Rachel Thomas

@math_rachel

co-founder + professor | past: Duke math PhD, quant, early Uber eng | now: making neural nets uncool

San Francisco, CA
যোগদান করেছেন মে ২০১৩

মিডিয়া

  1. ৩ ঘন্টা আগে

    I regularly talk to people who are miserable in their current jobs, have highly in-demand skills, and yet are convinced they couldn’t find another job and don’t even bother applying.

    Is leaving even an option?
In my experience, people in the tech industry consistently underestimate their employability and how many options they have. And I understand, because part of toxic jobs is making you feel like the problem is you, not the job. But please apply!
  2. ১২ জুলাই

    Even experts often struggle with training new deep learning models.

    One AI researcher at Stanford told me, I taught a course on deep learning and had all the students do their own projects. It was so hard. The students couldn’t get their models to train, and we were like “well, that’s deep learning”. Ali Rahimi, an AI researcher with over a decade of experience and winner of the NIPS 2017 Test of Time Award, complained about the brittleness of training in his NIPS award speech. How m
  3. ১২ জুলাই

    husbands are ten times more likely to leave women with brain tumors than wives are to leave their husbands

    a study from the American Society of Clinical Oncology in which researchers found that husbands are ten times more likely to leave women with brain tumors than wives are to leave their husbands. For other illnesses, such as lung cancer, the risk of divorce when the wife is diagnosed is 5.7 times higher than when it’s the husband.
    এই থ্রেডটি দেখান
  4. ১২ জুলাই
    It’s probably for that reason that men receive pain meds more quickly (within 47 minutes, on average, compared to women’s 65) even when they report the same pain scores. If a man is complaining about pain, it must really hurt. Women, on the other hand, might be experiencing pain — or maybe they’re just being dramatic.
Because of the lack of trust in women’s assessments of how they’re feeling at any given moment, the
    এই থ্রেডটি দেখান
  5. ৭ জুলাই

    there is no need to wait and see if the police or government agencies will abuse facial recognition technology

    As workers at Amazon explained in a letter to Jeff Bezos, there is no need to wait and see if the police or government agencies will abuse the technology. It is already clear that ubiquitous, automated facial recognition is well suited for discriminating against people of color, targeting political activists, and otherwise supporting militaristic and authoritarian modes of government. There is little doubt that the a
  6. ৬ জুলাই

    Surveillance is Orwellian when accurate, Kafkaesque when inaccurate. -

    Surveillance is “Orwellian when accurate, Kafkaesque when inaccurate,” Privacy International’s Frederike Kaltheuner told me. These systems are probabilistic, and “by definition, get things wrong sometimes,“ Kaltheuner elaborated. “There is no 100 percent. Definitely not when it comes to subjective things.”
  7. ৪ জুলাই

    Another example of how informal social interactions can have a big impact on who succeeds is tech startups that lack onboarding

    3. Institute onboarding. Even if you succeed in hiring more women and people of color, you are setting them up for failure without proper onboarding. Engineer Kate Heddleston noticed that for employees starting with the same experience level, again and again men were getting promoted much faster than women. Lack of onboarding was the source of the difference. Valuable information is shared through informal social net
    এই থ্রেডটি দেখান
  8. ৪ জুলাই

    Here are some tips from on how to counter clique culture:

    Assume that everyone here has something great to contribute.  Don't constantly gauge whether I'm worth your time or not.  Don't try to ONLY approach famous people.  If you're a group going out, invite some people who might be alone to come along.  Be friendly
    People become famous partly because someone decides to make them famous.  Don't just think of the people who are already famous, think of the people who you think *should* be famous
    এই থ্রেডটি দেখান
  9. ৪ জুলাই

    Important talk by at on avoiding "clique culture": cliques push people out, harm diversity, & we all miss out on good ideas

    এই থ্রেডটি দেখান
  10. ৪ জুলাই

    the problem with previous attempts at drag-and-drop machine learning interfaces is that they don't make the hard parts of ML easier. -

    Howard says that there isn’t really much point to these visual interfaces; the process is essentially the same as writing code, but it’s “more awkward, takes longer, and you see less on the screen at once.” He points out that to build anything other than the most basic application (like I did), you still have to know which components you want to use, how to connect them together, and so on. But instead of making that
  11. ২৫ জুন

    You Say This Can't Really Be America (2017) by

    You say We've never seen anything like this before.
I say Yes This is the America I know.
You say This can't really be America.
I say Yes This is the America I know from history.
  12. ১৪ জুন
    What bias is in the data?
Can the code and data be audited?
   Are they open source?
What are error rates for different sub-groups?
What is the accuracy of a simple rule-based alternative?
What processes are in place to handle appeals or mistakes?
How diverse is the team that built it?
  13. ৮ জুন

    The idea of "structurelessness" does not prevent the formation of informal structures, only formal ones. Structurelessness becomes a way of masking power. (recent h/t , I first read it via )

