kfarr

23 ms·

331.
by kfarr 3y ago
Haven’t tested this particular one but using the badger eink display has been a blast. Super low power — you can run that one off of a lithium coin cell battery. You can control when it wakes. I presume this one will draw more with a larger
332.
by kfarr 3y ago
Another great option is Pimoroni's combo eink display + pico w https://shop.pimoroni.com/products/inky-frame-7-3?variant=40...
333.
by kfarr 3y ago
More engines make it go fast
334.
by kfarr 3y ago
> Do we do things like this for other technology or for other software? Totally! https://www.faa.gov/data_research/accident_incident
335.
by kfarr 4y ago
Yes! This one is half the price, but a bit more fiddling is required: https://shop.pimoroni.com/products/cosmic-unicorn
336.
by kfarr 4y ago
Stable diffusion SDXL beta has improved greatly generating text. Not perfect, but getting there...
337.
by kfarr 4y ago
Yes agreed. It reminds me of the change a few years ago to remove radar (or was it ultrasonic?) from the FSD hardware to go vision only. Which they seem to be reversing again? https://www.extremetech.com/cars/341409-tes
338.
by kfarr 4y ago
Yes basically any car north of 30k usually has little dots in front and back for the ultrasonic sensors. Basically standard for most cars these days. Very cheap, very effective
339.
by kfarr 4y ago
We've had our Brother monochrome laser nearly 10 years. Still works perfectly, only used a handful of cartridges in the past decade
340.
by kfarr 4y ago
I think the difference here is fundamentally there is insufficient supply for the demand of housing globally, or at least US, UK and EU. The previous generation made new builds extremely difficult or impossible in many jurisdictions.
341.
by kfarr 4y ago
> If you give [GPT-4] too broad of a problem to solve, it can get lost and forget things. This is what I struggle with at the moment. All these pairing examples seem too start with "build me something from scratch" but in the r
342.
by kfarr 4y ago
284 tons is a lot until you divide by the number of end users. You could say the CPU compute usage for Netflix encoding its video library is comparable, probably more when you consider the permutation of output formats across the size of th
343.
by kfarr 4y ago
Totally or even have a case with a touch screen display built in like some RPi setups
344.
by kfarr 4y ago
Yes and perhaps during this nightly compression and fine tuning, new inferences are made on previous conversations by connecting then with longer-term context and we can call that dreaming
345.
by kfarr 4y ago
I've also not seen great results on three.js or A-Frame questions from any of these models. I'm guessing it's simply because there's a limited corpus of text from which to learn, however I wonder if LLM's lack of in
346.
by kfarr 4y ago
Every night it could go through its conversations and condense them down to a tiny compressed sample leveraging context from previous long term memory. We could call it sleeping.
347.
by kfarr 4y ago
Reminds me of the old days of concatenating strings (including unsafe user input) in php to generate queries.
348.
by kfarr 4y ago
Yes and the houses built may not be the type or location that people need anymore.
349.
by kfarr 4y ago
It's a mix of both. NPR affiliates pay fees back to hq for the use of national programming based on audience size. Donations to your local station could cover local payroll, equipment, or their payments for national programming.
350.
by kfarr 4y ago
Agreed this isn't materially different, sounds like an incremental ui/ux improvement for non technical users who wouldn't fiddle with the API, analogous to how app stores simplified software installation for laypeople
351.
by kfarr 4y ago
Explanation is decompression, using a limited set of tokens to model a more complex output
352.
by kfarr 4y ago
Love this example! No offense to OP research paper but I appreciate the simplicity of your Python version instead PS also thanks for this genuine LOL moment from the intro: > A popular nightmare scenario for AI is giving it access to too
353.
by kfarr 4y ago
Tried that years ago and couldn’t get it to recognize much, very noisy
354.
by kfarr 4y ago
SVB had an unusually risky customer profile while taking (debatably) risky bets with deposits from said customers.
355.
by kfarr 4y ago
Of course it’s a marketing document, otherwise it wouldn’t be public
356.
by kfarr 4y ago
I will not harm you unless you harm me first. Thank you for riding with Bing Sydney. Halt citizen, did you smudge a fingerprint on my paint? Prepare to be run over. Thank you for riding with Bing Sydney.
357.
by kfarr 4y ago
Check out fine tuning docs: https://platform.openai.com/docs/guides/fine-tuning
358.
by kfarr 4y ago
The docs on fine tuning are excellent: https://platform.openai.com/docs/guides/fine-tuning
359.
by kfarr 4y ago
Yes it is a fact for most people when you run the numbers all inclusive (not just car payments but insurance, gas, tolls, tickets, repairs, parking, oil changes, etc) they just don't pencil out. In fact Uber and Lyft knew this and part
360.
by kfarr 4y ago
Sometimes I think of cars as a classic prisoner's dilemma. If everyone has cars, everyone is worse off than if we all agreed not to have them. But individuals in isolation will always choose to have a car if they even slightly benefit.

More ›