cfcf14

Completely agree! It's not even clear who is asking. Causality is erased, as you say.

"I have an ask for you" - is it my ask? Am I conveying somebody else's ask? Who knows? It's atemporal, acausal, you can choose to investigate by querying but there's a good chance the response is going to be a shrug and a small, apologetic smile accompanied by "ah - you know, the usual" at which point you'll sigh and respond with "yep - ok, let me see if I have the bandwidth and circle back to you on that".

Exactly - this kind of research is repugnant and hopefully will be seen the same way phrenology is seen.

A paper I once found really illuminating on what a kalman filter was and how it related to a similar models is "A Unifying Review of Linear Gaussian Models", by Sam Roweis and Zoubin Ghahramani. It can be found here in pdf format: http://www.stat.columbia.edu/~liam/teaching/neurostat-fall20...

Meta comment: I really appreciate that the author is taking the time to reply to people with queries (not this post). Thanks! It makes this community better.

I've had a few short-term contracts (3 months or less) at £950-1100/day as a staff level data scientist, although those are usually 'exceptional need' situations where the company has their back against a wall for some reason or another.

These days it feels like the market is actually cooling down to me, but that could just be because I'm looking in the wrong sectors.

Modern sound mixing makes it so difficult to hear dialogue that it's practically required for most films in my experience.

Interestingly enough, passing even the most complex example in the link to GPT-3 with the prompt "What does this code output when run?" returns the correct result.

Essentially means to play poorly on purpose, with the intent on making the opposing team stronger, or to give them a strategic advantage against your team. People do this to ruin games for whatever reason, it's a phenomena which persists throughout many multiplayer games online.

(Assuming you mean AC) - HVAC systems are generally designed to keep air flowing through the building optimally, which may include control systems for attenuating fans, exterior building air intake, and whatnot as a function of sensor measurements. I'm not sure how often CO2 is taken into account during building design or operation to be honest, aside from rough targets on designed occupancy rates and volumetric estimates of airflow.

In the UK, most buildings are naturally ventilated. There's no mechanical ventilation moving air around automatically, you gotta open your window if you want airflow. In the winter this generally puts people between a rock and a hard place: either you are slightly cold and sitting in a stuffy roof, or you open a window and are REALLY cold and also go broke from the energy bills :P

I wonder how this will affect attitudes towards schooling as an institution in general in the future? These children will grow up and take their places as policymakers, parents and so on - will they value school more, or less?

Perhaps we'll see some large scale restructuring of the school system altogether, which at this point doesn't seem like the worst outcome necessarily, every teacher I've met in my adult life has expressed pretty vocal grievances about almost every administrative aspect of their jobs.

I hadn't considered how bad public transport might be in terms of ventilation, but it makes sense. I wonder how bad the London tube might be on a busy day?

It's unfortunate that research on the impact of Co2 concentration on cognitive function and general wellness is so inconclusive (ie: effect size of high concentration on most tasks seems small), when subjectively it feels really high.

Some years ago I did some datascience work with a housing association wherein they installed high-frequency sensors in various rooms of housing units across the country (with the explicit consent of the inhabitants, data anonymized) with the intent on understanding how building properties and use correlated with the sensor data. In general there was (and still is) a lack of understanding how to model the relationship between things like building material, climate, occupancy, tenant behaviour and negative outcomes like mold growth, fuel poverty/health outcomes, and so on. I mean - I'm not saying there's no understanding, but it's definitely a field where more work is needed.

Anyways, one of the things I found most shocking is that in many of these flats, Co2 concentrations would get mind-bogglingly high in bedrooms during the months where tenants would likely have the windows closed (this is something you can infer by temperature gradients and heating usage). By mid-way through the night, some individuals were in rooms with 4k+ co2 ppm. A common side effect of breathing air with such a high concentration is waking up feeling groggy, disoriented, or with a headache. But what are these people supposed to do - many of these flats were low-income housing, so they're badly insulated to begin with, and they probably can't afford the heating bills of keeping windows open for ventilation.

Sorry for the long story - but the point is that there are many negative health externalities associated with poor ventilation in buildings and interior spaces, and at least we're starting to get attention on the problem now with covid. Hopefully this leads to improvements across the board.

