guy98238710

12 ms·

91.
by guy98238710 3y ago
Recursive training of generative models degenerates into reinforcement learning with random feedback. You need strong feedback signal in the loop to survive recursive training. The feedback does not have to come from humans though. You can
92.
by guy98238710 3y ago
I think the main error of the author is to assume that short content equals low quality content. There's value in conciseness.
93.
by guy98238710 3y ago
The guy starts with a clickbait title and then proceeds to present a lecture on content quality.
94.
by guy98238710 3y ago
Plant-based food is actually cheaper, at least the ordinary stuff like bread, pasta, legumes, oat flakes, and even the hated sugar. And so is public transportation where available.
95.
by guy98238710 3y ago
The core three tenets are drive encryption (easy), regular wipes (10h/year), and sandboxing (crazy expensive at the moment). That gives you decent security.
96.
by guy98238710 3y ago
Most people don't have time to study stuff like this. Even if they do, it's hard to tell whom to believe. Cost of pollution must be factored into products. That will change behavior.
97.
by guy98238710 3y ago
I think the main limitation is currently in Flatpak sandbox. Lots of apps assume control over the whole computer (IDEs, Rust, Android tooling). Lots of apps still rely on X11, which allows trivial sandbox escape. And then a lot of apps just
98.
by guy98238710 3y ago
I am nuts into linux desktop security, but I cannot imagine recommending this to even other developers, let alone non-dev users. There are just too many usability tradeoffs, the whole thing is a major time waster, and you have to deal with
99.
by guy98238710 3y ago
Artificial intelligence is a cognitive augmentation tool. It makes people smarter, more competent, and faster. That cannot be. Intelligent people are dangerous people! Just consider some of the more sinister hobbies of intelligent people: -
100.
by guy98238710 3y ago
More like intelligence == human. ChatGPT is superhuman in many ways.
101.
by guy98238710 3y ago
Personally, I wouldn't mind paying for the service as business user if the free accounts were worth recommending to people I know. Just yesterday I found out that my son's free account does not let him search emails older than 1 m
102.
Tutanota Is Raising Prices(tutanota.com)
1 points by guy98238710 3y ago|4 comments
103.
by guy98238710 3y ago
The underhanded way SE has handled this is a problem, but deleting AI-generated answers is silly. They should instead ask SE Inc. to automatically provide one AI answer for all new questions. Nobody wants to answer the same trivial question
104.
by guy98238710 3y ago
But isn't that the advantage of having ChatGPT in the ecosystem? Who wants to moderate a 1000th variation of the same trivial question? Let ChatGPT handle the easy cases and leave the hard stuff for SO. Perhaps SO management does not
105.
by guy98238710 3y ago
Proposed as a wiki project: https://meta.wikimedia.org/wiki/Wikiask/Introduction It never got anywhere. If it ever does, I would consider being a contributor.
106.
by guy98238710 3y ago
Once you have certain degree of automation, the hard-working people just drag everyone into pointless rat race.
107.
by guy98238710 3y ago
The stones must be good reflectors of noise, which makes the ballast rather undesirable under trams in cities. I wonder why is it still used in such circumstances instead of having noise-absorbing soil and grass under and around the tracks
108.
by guy98238710 3y ago
You know, it's not just labor that contributes to economic output. Natural resources and accumulated knowledge and technology contribute too. Both belong to everyone with equal share. Giving people what their share of resources and tec
109.
by guy98238710 3y ago
Huh, UBI has its own moral panic. Never thought I would see that, but it seems obvious in retrospect. Both capitalism and communism worship work like some insane religion. If you work, you are a good person. If you don't, you are scum.
110.
by guy98238710 3y ago
The only risk with AI is that it will be abused by the wealthy and the powerful, especially autocrats, who will no longer need labor, only natural resources. Hence the solution is to promote worldwide democracy and public ownership of natur
111.
by guy98238710 3y ago
It's not just university students anymore. My 11yo kid got accused of being a robot on a physics competition where the only reward is a (paid) summer camp full of extra physics lessons. All that was needed to trigger the accusation was
112.
by guy98238710 3y ago
Is that really a bias? There are more female students. Not sure about professors, but I would guess there are (or at least recently have been) more male professors. Sounds like correct incorporation of related knowledge.
113.
by guy98238710 3y ago
The problem is still the same as it was with the cookie law: Should we regulate capability or intent? Regulation by capability is easy to specify in the law, but it cripples useful technology as much as predatory technology and thus ends up
114.
by guy98238710 3y ago
The low hanging fruit theory assumes there is only one tree and all fruit on it was visible from the beginning. Science has an exploration component that makes more low-hanging fruit visible and scaffolding component that makes existing fru
115.
by guy98238710 3y ago
This is an illuminating point of view. Science is weak-link on its output, because of the downstream damage it can do. But article's claim still holds on the input side of science - sorting ideas for funding and perhaps publishing earl
116.
by guy98238710 3y ago
It's a mistake to just protest these proposals. Counter-proposals that push the balance in opposite direction should be made and pushed through, including some that cut down on existing legislation. A lot of good arguments can be made
117.
by guy98238710 3y ago
The proposal claims to be proportionate, but I don't see any mention of minimum income threshold or even a broad "significant revenue" statement in the proposal. Are hobby developers who make 100€ per year going to be exposed
118.
by guy98238710 3y ago
I like to tell people I am a pirate though. I think it sounds cool.
119.
by guy98238710 3y ago
Why would you train the model on shorter context than you can provide? Why not provide all context you have? Sure the model has to learn to handle short context, but that occurs naturally at the beginning of the document. Anyways, this stil
120.
by guy98238710 3y ago
There's one aspect I never saw explained. Why is masking used instead of a sliding window? Why even bother with masking when future tokens can be easily hidden by simply positioning the context window before the current token? Isn&#x27

More ›