sanjiwatsuki

I used to be a big keyboard enthusiast until I did a bunch of typing tests with many inexpensive keyboards. To my surprise, the data showed a statistically significant increase in my typing speed on several budget membrane keyboards.This disturbed me enough to fully convert back to membrane after all of these years.

For an idea of the scale, I typed about 140WPM on the HHKB and about 150WPM on the budget membranes.

If AMP had stuck around, we wouldn't have the quagmire of terrible UX on news sites today. The hostility towards AMP was wholly undeserved.

The current best open weights model is probably Cohere Command-R+. The memory requirements on it are quite high, though.

Almost no one can tell a difference with today's encoders (unless the track is particularly degenerate for the encoder) but I think there is an edge case: listening over Bluetooth.

When the already encoded audio is then encoded again into AptX or SBC or AAC, it starts to introduce audible artifacting - nothing most people would immediately notice but audible artifacting nonetheless.

No shot. Mistral Medium's outputs from API were virtually identical. Miqu really was Mistral Medium which happened to be a continued pretrain

The VRAM usage is closer to a 47B model - although only 2 experts are used at a time for inference, all experts are needed to complete it.

If you're looking for the best small model, I'd recommend using Berkeley's Starling-7B model [0].

It'll run on a lot of commodity GPUs and performs well in head-to-head comparisons against bigger models, edging out the most up-to-date GPT-3.5-Turbo [1].

[0] https://starling.cs.berkeley.edu/ [1] https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboar...

I really enjoyed Zume Pizza as a product. They sometimes could get a pizza to my door within 10 minutes of me ordering.

HR reasons are more common than you'd think. There's only so much HR and legal power to go around and it sometimes is easier to plan layoffs in multiple stages, rather than go for one big one.

The *majority* of times I've taken a taxi in my life, the driver has tried to scam me. It was ridiculous. I guess I just look like an easy mark or something. It was very routine for them to turn off the meter in the middle and say "don't worry, I know the cost of the trip" It's absurd that it's happened multiple times to me.

No Uber driver has ever tried to scam me. I'd gladly pay a premium for it.

This is one of the interesting parts of gun control legislation. It often polls well in isolation but massively under-performs when it's a ballot initiative, even when the gun control side spends substantially more.

Although there are certainly some servers with gauntlets to join, I think that's only for some really huge ones that have scaling issues.

I have no issue joining a Discord for technical assistance.

I do recall that one of the features that Wealthfront had was to design their UX in a way that discouraged behaviors like frequently checking the valuations, making it annoying to make emotional transactions, etc, etc. Rather than having a human tell you to be calm, they tried to mediate behavior through UX patterns.

AFAIK, every tool that integrates with these things apply a negative boost equal to the peak added to avoid clipping.

IIRC, Apple was already scanning iCloud server-side for CSAM [0] so I'm not sure if that excuse holds.

[0] https://www.macobserver.com/analysis/apple-scans-uploaded-co...

The metrics don't really change much if you account for CO2 emissions needed for production/transportation:

https://ourworldindata.org/grapher/prod-cons-co2-per-capita?...

This accurately describes the magnus effect which powers the majority of the movement that a pitch has.

I don't think this is accurate. [0] and [1] have 3DMark Ice Storm Unlimited Graphics benchmark tested on both the M1 and the 4800U in the Lenovo Yoga Slim-7-14ARE and the M1's GPU stomps the Vega 8 R4000. It outscores even the Ryzen 5000 series iGPU.

I've seen no evidence that the M1's GPU is anything but best in class for integrated graphics.

[0] https://www.notebookcheck.net/Apple-M1-GPU-GPU-Benchmarks-an... [1] https://www.notebookcheck.net/AMD-Radeon-RX-Vega-8-Ryzen-400...

Consumption taxes can be made arbitrarily progressive. Imagine a hypothetical one where the rate on spending below $1,000,000 a year is 0% and 200% after that point. There's a reason why the wording is specifically "a progressive tax on consumption."

This article is really inaccurate. The impedance isn't a good signal for how much power needs to be pumped into the headphone. The impedance also doesn't act as a good signal for what devices could generate enough volume. It also isn't a good signal for being able to "more accurately and vividly reproduce sound." Almost everything in this section isn't accurate.

Lower impedance headphones are "harder" to drive at the same voltage because they require more current. Often, amplifiers will have lower maximum outputs when driving into a lower impedance.

The sensitivity section isn't too bad but a lot of this article is pure snake oil and bad science. A lot of the amplifier talk in this article also doesn't hold up to scrutiny.

