What are the advantages of Whisper over a custom app running on a cheaper Andriod tablet? How much benefit is derived from the dual microphones, for instance? Are there hardware-level APIs that Whisper offers which a custom Android app wouldn't be able to access?
I don't mean to nitpick, this looks really cool. I've just got a few old tablets lying around, and I'm trying to decide whether to spring for one of these instead of trying to make something work on what I've already got.
Willow uses a local ML model for wake word detection. Once wake is detected, the actual speech recognition has one of two user configurable modes:
Local - Willow also includes the latest ESP SR Multinet 6 command recognition model. Willow will automatically pull entities from Home Assistant (when using Home Assistant) and define the speech grammar based on the friendly entity names. In this mode, the speech/audio never leaves the device and the speech recognition result is sent directly to Home Assistant.
Server - In this case, after wake is detected we immediately begin streaming audio to our highly optimized inference server implementation (release next week). Once end of speech is triggered using the ESP BOX Voice Activity Detection we send an end marker to execute Whisper on the server side and then take the results and send them to Home Assistant (when Home Assistant is configured).
Doing reliable wake word detection and getting clean far-field speech (generally defined as 3 meters or more) in random/unknown environments that are typically less than ideal (background noise, acoustic echo, etc) is actually quite a challenge. Willow uses the dual microphones and the ESP SR AFE (audio front end) to do a variety of signal processing on the device to clean the speech.
The integration and engineering for anything resembling an Echo-like experience is very involved, down to physical attributes of the enclosure, microphone cavities, etc. There is an entire field and cottage industry of acoustic engineering on the hardware design for these applications.
The point is, providing an Echo/Willow-like experience is much, much, much more than putting a random microphone in a room. So with that, we don't plan to specifically support random devices because the outcome is almost certainly very poor and not something we're currently interested in supporting.
All of this keeps comes coming up and we will certainly document it.
A highly-enriched uranium fission bomb is conceptually and practically simple to construct. The hard part is getting enough HEU. That part isn't even hard to achieve (North Korea managed it after all), just moderately expensive and hard to conceal. This process and the bomb design is explained in detail in the nuclear non-proliferation documentary, "Countdown to Zero" (2010): https://www.imdb.com/title/tt1572769
The amount of HEU needed for a device of this type with a yield comparable to Little Boy could fit in a shoebox. Seal it in some lead pipe, stick it in a shipping container and no one would know. The above documentary illustrates this powerfully with animations and expert testimony. Strongly recommended, especially as a companion to "Command and Control" (2016): https://www.imdb.com/title/tt5598206/
This is a pointless observation since the male/female difference is the approximately equal for all races. In contrast, black women in the US commit more murders per capita than do white men: 5.8 versus 3.5 per 100k, as of 2021 per the FBI/Department of Justice. Black women are the only female group with this distinction compared to the male ethnic majority in the US, and all other countries of which I am aware.
It does not. Americans have a median income roughly 10% higher than Germans, far more guns per capita than Germans, and nearly 40% of male homicide suspects in Germany are non-German in any case:
>Can you not just accept that "white" (however you like to define the term presently) men commit significantly more murders in America than in Germany?
Sure, I agree and I'm pointing out reasons as to why that is. I have never before heard gun policy lumped in with "socioeconomic factors."
My original point is that socioeconomic factors alone don't explain the gap in murder rates between e.g. black women and white men in the US and in other countries. Your argument to the contrary is woefully insufficient. Comparisons between the same race and sex in two similar cultures at similar income levels isn't enough.
We already know wealth and income correlate somewhat with crime. This effect is seen within racial groups. What we don't know is why large gaps exist between different groups at similar income levels. The effect is particularly stark and peculiar when it comes to violent crimes e.g. rape which have nothing to do with stealing to feed oneself.
Violence perpetrated ON immigrants however, would inevitably rise since the same statistical validity applied to most women being the victims of violence.
That heritable factors which correlate with race significantly impact rates of violent crime, including large differences in crime rates between groups; that these differences matter more than sex differences, particularly given taboos around race; and that this knowledge should drive expectations for fair criminal justice policy.
No one believes that since most imprisoned murderers are men, there's systemic anti-male bias in the justice system. The relevant biological differences between sexes are acknowledged and accepted. I'm asking HNers to accept similar well-studied differences between ancestral genetic groups.