The author could have probably shed a tad more light on why they feel music apps (spotify et al) are bad. I agree with them though - these apps are atrocious to use, and seem to be continually getting worse. Some of my ongoing complaints about spotify are the following:

* Offline mode still doesn't work properly, after literally years of development. If you lose connection while streaming and want to switch to your 'downloaded albums'? Nope, the app will hang on a blank screen or crash. As londoner, this means spotify basically doesn't work on the tube at all - again, _even in offline-mode_.

* App will generally blank screen, hang, or crash semi-regularly for no apparent reason at least a couple times a day.

* Audio quality is awful. The default is 96kbps, which I believe even the most untrained listening can hear the crumminess of.

* Tons of missing music, even from mainstream artists.

* UI makes it difficult to navigate playlists <-> artist <-> song <-> album, traversing the hierarchy of music associations

* UI prioritises pushing podcast & other unwanted commercial content at you, make it difficult to find what you want and navigate the library

Then you've got the scumminess of the company itself that you alluded to (not paying artists, toxic corporate politics, etc).

For individuals with certain auto-immune diseases (like rheumatoid arthritis), the medication they're on can also interfere with b12 absorption - such as methotrexate. This is a possible consequence of what you mention regarding absorption digestively. Injections are a solution but are not always accessible in every country - in the UK they are extremely hesitant to permit b12 injections, for whatever reason.

Hey, I noticed that you explicitly mentioned that you found a female md that was the one to ultimately make the correct diagnosis. Do you feel that the previous md's (males?) were unwilling to consider they should conduct a tick panel due to bias? I ask because my partner has had comparable resistance for a very different ailment and I'm interested in possible systemic errors occurring.

I agree with the other commenter and feel the connotations of your post, the hints at your enjoyment at the prospect of this potential future - are disturbing.

Yep - I also used to use airbnb without even thinking twice about it in Europe. Last few trips have been getting steadily worse, culminating in a stay where the residence was completely unsuitable for habitation (extremely cold & heating was broken, no linens/towels, mold & damp issues) and airbnb caught us in some sort of bizarro catch-22.

Basically they said: "we can't refund you until you give the host a fair opportunity to address the issues, so get back to us in 24 hours if the host hasn't done anything". When the host failed to do anything and we re-contacted airbnb, they said that the period of validity for us to file a complaint and get a refund had expired, and they would not do anything to help us.

My wife and I fought this, HARD. One of the few times I've basically seen red while trying to sort out a simple issue like this. The host was contacted by airbnb and was livid, and became extremely hostile and unfriendly to us in communications. We ended up getting a hotel nearby and working out the whole thing through travel insurance. Honestly, fuck airbnb.

Isn't the VQ-VAE/dVAE generator approach in the DALL-E models quite a bit cheaper computationally than latent diffusion models?

My understanding was that diffusion models were quite a bit more expensive, but yielded richer latent distributions and better images (for some definition of better).

Practice! I forced myself into so many social situations, over and over again. I went to speed date nights on my own. I went to bars to try to talk to strangers.

It was so painful at first. I embarrassed myself. I was foolish and awkward. But I got better. The terrifying became the comfortable, and the mundane.

The journey continues! All things come with practice.

This blog seems to come up because the author voices a sort of plainly spoken view of their experiences in tech companies and people kinda commiserate with it.

Maybe it's a condemnation of the regular viewership here (and myself), but I find Rachel's posts pretty easy to identify with.

I think the startup has to genuinely care to begin with. And often they just don't have the expertise to realise what they're falling short on.

One of the things that endeared me most to my current big tech employer was the understanding that it would be weeks before I could do anything.

Yeah. I sat through the sessions and followed along. But the internal docs were sorta out of date, and basic stuff seemed to just... not work. And SREs do what they do- ask a ton of intimidating questions and make you feel on the defensive.

But I wasn't expected to get set up. Part of onboarding was slogging through the tech. The k8 madness, the permissions and salts and accesses and everything else. And although it fueled my imposter syndrome at times, I appreciated knowing everybody else went through it too. I wasn't expected to just get it.

We expect people to take a few weeks to settle in. I value that - because onboarding is intimidating as hell. A company that doesn't see that - red flag, for sure.

In addition to this, their company (or its parent) funded the trial too.

Borderline of being a staff-level machine learning engineer at a large technology company, and despite getting an exceptional review (and completing a major product delivery), I will also be getting a sub-inflation comp adjustment.

Already started interviewing, early indications and offers suggest a 40-60% pay jump is within reach, possibly more with equity. I just don't understand why companies do this to themselves.

