spott

13 ms·

151.
by spott 2y ago
> We find that, when compared to census tracts that were scheduled to receive a Starbucks but did not do so There still might be a confounders, but they at least tried to control for that.
152.
by spott 2y ago
Maybe the first time it happens you pop up with a confirmation dialog and it has a checkbox that says something like “always ask” (and an option in the settings).
153.
by spott 2y ago
So, it looks like if you try and add a kernel for a virtualenv that doesn't have ipykernel installed you immediately install it in the virtualenv with `pip` It would be great if you asked the user before doing this. My environments ar
154.
by spott 2y ago
Because easy scams to detect prevent false negatives for the scammer. If you get a response to a badly written scam, chances are the respondent will be a sucker and the scam will be successful. If you get a response to a hard to detect scam
155.
by spott 2y ago
(2023) Still very useful, but it should probably have a date in the title.
156.
by spott 2y ago
On the other hand, I feel like I see it all the time. Probably just different circles.
157.
by spott 2y ago
From the page: +6/-4 gs. That doesn’t seem like nothing, but I don’t know anything.
158.
by spott 2y ago
I mean, if you are attempting to use multiple sampling to avoid this kind of error, just use temperature 0 sampling and do it once. In your example you will get Pete every time.
159.
by spott 2y ago
Cnc kitchen did some tests on brick layers: https://www.cnckitchen.com/blog/brick-layers-make-3d-prints-... He saw some strength gains, though not huge gains.
160.
by spott 2y ago
A lot of cad software actually has an editable history: you can go back 100 steps and modify a dimension or a step and have the rest of the model updated to take that new dimension into account. I wouldn’t work with any cad software that
161.
by spott 2y ago
I can’t imagine how you could build such a study and believe it. In other words, I wouldn’t believe the other 20%…
162.
by spott 2y ago
Yea, but wood isn’t actually uniform, so a finite element analysis might find a joint that is stronger on paper, but the particular piece of wood it needs to work on makes it weaker in practice. It might be neat to do it automatically based
163.
by spott 2y ago
> Apple's refusal to make RAM upgradeable I'm not sure they can make RAM upgradeable at this point and still keep their unified memory. If you look[0], they have the memory under the heat spreader, butted right up next to the
164.
by spott 2y ago
Mostly copilot is a nice autocomplete. Sometimes it writes what I would write and then I don’t have to type it out. Sometimes it helps when I’m writing code in a domain I don’t know. It can pull in a library function I wasn’t aware of for
165.
by spott 2y ago
https://github.com/NVIDIA/Megatron-LM This is probably a good baseline to start thinking about LLM training at scale.
166.
by spott 3y ago
This is one massive blob of text in reader mode. No line breaks show up.
167.
by spott 3y ago
Because as companies get bigger they worry more and more about the long tail of risk. They do this for many reasons, but generally predicting the risk of very low probability events is very difficult, so they don’t understand the risk vs co
168.
by spott 3y ago
I think most of them do… but they are too easy to remove. It isn’t just quality drop resistance, it is also filtering and post processing resistance. It is kinda like antivirus signatures: those who want to abuse it just filter and tweak
169.
by spott 3y ago
The peters principle makes an assumption about the convexity of competency that I’ve never thought was great. Essentially someone could get promoted to their level of incompetence, when if they were promoted again they would actually get b
170.
by spott 3y ago
NIST, but they required password rotation up until very recently, against their own advice.
171.
by spott 3y ago
> Anything that requires taking 2 things into account that aren't right next to eachother is just not going to work. They kinda address that in the technical report[0]. On page 12 they show results from a "multiple needle in a
172.
by spott 3y ago
> Gemini Pro seems slightly better than GPT-4 original in Human Eval (67->71). Though they talk a bunch about how hard it was to filter out Human Eval, so this probably doesn't matter much.
173.
by spott 3y ago
> it's just voltage and all you have to do is step up or down. It isn't quite that easy. The high-output 18v batteries can dump over 100 Amps for short durations. Stepping that kind of power up or down isn't really possi
174.
by spott 3y ago
The configuration language evaluates to YML that defines a GitHub action. The configuration language doesn't run GitHub actions.
175.
by spott 3y ago
Did they reduce the cost of everything that it was baked into? Did EC2 instances get cheaper?
176.
by spott 3y ago
Probably on some fixed date every month, rather than the the day with the lowest price every month.
177.
by spott 3y ago
How is this different from continue.dev?
178.
by spott 3y ago
Microsoft invested $10B after GPT4 came out… their initial investment of $1B might have been for R&D, but their $10B investment was for tech that they knew about.
179.
by spott 3y ago
I wonder how much heavier it is than down?
180.
by spott 3y ago
For the chemists here, I have a question I’ve always been curious of. Would you eat/drink out of glassware that had been made “quantitatively clean”?

More ›