tlb

32 ms·

511.
by tlb 3y ago
Pop music contains 10^5 or so versions of “I love you”. I think there could be a similar number for writing.
512.
by tlb 3y ago
As usual, thought leadership is mainly marketing for the business. He made it work well for that purpose. Everyone knew that it was the shoe warehouse with happy employees. It was a clever theme. Talking about shoes wouldn’t have been as co
513.
by tlb 3y ago
If any time a proponent of a philosophy has a bad life you declare the philosophy failed, there must be very few left.
514.
by tlb 3y ago
The placebo here was saline (no physiological effect). Sometimes they give amphetamines as a placebo so the control group also feel that something is different. But I’d expect that to affect RL hyperparameters too.
515.
by tlb 3y ago
The real morality play here is to publish your work. I suggest that Shepard had a net-negative effect on the field. Anyone who might eventually have done the same research would have heard about Shepard's work and been discouraged from
516.
by tlb 3y ago
One surprising thing is that MS can be receiving an API call per web page from all Edge users, without noticing that they have a vast increase in traffic. It must be many times more requests than Bing itself. That must have required some se
517.
by tlb 3y ago
Interruptions are costly to both programmers and writers. But I don't think writers get interrupted as much, so they can work during normal waking hours. Programmers get interrupted a lot if they work during business hours because thei
518.
by tlb 4y ago
Not what you're asking, but tr a-z A-Z </usr/share/dict/words | grep '^[AHIMOTUVWXY]*$' gets you a subset of English. You can say things like AHOY OTTAWA AUTOMATA! YUMMY HAM: MOUTH VOMIT. MAXIMUM
519.
by tlb 4y ago
The moral position is against wearing fancy clothes that most can’t afford. Clothes are cheap now, but a modern parallel is to take a moral position against flying private jets.
520.
by tlb 4y ago
There’s a little more to it than that. Matrices have several properties like being symmetric or unitary (maybe even diagonal) that scalars don’t. Those enable rewritings to make systems more stable or computationally efficient.
521.
by tlb 4y ago
Yes, in some sense the recent LLM results show us how big the Chinese room has to be: enough for 10^11 parameters. If the index cards Penrose imagined could hold one column of a matrix, it's around 10^8 cards. It's not surprising
522.
by tlb 4y ago
I haven't heard of that being tried (though I don't read everything.) Someone could do the experiment and write it up, and maybe get it published. The main ML conferences rarely publish anything that's not an improvement on t
523.
by tlb 4y ago
If you make N copies of a model, train them independently for a little while on N machines, and average them back together, it sort of works. But not if you train for very long, as the internal structure diverges. It becomes an empirical en
524.
Emergent abilities of large language models(jasonwei.net)
157 points by tlb 4y ago|190 comments
525.
by tlb 4y ago
Indeed, it fails if you invoke UB. For example, an int overflow in a constexpr causes a compile error: <source>:5:17: error: static assertion expression is not an integral constant expression static_assert(1024*1024*1024*3 !=
526.
by tlb 4y ago
The best silicon process for DRAM is significantly different than for CPU logic. So it's not practical to combine them on one die -- you'd have to compromise the speed or density of one or the other. But the chiplet approach works
527.
by tlb 4y ago
That can work well for many programs, but no one has succeeded in writing a multithreaded kernel that way.
528.
by tlb 4y ago
Most of the YC-funded companies that have done a Launch HN ( https://news.ycombinator.com/launches ) in the last couple months will be raising money now. Pick a few whose products impress you and ask if they're open to n
529.
by tlb 4y ago
I think an AI can learn everything it needs from books and other text. But others claim that it can never understand qualia like the feel of grass between your toes or the sensation of the color red without some direct experience. Does an A
530.
by tlb 4y ago
Predict log(votes + 1) instead.
531.
by tlb 4y ago
And by all objective metrics, there are many $500 sound systems that sound as good as any system at any price. You can't be both objective and high-end.
532.
by tlb 4y ago
I mean, most dictators didn't "do" much. They just said things and gesticulated dramatically and convinced other people to do things. Perhaps a body is necessary to have massive psychological effects on people, but we don&#x2
533.
by tlb 4y ago
It seems correct, but verbose. GPT-4 gives the more concise and idiomatic: module full_adder(input [3:0] a, input [3:0] b, input cin, output reg [3:0] sum, output reg cout); always @(*) begin {cout, sum} = a + b + cin; end
534.
by tlb 4y ago
Conspiracy theory: this idiom used to be widespread on the internet, but a shadowy cabal of AI maximalists erased millions of such expressions from search engines so GPT could appear to re-invent them. It is a good expression though -- evoc
535.
by tlb 4y ago
There could be (and probably is) some little app that sets good defaults in Windows and turns off all the abuseware. But who would trust such an app? At some level, the problem is that any company big enough to build a trusted brand is goin
536.
by tlb 4y ago
GPT does encode position information in its training data (an idea going back to Vaswani's 2017 paper). But indeed, it can get it wrong if it hasn't seen contrasting examples where different order means something different. In thi
537.
by tlb 4y ago
No, you can instruct ChatGPT to respond as a liar. If you wanted, you could build a clone that did this implicitly by having the instruction hidden.
538.
by tlb 4y ago
The bad guys would also have to have MITMed it every time I connected for the last 15 years, or I would have seen authentication failures when it connected to the real thing. MITMing someone once isn't that hard, but doing it consisten
539.
by tlb 4y ago
As I understand it, the distance over which the increased field gradient from a sharpened rod is greater than the variation in field gradient due to rain & wind is pretty small. Once the discharge has ionized the air to within several m
540.
by tlb 4y ago
High-speed cameras are fascinating to observe things with. I use them a lot for robotics, but almost any physical phenomenon has a great deal of detail only visible at higher speeds than the eye is capable of. Your phone camera can probably

More ›