bandrami

10 ms·

211.
by bandrami 5mo ago
And I think it would be an interesting question of discovery to determine if the models were or weren't trained on CSAM. (Diffusors were trained on both pelicans and bicycles, you know...) I still think there's a lot of legal an
212.
by bandrami 5mo ago
This is what I keep coming back to: I also have access to an LLM. Just send me the prompt you used
213.
by bandrami 5mo ago
In the US possession of CSAM is a strict-liability crime. If the models contain the data to produce it, the modelmakers should be liable.
214.
by bandrami 5mo ago
True but they're still committed to providing the compute they're getting from SpaceX to their clients. The cynic in me says this is a numbers game by Elon and Dario to screw up Sam's IPO, along with SpaceX's filing (usi
215.
by bandrami 5mo ago
Along with the "borrow tax-free against unrealized equities gains" hack.
216.
by bandrami 5mo ago
Anthropic is profitable only if you carry forward the discounts they get for the first two months of this deal. That's why they're specifically talking about Q2. They start having to pay $1.6B/mo to SpaceX in July.
217.
by bandrami 5mo ago
Kagi was the best service provider change I've made in years and if anybody cares for an anonymous HN endorsement, they have mine.
218.
by bandrami 5mo ago
Sometime about 20 years ago we all got used to the idea of companies not having to actually make money on their business operations and I worry the bill for this is coming due soon
219.
by bandrami 5mo ago
I think the point of the phrase is that it is obvious but people refuse to talk about it
220.
by bandrami 5mo ago
That will be interesting, to say the least
221.
by bandrami 5mo ago
I am curious if LLMs are better at some kinds of problems than others. IIRC this and another big recent one were cases of the LLM producing a counterexample to a conjecture.
222.
by bandrami 5mo ago
The bigger issue is that the AI we currently have available has been produced by setting a giant pile of money on fire with no real plan for ever earning it back. And that is not sustainable because at some point the checks dry up. We hav
223.
by bandrami 5mo ago
Cool so will they release revenue and expense numbers in line with GAAP?
224.
by bandrami 5mo ago
Cool does this mean they'll disclose their revenues and expenses in line with GAAP?
225.
by bandrami 5mo ago
> that's already a huge saving to any company's bottom line Possibly not if they are paying the full cost of inference
226.
by bandrami 5mo ago
But Anthropic is adding employees
227.
by bandrami 5mo ago
I think it's "molten" while it is still liquid but not once it crystalizes; at that point it's "melted"
228.
by bandrami 5mo ago
There was a neat helicopter landing game written in prolog way back in the day
229.
by bandrami 5mo ago
You could but the other guy would have to book 100 of goodwill in the interim, matched by me booking goodwill later, and that brings it's own problems
230.
by bandrami 5mo ago
Somehow my EMP blast idea is the less intrusive one...
231.
by bandrami 5mo ago
Faraday cage. EMP blast if that doesn't work.
232.
by bandrami 5mo ago
The procurement and certification processes we go through are basically specifically designed to keep a scrappy startup with the next new idea from ever winning a contract without significant institutional buy-in, for reasons that will prob
233.
by bandrami 5mo ago
Oh absolutely. The level of antisocial behavior the US tolerates compared to multiple other countries (developed and developing) I've lived in is astounding.
234.
by bandrami 5mo ago
For me the much bigger problem is the data (and God knows what else) going to a third party. But yeah the non-repeatability doesn't pass the DoD audits either.
235.
by bandrami 5mo ago
Yeah I'm realizing now how many of you guys work in industries with no data security/protection requirements
236.
by bandrami 5mo ago
As somebody who's spent most of my career in Fairfax County I find nothing about this story even remotely surprising.
237.
by bandrami 5mo ago
I'm not a big "lock them up" guy but seriously people don't seem to understand how hard it is to actually get the state to put and keep you in jail. You have to do really really bad things multiple times. The US prison p
238.
by bandrami 5mo ago
Isn't that exactly why MS ties O365 to OneDrive like they do?
239.
by bandrami 5mo ago
This anecdote may be the key to the productivity mystery of LLMs: the output is appearing in the form of software that wasn't ever economical to make before so kind of by definition won't appear in productivity metrics. Though at
240.
by bandrami 5mo ago
I wasted most of my high school years on the OG (1991) version. I love how such a simple concept can make for such a great game

More ›