sterlind

33 ms·

601.
by sterlind 3y ago
archive.is is messed up like that for me too, lately. I have no clue how to work around it, but it seems to be tied to my device rather than IP.
602.
by sterlind 3y ago
Probably more of a general Southern dialect thing than AAVE, but definitely intentional either way. Maybe a reference to Huckleberry Finn but I'm not well-read enough to know.
603.
by sterlind 3y ago
How does this work? I know that LLMs predict the log-likelihood of the next token in a sequence.. so I guess you initialize the LLM from scratch, say that each token of the input text is the Nth most likely possibility, and record what N is
604.
by sterlind 3y ago
Notably, I think this would fail on the Wikipedia compression contest because that counts the size of the decompression program (model in this case), so that it measures Kolmogorov complexity. And RWKV is surely too big. A neat thing to thi
605.
by sterlind 3y ago
seriously! it's literally a Scooby Doo plot. what's next, Shaggy and Velma manage to tie them up, and Fred unmasks them to discover it's an evil developer who wants to scare the locals away?
606.
by sterlind 3y ago
The surge in anti-AI clauses reminds me of robots.txt, albeit with more (gnashing of) teeth. But robots.txt mostly served to entrench Google and Bing, and frustrated attempts to build Common Crawl (a project that's also in the crossfir
607.
by sterlind 3y ago
If package had explosive residue but was not a bomb, then: 1) package is safe, AND 2) either: a) package had residue for harmless reasons, OR b) package was used to transport bomb. conducting additional sweeps rules out case b.
608.
by sterlind 3y ago
SLI isn't used at all for CUDA. if you meant NVLink, it's apparently not useful at small scales - I think the PCIe lanes are enough.
609.
by sterlind 3y ago
how different is the Hurd driver model from Linux's? is it a straightforward but tedious process to port drivers, or does it require major rethinking? I wonder if machine translation is possible, or even LLM-assisted automatic ports as
610.
by sterlind 3y ago
iirc you go to a place that can verify your identity through other means, and they'll update your iris scan or give you a yubikey you can use instead or something.
611.
by sterlind 3y ago
the DEA wouldn't care. tirzepatide isn't scheduled, nor is it a drug of abuse.
612.
by sterlind 3y ago
why can't that someone be you? if you've saved up enough for a home with a garage, can't you further save another $10k?
613.
by sterlind 3y ago
FOGBANK, the classified styrofoam-like material used as a filler in US thermonuclear warheads, had a similar issue. They needed to restart production, but the new samples didn't work! It turns out an impurity present in the original ba
614.
by sterlind 3y ago
Prediction markets seem blessedly free of externalities though, compared to, say, the energy market (CO2) or textiles (child labor, sweatshops.) except for incentivizing action to tilt the odds, which is weirdly amoral. if you bet on a bad
615.
by sterlind 3y ago
> Is it fascinating the risk of a nuclear weapon detonation by December 31st of 2023 is accessed to be around 9%? If a nuke goes off, full-scale nuclear war becomes much more likely. In the event of nuclear apocalypse, your money will
616.
by sterlind 3y ago
generally, insurance companies do require efficacy. many of them have decided not to cover aducanumab for that very reason. though this also gives insurance an additional excuse to deny stratospherically-expensive gene therapies and other l
617.
by sterlind 3y ago
there are a lot of freelance hackers. if companies pay more for bugs than the bad guys, it makes financial sense for hackers to sell bugs to the good guys. $10k seems quite undervalued, and the only reason they're getting any is becaus
618.
by sterlind 3y ago
arbitrary r/w via WebGL, apparently: https://bugs.chromium.org/p/chromium/issues/detail?id=142786... looks like a pretty powerful primitive for only a $10K bounty. what would it go for on Zerodium?
619.
by sterlind 3y ago
on the other hand, people go to MIT to interact with brilliant classmates and faculty. MIT's value proposition is that the smartest people are there, and funding will find those people (and vice versa) on the merits of their intellectu
620.
by sterlind 3y ago
Do you get what you pay for? Are the executives that get paid $50M/year bringing in >$40M/year more than one who only asks for $10M/year? On average? This seems like a question that can be answered with data. My gut feelin
621.
by sterlind 3y ago
My old team used to buy donuts on computers that were left unattended and logged in, as a game to get people to lock their screens. Dvorak alone saved me a couple of times!
622.
by sterlind 3y ago
It's hard to take the idea seriously, but in the interest of honing my steel-manning skills: You'd only want to rate-limit writes to existing files, since reads and creates are harmless. You'd have to rate-limit deletes too,
623.
by sterlind 3y ago
your example of the blind spot is quite elegant and convincing. I think it's partly so convincing because there's a large fossil record and diverse phylogenetic tree, with many gaps covered. conversely, we're missing direct e
624.
by sterlind 3y ago
> Has anyone been banned from the platform? Yes, evidently: https://www.cnn.com/2022/12/15/media/twitter-musk-journalist... > Do you have any evidence that people cannot disagree with him when work
625.
by sterlind 3y ago
his conduct at Twitter, primarily, as he rapidly abandoned his promises of free-speech absolutism and started to micromanage its moderation to benefit himself.
626.
by sterlind 3y ago
are the researchers allowed to disagree with Musk? if the supposedly-objective AI they produce contradicts Musk's views, will he accept that he's wrong? or will he call it a bug, or pressure his researchers into training it to sup
627.
by sterlind 3y ago
Insurance companies can't use genetic data, that's prohibited under GINA.
628.
by sterlind 3y ago
> There is also no biochemical reason why base2 life wouldn't work. are you sure about that? are you sure there's no weird effects that might destabilize very long sequences of 2-nucleotide DNA? or on how wide DNA-binding dom
629.
by sterlind 3y ago
Unironically, this feels right. It's not an error, it's not a success. It's just announcing what it is, a teapot - er, web server.
630.
by sterlind 3y ago
Your summary is incorrect. Step 3 doesn't match what GPT-4 actually said: > When the 7-minute hourglass runs out, don't flip it yet , but note that the 4-minute hourglass has now been running for 3 minutes on its second run. (

More ›