srcreigh

28 ms·

451.
by srcreigh 4y ago
In UWaterloo CS there's at least two database courses. One covering SQL/querying/normal forms, another looking at how DBs are implemented. They are not required courses. I agree with GP that DBs are not fundamental to CS in a
452.
by srcreigh 4y ago
Enter the formula on row 1, copy it, select the column and paste
453.
by srcreigh 4y ago
Antibiotics, poor diet over time. I wonder if alcohol has any effect too.
454.
by srcreigh 4y ago
Stanford or MIT or UWaterloo
455.
by srcreigh 4y ago
OpenAI already has access to any prompts anybody uses to write programs using GPT. Who’s to say GPT-4 isn’t some ploy to gather data to train private AGI?
456.
by srcreigh 4y ago
In 2008-2010 there was something that lit up the night sky like day over haldimand county in Ontario Canada
457.
by srcreigh 4y ago
Physics has plenty of holes. It’s a changing model of measurable quantities in our world. It doesn’t take a lot of brainpower to fit other systems either into the gaps in physics, or independent unmeasurable areas
458.
by srcreigh 4y ago
I do. It’s not a very controversial belief except if you only train using Internet comments.
459.
by srcreigh 4y ago
Yep. Another commenter said that likes and clicks is enough for AI to generate new relavant information, not new content. Think language change every 1000 years. Would AI adapt to that? What new training data would it require? Surely obviou
460.
by srcreigh 4y ago
The key difference is that we know how LLMs work but we still haven’t solved the problem of consciousness, or even neuroscience. You can’t equate something you understand with something you don’t understand. Something you can create with so
461.
by srcreigh 4y ago
Any automated process for generating training data is part of the model, not part of the training data. Authentic human generated information has irreducible newness.
462.
by srcreigh 4y ago
I wasn’t reader user, but it seems to me like when advertising self interest took over
463.
by srcreigh 4y ago
It’s kinda like prisoners dilemma, except the prisoners can meet and discuss a cooperative strategy.
464.
by srcreigh 4y ago
It’d be cool to parse the JSON using the same memory holding the string. Intern the object keys and it should fit
465.
by srcreigh 4y ago
There was a thread this week (?) about how AirBNB will ban you for traveling with a banned person
466.
by srcreigh 4y ago
The reusable dog poop bag is probably a bucket. Maybe smaller one with a lid.
467.
by srcreigh 4y ago
Haven’t tried others, but I really like boon. It’s so simple and it has the verbal actions like delete I’ve patched the boon code to support switching between Dvorak and QWERTY
468.
by srcreigh 4y ago
It’s mind blowing how many of these takes ignore national intelligence/security. As if NSA spying on Americans is seriously the same as the Chinese government spying on Americans. Enacting data privacy laws is touted as a solution. Is
469.
by srcreigh 4y ago
The guy, Seamus Blackley, a game programmer who proposed the original Xbox, no less. Also made Trespasser (1998) realistic open world 3D game with physics engine [1] [1]: https://www.youtube.com/watch?v=OSDa4VZbJzI
470.
by srcreigh 4y ago
So if a new model is trained with 10x parameters of gpt3, is gpt now no longer an llm?
471.
by srcreigh 4y ago
I signed up for their premium service a couple months ago. No joke there’s like 3 entire seconds of extra super animation whenever you open the app. It doesn’t make any sense at all.
472.
by srcreigh 4y ago
"This proposal is vague in its usefulness" is the same thing as not adhering to specific, written, agreed upon goals and principles. So you could say "yes if this becomes more specific in relation to X". However, this si
473.
by srcreigh 4y ago
In a real-life situation where I tried this the keys were very wide, compound, with text keywords. That's where most of the savings came from since before each row was mostly key and not very much actual associated data. Sorry to misle
474.
by srcreigh 4y ago
PG arrays also compressed on disk. It may not be all that different than timescale. So ya if you store data in ts|id|value row format, it’s 70GB, but id|values array the values array is all compressed. I’ve seen 40x disk size reduction in p
475.
by srcreigh 4y ago
Very cool. with the define module the last annoyance with SQLite can go away, I can easily define functions to convert between epoch and iso times.
476.
by srcreigh 4y ago
I doubt you’ll find any decent audio or image related models which don’t use sine inputs. It’s ubiquitous. It’s why jpgs and mp3s work.
477.
by srcreigh 4y ago
It already is.. can’t find the paper, but some image related deep learning model used sin(x) type of inputs to their model. This was a core improvement. Not unlike how you can linear regress on x1,x1^2,x1^3,x2,x2^2. Encoding prior knowledge
478.
by srcreigh 4y ago
GCP supports lots of extensions including hint query etc https://cloud.google.com/sql/docs/postgres/extensions#postgr... As an aside, I question timescale use case. It’s highly niche, where you need to store
479.
by srcreigh 4y ago
Nobody asked, but that person is Donald Knuth
480.
by srcreigh 4y ago
For me the geopolitics are a lot more adversarial than that. For ex can't use FB or Google in China why - its American espionage against Chinese citizens. The question is why America (via Apple/Google) are allowing TikTok to spy o

More ›