It's not clear what we can actually do to help. I work as a web-sec machine learning researcher, and I'd like to help. But anything I can do is probably illegal in my country or a breach of contract with my employer :(

Hey thanks for posting about this! I never would have discovered it on my own :)

When I read the title, I had a very different idea in mind for what the content of the article would be.

I think it's interesting to apply a sort of topographical analysis to the virtual regions of WoW to try to understand why they're so important to us. Kudos to the author, I enjoyed reading his analysis. However, as other commenters have noted - I believe it's impossible to understand the allure of the game independently of the social elements.

To me the most important part of what made these environments compelling was the ability to actually quantify my interaction with them, exactly. Almost 20 years ago - god, has it been that long? Going into this virtual world with a counter that exactly captured how 'well' or 'good' I did was some kind of revelation. To people who played: these were the first DPS meters or counters. The foundations of quantitative gaming analysis for these kinds of games.

Of course, in the real world we have tests and races and everything else. But it wasn't the same. In World of Warcraft you could bend your entire intellect against the goal of maximizing some number which summarized your interaction with a virtual agent in a public way. This wasn't the first game to do this, but it was the most successful. If you were were good at the task (of interacting with the game environment), this gained you prestige in a highly visible way. For a million faceless and lost teenagers across the world, this mechanism was an indispensable lifeline.

Although it would be a world very different from ours, I wonder what the consequences of an ad-ban would be? You create a business or service, but the only way an individual could discover it is by intentionally going out and seeking that information (perhaps you've made a new washing machine, and they seek out information for washing machines).

Perhaps specialised 'discovery' services would pop up for people wondering what sorts of new stuff/services was out there?

On the surface of it, I'd assume that if reaching out to customers to 'let them know you exist' via advertisements was prohibited, the amount of innovation and investment into products would decrease. But at the same time, the amount of money allocatable to product development within a given budget would increase, as you wouldn't have to expend so much on marketing (which in some cases, dwarfs the cost of R&D). But maybe the markets would just adapt anyways. Seems a fun thought experiment.

I've added one, will be pleased to hear from you :)

For many people, it's more than just a jab, it's 1-3 days of being quite unwell.

I have no underlying health conditions, but I've been bedridden by my 2nd shot and booster shots, with a high fever, fatigue, dizziness, and aches for at least 48 hours. My partner who is immuno-compromised is worse off, generally being very ill for 1-2 weeks after each jab. Her case is somewhat different though, as her situation is just all round unfortunate.

I'm not saying I won't continue to comply with all health advisories for vaccination or other public safety measures, and I'm not saying vaccines aren't worth it for society. But it's not "just" a jab.

Hi, do you have an email I can reach you at? Would love to chat more about this.

For any kind of research or experimental work, I cannot imagine using anything other than PyTorch, with the caveat that I do think JAX is extremely impressive and I've been meaning to learn more about it for a while.

Even though I've been working with Tensorflow for a few years now and I feel like I do understand the API pretty well, to some extent that just means I'm _really_ good at navigating the documentation, because there's no way to intuit the way things work. And I still run into bizarre performance issues when profiling graphs pretty much all the time. Some ops are just inefficient - oh but it was fixed in 2.x.yy! Oh but then it broke again in 2.x.yy+1! Sigh.

However - and I know this is a bit of a tired trope, but any kind of industrial deployment is just vastly, vastly easier with Tensorflow. I'm currently working with ultra-low-latency model development targeting a Tensorflow-Lite inference engine (C-API, wrapped via Rust) and it's just incredibly easy. With some elbow grease and willingness to dive into low level TF-Lite optimisations, one can see end to end model inference times in the order of 10-100us for simple models (say, a fully connected dnn with a few million parameters), and between 100us-1ms for fairly complex models utilising contemporary architectures in computer vision or NLP. Memory overhead and control over inference computation semantics are easy.

As a nice cherry on top, we can take the same Tensorflow SavedModels that get compiled to TF-Lite files and instead compile them to tensorflow-js for easy web deployment, which is a great portability upside.

However, I know there's some incredible progress being made on what one might call 'environmental agnostic computational graph ILs' (on second thought, let's not keep that name) which should open up more options for inference engines and graph optimisations (operator fusion, rollups, hardware dependant stuff, etc).

Overall I feel like things have been continuously getting better for the last 5 years or so. I'm pleased to see so many more options.