vrc

I'd correct this to be "a smaller sized EV" versus "the smaller size of an EV" as many of the EVs on the market are surprisingly big. Without pointing to the obvious behemoth truck/SUV EVs, my Tesla MY is considerably larger than a lot of CUVs on the market and has a much worse turning radius. What it does do well is stop and start fast meaning I can be more decisive and navigate through messy NYC/BOS traffic with more confidence than in my CRV, and regen braking + having lower "idling" costs does improve its city-profile.

Out of curiosity and semi unrelated — why do so many of these projects with customized encoder-decoder setups use earlier Qwen versions like 2.5 and 3 and not the smallest 3.5? Purely the few 100m params, or something else in the latter’s arch or pretraining?

Counterpoint: you can't replace Reels/TikTok or smoking time with exercise and you _have_ to displace something meaningful to do dedicated exercise.

Phone activity time is almost irreducible because it's not dedicated or de-fragmentable by design. It's almost always concurrent with something else, or in the "in-between" periods where your alternatives are limited.

One successful strategy here is to offload exercise into "snacks" that compete for this time. And that might buy back time you spend at the gym. Strength training and mobility work where the emphasis isn't on getting your HR up can easily be squeezed in here. So you might go to the gym, do your pressing exercise in 20 minutes, then come home and do pushup press variants as your accessory work in small increments over the day.

The one area this does not work on is CV training, where you need a dedicated amount of time in target zones to get the benefit. For that, you need to treat it as a first-class obligation, find some protocol that meets your constraints, and hold it sacred to get it done.

Somehow I keep hearing the rumblings of crypto maximalists trying to merge tokens. I actually wouldn’t mind since I signed up directly with some providers I’ve stopped using and have small amounts of credits strewn across the web.

It wins on two fronts if this is true. Provides the cheap alternative for the West, and maximizes returns against their homegrown audience.

I’ve used both Belsomra (suvorexant) and Quviviq (daridorexant) and compared to the litany of other sleeping aids, Quviviq in particular is great for me in terms of not being groggy, not knocking me out standing up in case I somehow don’t sleep or need to do something. No rebound on days I run out or if I stop. Feels quite natural—I imagine it feels like what most normal sleepers feel like every evening. It’s nearly impossible for many to get them, tho…

Edit: To add some academic merit — there’s been research for its use for cognitive decline [0]

[0] https://www.nature.com/articles/s44323-025-00025-5

What are your thoughts on the detector --> VLM pipelines, and if there's ever a world where a small LM or LM augmented detector can be efficient enough to play a role as router. I ask because I recognize you from your handle and am very familiar with your work in the doc+detector space.

Paddle-VL and GLM-OCR do this by using PP-DocLayoutv3 as their "detector/slicer" and then just batch the OCR on the clips to do pretty darn well at a tiny size.

A lower-tech version is to use a good detector and XY-cut or just a naive Y-cut or orientation-away cut to slice up the page. But if you're doing that you're getting closer and closer to DocVLM style OCR+low res image. Been playing around with something like this using the new PPOCRv6 which itself punches well above most traditional OCRs and is multi-language without the hassle of language detection and dict-loading for rec.

It's widely hypothesized that dogs anthropomorphized themselves, so to speak, accentuating their expressive eyes and eyebrows over generations to be more human-like in how they communicate. And very few humans today view their dogs as pure working tools -- most at least say "good boy".

Does it prompt logging? For example, when I was trying to monitor my BG after diagnosis, I tried to log my meals to correlate later, but 1) would forget and 2) wouldn’t have the energy to time align the stats. So a tool that even saw changes in BG and shot me a text or message (did you eat/exercise do something @ [time]?) and used the LLM or something else to capture and enrich the metadata. Paired with boring things like med reminders (I just realized I forgot my metformin while typing this) and giving me an easy visualizer with these meta points would be useful. If I’m tracking sleep on a device etc.

As others have said, the analysis might be risky. I don’t want to trust interpretation to anyone but myself (bear my own risk) or my clinician. But just remembering to capture the data and making it easily time alignable and possible augmentable in the future would be useful.

Does Wallet allow apps to interact with the meta-data of cards, and/or update them in any way? This could be interesting for insurance cards, in particular. Upload & verify status periodically with a prompt to update, for example.

Of the examples you listed, politicians are the only ones you directly fund and supposedly work for you. Your lawyers and doctors aren’t your employees, and they also don’t work on your property (though lawyers might handle your documents). The biggest thing this points to is that the mask is almost entirely off between employee-employer relationships in the US, and it looks like by ensuring everyone depended on employment for insurance before turning this corner, there’s not much resistance left.