    This means that to strive for a structureless group is as useful, and as deceptive, as to aim at an "objective" news story, "value-free" social science, or a "free" economy. A "laissez faire" group is about as realistic as a "laissez faire" society; the idea becomes a smokescreen for the strong or the lucky to establish unquestioned hegemony over others. This hegemony can be so easily established because the idea of
  14. ৭ জুন

    medical data is often collected through systems designed by, and for, the needs of insurance providers first, and patients second.

    the surgeon and author Atul Gawande told the host: “Our systems are incredibly optimized for sending bills. I can send the bill in three keystrokes. But recording an allergy can be four different screens. So it’s not built to set a goal for care and then accomplish it.” In fact, beyond just creating an administrative burden on medical staff, an ethnographic study conducted in an obstetrical unit found that nurses fre
  15. ৭ জুন

    "software is training most users to spend their days oscillating between being interrupted by notifications and being distracted by the compulsive tendency to stay on top of inboxes"

    Even for the minority of users who successfully disable unnecessary notifications and curate their inboxes, it won’t fix the broader problem that software is training most users to spend their days oscillating between being interrupted by notifications and being distracted by the compulsive tendency to stay on top of an array of disparate inboxes (for whom what little time that remains is wiped out by the attention
  16. ২৯ মে
    That is to say, there is a growing consensus that the government has neither the power nor understanding to regulate tech’s algorithms.
This consensus misses the point entirely; there are actual humans behind these algorithms, and humans can be responsibly regulated. For years, tech companies have tried to obfuscate this essential truth, attempting to convince us that the behaviors of tech platforms are somehow separ
  17. ২৩ মে

    Numerous Facebook community "support" groups target addicts & their families. Many are run by marketers of for-profit rehab centers who often don't disclose their conflict of interest & hound members w/ predatory tactics

    He kept asking for my insurance info and even told me how wrong it was for me to keep hurting my family by not coming to [their] rehab center,” she wrote.

When contacted by The Verge, the woman said she’d been contacted by six different treatment center marketers since joining the group.
    The Verge interviewed three women who spoke with Knight about a referral for a loved one. All of them described Knight as pushy, even aggressive, in her sales tactics for rehab in Florida. One said Knight told her that treatment would cost over $20,000, suggesting she take out a home equity loan to pay.
    Marketers from the treatment center had to approve every post in the group, which gave them the first opportunity to privately message good candidates for their rehab and try to talk them into going to Windward in California. They needed that edge, Mendoza explained to me a few weeks ago, because they knew a Facebook group that big would be full of other marketers, waiting to swoop in as soon as a juicy message was p
    Facebook, by making desperation so easily searchable, has exacerbated the worst qualities the treatment industry. A word-of-mouth industry with a constant supply of vulnerable and naive targets who feel stigmatized and alone is a scammer’s paradise. Facebook does have tools to report groups that are abusive, but given the murky definition of patient brokering, Facebook’s legendary lack of transparency, and the fact t
  18. ২২ মে

    Lung cancer screening guidelines are based on a study in which only 4% of participants were African American. This impacts who qualifies for screening, cascading effect on future research. report on bias in precision medicine

    This all came from a study that was done by the National Lung Screening Trial, one of the largest stud[ies] for lung cancer screening to date. 53,000 people. Of those 53,000 people, only four percent were African American. So, you mean to tell me that we now have lung cancer screening guidelines based on four percent of the African American population. . . and because you have such low minority
enrollment in the stud
    African Americans may not be adequately
represented in this data, not because they chose not to get screened, but because some may not have qualified for a screening. This bias has cascading effects,  as those who are not screened have potentially important clinical data missing from their medical records which not only impacts their care, but recursively biases medical research since those biased screening data may
  19. ২২ মে

    The FBI: - has access to 411 million facial images - if true candidate exists in database, will be listed by their system in top 50 choices just 85% of the time - claims they don't need to worry about false positives from report

    The GAO report disclosed for the first time that FBI had access to over 400 million face recognition images—hundreds of millions more than journalists and privacy advocates had been able to estimate before that. Totaling 411.9 million images, this is an unprecedented number of photographs, most of which were collected under civil and not criminal circumstances.
    Because the system is designed to provide a ranked list of candidates, FBI has stated the IPS never actually makes a “positive identification,” and “therefore, there is no false positive rate.”60 In fact, FBI only ensures that “the candidate will be returned in the top 50 candidates” 85 percent of the time “when the true candidate exists in the gallery.”
    FBI indicated in a 2010 presentation that it wants to use NGI to track people’s movements to and from “critical events” like political rallies, to identify people in “public datasets,” to “conduct[] automated surveillance at lookout locations,” and to identify “unknown persons of interest” from photographs.
  20. ১৬ মে

    Machine Learning Arxiv papers are growing at a faster rate than Moore's Law. -- Zak Stone of TensorFlow

লোড হতে বেশ কিছুক্ষণ সময় নিচ্ছে।

টুইটার তার ক্ষমতার বাইরে চলে গেছে বা কোনো সাময়িক সমস্যার সম্মুখীন হয়েছে আবার চেষ্টা করুন বা আরও তথ্যের জন্য টুইটারের স্থিতি দেখুন।

    আপনিও পছন্দ করতে পারেন

    ·