman, August 2017
Seriously going to use AI to write a reply that would take you 30 seconds?
Thank you for the kind words! :)
This is a desperation move from the LLM providers here. Just think: if you were Sam Altman, and you actually had strong evidence that practical AGI was as imminent as claimed - would you be spending your time building AI-slop TikTok and a smut bot?
Please don't post paywalled articles to HN.
There is a beautiful irony in your misspelling of Anno Domini as Anno Domani, for the Italian speakers in the room.
What is the benefit of this over `watch nvidia-smi`, possibly prepended with an `ssh` in the case of a remove server?
I love the concept of this and hope Slate survives, but $27.5k is too expensive for what they're offering. With the Kia Soul at $21.5k and with more features, and the Honda Civic at the same cost, the value proposition isn't there for many people.
Minimialism is great, so long as the price also reflects that.
This was a disappointing article: based on the provocative title, I was hoping to learn something new. Everything in here is already clear to anyone who has read the "Getting Started" page of uv, and I don't know anyone who is making these mistakes (system-wide package installs on a venv-first package manager?) with uv.
Yes, it is completely wrong. If this were a valid explanation, flat-plate airfoils could not generate lift. (They can.)
Source: PhD on aircraft design
Funny you should mention that! The surface of a trampoline can actually be modeled as a 2nd-order wave equation PDE, which uses a Laplacian spatial kernel - for the same reasons this is called Laplacian smoothing!
You imply that the title "King of America" is pejorative, but did he or did he not refer to himself as a king? As far as I can tell, he endorsed this title.
To add on this, prediction markets currently put Trump Sr. as 8.5% likely to win the 2028 GOP nomination (electionbettingodds.com). So, I wouldn't take your "he'll be done in four years" as certainty. The market thinks things are far more precarious than you do.
> when it stops being laminar the flow separates
Not true, laminar/turbulent and attached/separated are independent descriptors. Flow can be laminar+attached, laminar+separated (e.g., laminar separation bubbles), turbulent+attached (e.g., the majority of boundary layers on large aircraft), or turbulent+separated.
> literally no errors in its operation are tolerated
Aircraft designer here, this is not true. We typically certify to <1 catastrophic failure per 1e9 flight hours. Not zero.
Disagree. The kind of serious plagiarism that's alleged in cases like this, Claudine Gay, Marc Tessier-Lavigne doesn't happen by accident, it's deliberate. There are plenty of academics, both minorities and not, who are capable of original thought (and not resorting to plagiarism).
I can't wait for the day that AI is used to expose the fakers en-masse. Previously, they got away with plagiarism through obscurity - no longer. The sooner we can excise these freeloaders from our professional institutions, the better.
I did not go though the grueling process of writing an original PhD thesis, just to share a title with those who plagiarized their thesis.
Solar can absolutely be modulated; in fact, it's perhaps the easiest grid-scale power source to modulate since this can be done solid-state: just tune the MPPT to operate off-peak if needed.
This is not how unit conversions work.
This is exactly the wrong time to use a LLM. Case in point: those coordinates are not in the Atlantic ocean.