Refusal to acknowledge and accept these differences has already led to bad policy, such as de-policing America's black communities, which has cost hundreds of black men their lives in just the last few years. A minority of this minority is allowed to run wild and victimize their neighbors and society at large. This is plainly unsustainable. We can take our medicine now and accept the need to police violent criminals, even if they are black, or we can keep our heads in the sand until something far worse boils over. This is the core message of all my comments on this link and race issues at large.
The reality is, if we were using demographic factors to prevent immigrants who cause violence from coming into a country, there's a few that are much more predictive than race.
First, we'd ban all men.
Second, we'd ban everyone under the age of 30. Maybe everyone under 50 if we were feeling really fascist.
With these two evidence based restrictions, we'd effectively eliminate violent crime perpetrated by immigrants. That's the stark reality, and its so funny to see racists try to avoid these points.
> We already police men way more than women, what's your point?
We would need to put more police in places where men congregate to be truly consistent - like software companies. The de-policing of men's places of work like these, may be putting us at risk, after all.
I could say that no one has seriously proposed "eliminating ALL violent crime perpetrated by immigrants," and you're using this absurd strawman as some sort of shield to avoid discussing race, and that people generally accept other groups who commit violent crime at similar rates to their own, and it's just one group in the US in particular who falls way outside the bell curve (per link theme)... but it's clear you aren't here for earnest or honest discussion, so I'm out.
Oh no you've got me all wrong! I love discussing race, and in particular how made up the concept of whiteness is. It's just most racists don't like to have that discussion, they'd rather try to throw stats around to justify more policing, but the stats don't lie - it'd be by far more effective to focus on and police men of all persuasions, rather than focussing on different made up race categories.
"Made-up race categories" are visible to the naked eye, to principal-component analysis of alleles, to measures of genetic distance, to surveys of visible and invisible traits, to even the smallest child who's only begun to speak.
What was the Human Genome Project for? Your mockery and denial, I guess. Why have scientists spent lifetimes improving DNA sequencing technologies, carefully gathering data, and drawing conclusion based on the evidence? So you could laugh it all off, I suppose.
Your refusal to acknowledge reality fails to budge reality. The natural world doesn't care about race or sex solipsism. Policy founded on reality-denial will always eventually fail and be replaced. Whether it fails hard or soft and what is allowed to replace it is the question of our time in America.
> "Made-up race categories" are visible to the naked eye, to principal-component analysis of alleles, to measures of genetic distance, to surveys of visible and invisible traits, to even the smallest child who's only begun to speak.
Initial definitions of white, didn't include Scottish and Irish people.
I believe they are simply pointing out the hypocrisy of the racist rant. Racists claim that the reason they are against immigration from certain countries because of the increase in crime rate they would allegedly bring.
It is true that I'm being condescending to racists. I'm also not expecting a positive response if the person (not necessarily you) I'm engaging is racist. It would be foolish.
> "allegedly"
Allegedly doesn't mean that they are wrong. It simply means that I don't know if they are wrong or right, because I don't have enough data. If I thought it is complete bollocks I would have written it as such.
Also you seem to ignore that the main point of my comment is the "hypocrisy", i.e. only pointing out racial statistics, ignoring the gender (and possibly other variables, such as age) related ones.
>I don't get the negativity expressed by top comments.
Lots of ideologically-opinionated users are mad Twitter no longer censors and bans their opponents. They're upset that their patrons who enforced this got fired by Musk. Much seemingly unrelated criticism flows from this.
As to the actual Twitter experience, I've always used a chronological feed and never understood complaints about the algorithm. Step one to not being bothered by an opaque algorithm's control of your attention is withholding that control in the first place. I haven't noticed any change in my daily Twitter browsing in this regard under Musk.
By "their opponents," I meant opponents of the previous Twitter owners.
The neutral platform is gone. As we were so often and so smugly reminded, Twitter is a private company. They can ban who they like. This is the dog-eat-dog public discourse world you chose.
>But as a public figure I think he's massively overexposed, largely by choice, and it's doing him no favours in terms of perception lately.
I think you'll find there's a significant split in perception between, for example, the average American and the pseudo-elites who inhabit Bluesky and Hacker News. Or the latter and the average actual hacker, for that matter.
Is there direct evidence of Elon's personal engineering prowess? All I can find are his own boasting and second hand reports of folks impressed by his familiarity with jargon.
Second and third hand accounts are nice. Even better when they come from folks no longer financially dependent on Elon's good graces.
But I was hoping to see some of his source code for Zip2, some CAD he'd done himself, or even a video of him solving some intractable problem in one of these rooms full of stumped geniuses.