For further background, there were two surveys of top economists, one for European economists, one for US economists, where the economists were asked if the EU used its antitrust powers to protect EU-based firms from international competition, rather than to promote greater competition in European markets.

They can be found at the following links: http://www.igmchicago.org/surveys/antitrust-and-internationa...

http://www.igmchicago.org/surveys/antitrust-and-internationa...

At least in the academic economist community, there is disagreement on the EU's use of antitrust powers (although it leans toward the EU promoting competition more than being protectionist).

I generally interpret this data to mean that viewing the application of EU antitrust as protectionist is a defensible, although uncommon, opinion.

Can you back up those assertions with some sources? This seems overly simplistic.

I see people saying that drugs in the US are twice as expensive as other countries in the comments. I'll summarize [0] which is one of the sources for that number.

People in the United States spend roughly 2x per capita compared to Germany, Canada, and the UK. This does not mean that drugs in the United States are twice as expensive.

The US tends to use newer molecules in the drugs that are purchased -- particularly ones that are still patented and more expensive. Other countries like Germany and the UK tend to use older drugs (e.g. molecules over 30 years old).

A naive analysis would notice that the US spends more on drugs for a lower number of sales and would conclude it's because prices are high, but that ignores the fact that US doctors tend to prescribe newer on-patent drugs which are -- drumroll -- more expensive and not necessarily more effective. I'd argue this is the fault of the doctors prescribing the drugs more than any drug company's fault.

If you examine Exhibit 6 in the study, you'll see that, yes, the United States does tend to have more expensive drugs than most other countries but it is NOT an outlier. For example, Japan and Mexico pay more than the US for the same drugs on average. Drug prices are not nearly as big of a problem as people make it out to be.

Also, to its credit, the US also has significantly cheaper generics and over-the-counter drugs.

[0] https://www.healthaffairs.org/doi/full/10.1377/hlthaff.27.1....

"Build Me Up Buttercup" ends with one of the most interesting fade-outs. It actually ends with a key change during a climax, ending unresolved. I've always found that to be one of the most interesting artistic fade-outs from the oldies.

For a soda tax to work and reduce consumption, the price for soda must rise.

The cost of the taxed drinks, indeed, sharply rose.

The cost of the other untaxed drinks that were caloric substitutes, like juice or milk, did not go up. This suggests that people were not just substituting soft drinks for other high calorie drinks.

Early results suggest that the tax may achieve its goal of weight loss, but it is too early to tell.

For those interested in how the sugar tax might affect things going forward, the NBER did an early study on the effects of the sugar tax on Mexico [0].

[0] http://www.nber.org/papers/w21197.pdf

This sort of "static tax analysis" makes tax increases look like they'll generate more revenue than they would really do. This, in turn, makes the policies sound more affordable since the tax revenues are exaggerated.

This article is really poorly written in order to paint Trump in the worst light possible.

First, Donald Trump's dad died in 1999. The article is calculating as if he received 1/5th of his father's estate in 1974. Furthermore, even if he had received his share of the estate in 1974 and invested it entirely in index funds, Trump outperformed that by all estimates (as seen in the graph in the article).

Second, the article ignores all taxes and expenses explicitly. The article states that it assumes that Trump "reinvest all dividends, not cash out and have to pay capital gains, and pay nothing in investment fees." This is a ridiculous assumption -- everyone is going to have expenses. Everyone is going to get taxed on it at some point. With more reasonable assumptions, Trump trumps the SP500 further.

Third, even if Trump wanted to invest his entire net worth into indexes, it wouldn't be possible. I can guarantee you that it would have been very hard to liquidate his entire estate and invest into indexes.

You can say that Donald Trump became rich because he came from a rich family, but this article goes out of its way to make outlandish assumptions to make him look bad.

Actually, according to the latest CNN/ORC International Poll, Trump is the leader in the second choice polling, as well as first choice. Granted, take early polls with a grain of salt.

Citation: http://i2.cdn.turner.com/cnn/2015/images/08/17/rel8a.-.gop.2... on page 9

The headline is misleading and clickbait-y.

The rule in question (3500 calories deficit results in one pound of weight loss) is mostly true, even if oversimplified. The fact that your metabolism drops as your weight drops just means that the calorie/day input needs to drop correspondingly to maintain the same deficit.

As the article states, the fact that the weight loss isn't entirely fat is a minor error. Thus, the 3500 calorie guideline is maintained, as long as you remember that it is a deficit and not a static amount.