Depends on if you believe it will ever become cheaper. Either hardware, inspiring more efficient smaller models, or energy itself. The techno optimist believes that that is the inevitable and investable future. But on what horizon and will it get “zip drived” before then?

I did the opposite. Tesseract to get bboxes, words, and chars and then mistral on the clips with some reasonable reflow to preserve geometry. Paddle wasn’t working on my local machine (until I found RapidOCR). Surya was also very good but because you can’t really tweak any knobs, when it failed it just kinda failed. But Surya > Rapid w/ Paddle > DocTr > Tesseract while the latter gave me the most granularity when I needed it.

Edit: Gemini 2.0 was good enough for VLM cleanup, and now 2.5 or above with structured output make reconstruction even easier.

I’m a PM and I’ve been able to do a lot of very interesting near production ready bits of coding recently with an LLM. I say near production ready because I specifically only build functional data processing stuff that I intentionally build with clean I/O requirements to hand to the real engineers on the team to slot in. They still have to fix some things to meet our standards, but I’m basically a “researcher” level coder. Which makes sense — I do have an undergrad and MS in CS, and did a lot of mathy algo stuff. For the last 15+ years I could never use anything in my brain to help the team solve things I was best suited to solve. I am now, and that’s nice.

The one key point is that I am keenly aware of what I can and cannot do. With these new superpowers, I often catch myself doing too much, and I end up doing a lot more rewrites than a real engineer would. But I can see Dunning Kruger playing out everywhere when people say they can vibe code an entire product.

Well technically AWS has never failed in wartime.

Goblet? Or is this something new? Deep goblets are great for opening the ankles and hips/SI area in ways that have helped my back. Some combination of improving mobility in other reasons prevents my back from overcompensating I guess

Yes. For very high risk patients, payers do want this. I’ve even heard of some paying pharmacies $100/fill if done on time for select people.

The problem is, prediabetic and folks who may have crossed 7.0 A1C once, and just overweight folks with docs who are willing to play fast and loose are demanding it. Skipping metformin and other first line treatment options that are way cheaper. For those folks, complications might be the next guys problem.

I was going to disagree but then realized I now shell out at least $100 when two families and their kids show up for 3-4 pizzas with toppings and chips and dip and some juices.

And god forbid I try and provide fresh fruit and beverages on that budget…

Wait til you get a load of Tamil/Malayalam transliterations’ use of “zh”. It was proposed by some German linguist to represent a really retroflex “r” and now makes outsiders pronounce kozhikode as “cozy-code” instead a closer “korikode”

Counterpoints: the detractors of this purported loop would likely neither fund the vast amounts of research they’d demand be done nor believe the results if they conflicted with their anecdata. I have yet to see a good faith argument against evidence based method that provides an effective and realistic alternative. Because that would take evidence.

4:3 only makes sense to you because you know which is length and width a priori. I for example, always have to recheck that. So if it was written as 1.33 or 4/3 it makes the same difference to me, and is similar in that way to dB

Satya had been at the company for quite a while himself. Perhaps not as long as others, but he was a well known veteran within the company

Locu. Acquired by GoDaddy. Morphed into a Yext competitor called GetFound.

It started as Jomax Technologies, I believe.

Historically in the US, Indian and Arab people were classified as “Caucasian”. In fact, it was marked on my relatives first speeding ticket in 1972.

There are more brown people than Indians… Usually these initiatives push for underrepresented brown people, ie Hispanic/Latino Americans.

Most diversity programs actively harm Indians as over represented, as they fall under the broad “Asian” category (see Harvard).

But I guess Indians are easy pickings these days.

This is the #1 reason for ad blocking in Eastern Europe and Asia (eg dolphin browser). It’s also why the current “solutions” big tech is providing viz. privacy and non-annoying ads doesn’t move the needle there.

Also, they program stuff like high rep clean and jerks and snatches along with high intensity cardio and gymnastic movements. Never a great idea to train high power, highly technical movements with fatiguing exercises and time pressure, let alone with poor form. I did a few classes in SF coming from a background in Olympic lifting and strength sports and I really found maintaining form and safety was quite hard.

I believe it’s because a doctor doesn’t prescribe cigarettes to you. So the only liability is your own.

It’s a weak line of reasoning to _allow_ cigarettes, but I’m not convinced it’s so bad to _disallow_ novel therapeutic uses that are understudied.