So this guy has founded multiple successful engineering-heavy companies, did not receive any actual financial support past 18 (despite inflated myths of an "emerald mine"), is attested by multiple skilled engineers, and has given videos where he talks in depth about engineering, but we're still looking for new & different kinds of evidence that he has skills?
Are we testing whether he knows anything at this point, or whether he's a literal Tony Stark? This thread started with discussion of whether he's any kind of engineer at all now we're looking for "rooms full of stumped geniuses" as if this was something every random engineer experienced.
Didn't see any links to talks by Musk. Perhaps you can share some?
Look I can go read Gates' code, so I have some direct evidence of whether his reputation as an engineer is worth the paper it's written on. The story for Musk is much less direct. Still, I'm open to studying the evidence of Musk's engineering chops.
The buyers do not approach those who have been legally deemed incompetent. As you say, such a sale would not be valid. But then again, can the isolated mentally incompetent even bring such cases to court? Vulnerable people who break a hip and can't afford surgery also don't deserve to lose their homes, especially not at a fraction of market value. There are clear gaps in the system which these buyers are exploiting and which you are repeatedly and willfully excusing. There's a chapter in the Bible about this type of behavior involving whips, overturned tables and an angry Jesus.
>or “white trash” which could both be a slur and the title of an edgy Vice documentary
It's only not considered a slur in the second case because Vice has money, you and your friends and powerful people like Vice, and the "white trash" don't reflexively assault you like beasts when they hear the words uttered. The meaning is understood and remembered regardless.
The evidence-based research on long COVID is limited and of mixed quality. From what I've seen, there are significant gaps in experimental methodology and a lot of studies based on extremely subjective surveys which don't effectively filter co-morbities. I would strongly caution you to examine the data before adopting a strong position on the issue and applying it to potentially unrelated phenomena.
Long COVID, from what I can tell / have read, is acknowledged by research[1][2]. The mechanism of it, and what is the best treatment for the people suffering from it, I think, are still the subject of ongoing research.
If you look closely at what those studies actually survey and how that data is collected, it's clear that many possible co-morbid conditions aren't adequately filtered out. There is _no_ objective test or evidence-based protocol that can conclusively diagnose "long COVID." Doctors must rely on checklists of broad symptoms and the patient's word. The CDC page you linked mentions this under "Data for Long COVID:"
>For example, some studies look for the presence of Long COVID based on self-reported symptoms, while others collect symptoms and conditions recorded in medical records. Some studies focus only on people who have been hospitalized, while others include people who were not hospitalized.
Can serious health conditions can arise as a lasting consequence of infection? Absolutely. Can we easily distinguish between those and other symptoms or conditions which may arise out of stress, anxiety, other unrelated conditions, or symptoms which in some (not all) cases may be psychosomatic? Can we accurately draw conclusions as to how many long-COVID patients there are in total, then apply that conclusion in other areas with confidence as the GP commenter has done? Not yet.
Though it is worth noting there is experimental clinical data providing a plausible pathway to suggest long COVID is a result of cell activity dysfunction with plausible biomarkers of dysfunction, and we've known this long enough to know long COVID is a real medical pathology (one distinct from acute infection based on biomarker evidence):
>A unique two biomarker profile consisting of ANG-1/P-SEL was developed with machine learning, providing a classification accuracy for Long-COVID status of 96%.
The first paper used a random forest-based decision tree classifier built on markers in blood assays. Neat.
However, this study has a major flaw. They rated their classifier's accuracy on classifying blood marker profiles of acutely ill COVID patients, long COVID patients experiencing "diffuse symptoms" referred with "no selection process", and a healthy control group. The control group consisted of healthy patients whose blood samples had been banked prior to the COVID-19 pandemic.
It's not clear whether they compared their classifier's results against people who've had COVID and recovered without issue, versus those who had COVID and continued to experience symptoms long after recovery. That is the entire point of developing such a classifier. This paper is worthless without that comparison.
Second paper has the same problem, and is honest about it:
>The healthy control subjects were individuals without disease, acute illness or prescription medications and were previously banked in the Translational Research Centre, London, ON (Directed by Dr. D.D. Fraser; https://translationalresearchcentre.com/). These latter samples were obtained prior to the emergence of SARS-CoV-2 in our region and therefore, were considered not to have been exposed to the virus.
I don't mean to nitpick, this looks really cool. I've just got a few old tablets lying around, and I'm trying to decide whether to spring for one of these instead of trying to make something work on what I've already got.