DennisP

5 ms·

61.
by DennisP 2mo ago
And yet, when I was a developer and the only way to automate things was by hand, I found that doing so could give me lots of extra free time.
62.
by DennisP 3mo ago
From the other comment: > this was partially tested last year in Thaler v. Perlmutter which affirmed that copyright requires human authorship Model weights are not authored by humans. Neither is AI output, which is why that's not co
63.
by DennisP 3mo ago
Normal software is copyrighted. Model weights don't seem to be copyrighted, as another comment pointed out here: https://news.ycombinator.com/item?id=49073127 If weights aren't copyrighted then there's no IP
64.
by DennisP 3mo ago
Sure, that's a plausible theory but I haven't seen that anyone's proved it. An alternate theory is that models need lots of input knowledge to learn complex reasoning, but don't need so much at inference time. An example
65.
by DennisP 3mo ago
I think the idea here would be to not use your super smart but specialized model to check the weather. It's not obvious that it's impossible to (eg) remove most of its biology knowledge, without removing much of its ability to dev
66.
by DennisP 3mo ago
For software, not only that, but run batches with a cluster of agents working different parts of the same task. Software like Yegge's Gas Town has one agent act as "mayor," and others work on writing or testing various pieces
67.
by DennisP 3mo ago
That last step contradicts my previous comment.
68.
by DennisP 3mo ago
Of course advertising increases their profit. It does that by increasing their revenue even more than the cost of the advertising. For the rest of us, that giant increase in revenue is an increase in our healthcare costs.
69.
by DennisP 3mo ago
Unless we came up with a different funding mechanism. Joseph Stiglitz for example has advocated a prize system for pharmaceuticals, though he doesn't suggest replacing the patent system entirely. https://www.project-syndicat
70.
by DennisP 3mo ago
So pharmaceutical companies spend far less on marketing outside the US, partly because every other country besides New Zealand makes those incessant drug ads illegal, and partly because governments negotiate prices and keep profit margins d
71.
by DennisP 3mo ago
So doing it on a big GPU would be pretty helpful. I wonder if anyone has used CUDA for this, or if any hardware synths have GPUs.
72.
by DennisP 3mo ago
Vitalik recently wrote about vibe-coding in Lean and assembly language, using proof verification in Lean, saying "if done right, this has potential to both output extremely efficient code, and be far more secure than the way programmin
73.
by DennisP 3mo ago
The article says Kimi is better at some things than Fable. That's probably not true of Sonnet.
74.
by DennisP 3mo ago
It'd be interesting to see how that works in various languages, because there may be some selection bias. Most English-speaking countries have decent protections for all but the most abhorrent speech, and people who don't find the
75.
by DennisP 3mo ago
Yep at this point we just have to do whatever seems least risky. It's not like we have a pristine planet. We have to balance the unknown consequences of phytoplankton (and other mitigations) against the known and unknown consequences
76.
by DennisP 3mo ago
768GB next quarter. The chip supporting 1.5TB won't come out until 2028 or 2029.
77.
by DennisP 3mo ago
What do you mean by "clean-rooming your prompts?"
78.
by DennisP 3mo ago
If the government provided funding to independent organizations that made public-goods AIs, that could be great, as long as the government had no editorial control. It sounds like he's imagining AIs only being trained and provided by g
79.
by DennisP 3mo ago
As a biological human I am glad that Google is applying AI to biodefense.
80.
by DennisP 3mo ago
But if Anthropic had just bought the authors' books instead of pirating them, then they would have paid nothing.
81.
by DennisP 3mo ago
Anthropic paid a large settlement for the copyrighted data they pirated. So far, US courts have found that it's perfectly fine to train AIs on copyrighted data for which you have legal access.
82.
by DennisP 3mo ago
A lot of mortgages were trash but the securities were a new and complicated financial product with a plausible story. A few people figured out the problems, but Michael Burry for example did it by combing through thousands of pages of prosp
83.
by DennisP 3mo ago
Some people actually do track their cars, though.
84.
by DennisP 3mo ago
I'm old enough to remember they weren't just buying and selling corporate bonds like most of the bond market. They were a hedge fund that used massive leverage to exploit tiny arbitrage opportunities between correlated securities.
85.
by DennisP 3mo ago
Institutional bond traders are pretty sophisticated. They're aware of all the possibilities you mentioned, and pricing the bonds low anyway.
86.
by DennisP 3mo ago
At this point I can only ask that you actually read my above comments more closely and think about them. I have exactly the level of security I described in my last comment, with complete anonymity, all with one design. If your claim is tha
87.
by DennisP 3mo ago
The only requirement I dropped is absolute perfect security, which no system will have and certainly not the nonsense "show your ID to the camera" stuff they're actually using. I showed how to eliminate proxies, and without t
88.
by DennisP 3mo ago
How? The websites don't see anything consistent from one session to the next. They can't tell when the credential is renewed, and don't see a credential id.
89.
by DennisP 3mo ago
You can fix the leaked token problem if your prover also proves that (a) the private token id is not on the public revocation list, and (b) the token has not yet expired. Use short expirations and auto-renew, this is just to keep the revoca
90.
by DennisP 3mo ago
Ideally, but in practice only if the bits are uncorrelated and each divides the population in half.

More ›